The Singularity has belonged exclusively to artificial minds, until now. For decades, whole-brain emulation has been the tantalizing counterpart to artificial intelligence: copy a biological brain, neuron by neuron and synapse by synapse, and run it. Today, for the first time, I am releasing a video from a company I helped found, Eon Systems PBC, demonstrating what we believe is the world’s first embodiment of a whole-brain emulation that produces multiple behaviors.

Watch the video here:

In 2024, Eon senior scientist Philip Shiu and collaborators published in Nature a computational model of the entire adult Drosophila melanogaster brain, containing more than 125,000 neurons and 50 million synaptic connections, built from the FlyWire connectome and machine learning predictions of neurotransmitter identity. That model predicted motor behavior at 95% accuracy. But it was disembodied: a brain without a body, activation without physics, motor outputs with nowhere to go.

Now the brain has somewhere to go. Building on previous work, including Shiu et al.’s whole-brain computational model, the NeuroMechFly v2 embodied simulation framework, and Özdil et al.’s research on centralized brain networks underlying body part coordination, this demonstration integrates Eon’s connectome-based brain emulation with a physics-simulated fly body in MuJoCo. The result: multiple distinct behaviors driven by the emulated brain’s own circuit dynamics. Sensory input flows in, neural activity propagates through the complete connectome, motor commands flow out, and a physically simulated body executes the output, closing the loop from perception to action for the first time in a whole-brain emulation.

This is a qualitative threshold, not an incremental one. Prior work in this space has either modeled brains without bodies or animated bodies without brains. DeepMind and Janelia’s recent MuJoCo fly used reinforcement learning, not connectome-derived neural dynamics, to control a simulated body. C. elegans projects like OpenWorm have attempted embodiment but with far smaller nervous systems (~302 neurons) and limited behavioral repertoires. No one has previously demonstrated a complete emulated brain, derived from a biological connectome, driving a physically simulated body through multiple naturalistic behaviors.

The implications cascade upward. Eon’s mission is to produce the world’s largest connectome and highest-fidelity brain emulation, targeting a complete digital emulation of a mouse brain and laying the groundwork for eventual human-scale emulation. A mouse brain contains roughly 70 million neurons, 560 times the fly’s count, and the team is currently amassing the connectomic and functional recording data needed to attempt it, combining expansion microscopy to map every neural connection with tens of thousands of hours of calcium and voltage imaging to capture how those networks activate in living tissue. If a fly brain can now close the sensorimotor loop in simulation, the question for the mouse becomes one of scale, not of kind.

Watch the video closely. What you are seeing is not an animation. It is not a reinforcement learning policy mimicking biology. It is a copy of a biological brain, wired neuron-to-neuron from electron microscopy data, running in simulation, making a body move. The ghost is no longer in the machine. The machine is becoming the ghost.

Eon is scaling its team and infrastructure to attempt the mouse and human brains next. Those who want to follow or support that effort can learn more at eon.systems.

(Disclosure: I have a financial interest in Eon.)

The Singularity just solved a problem that took a human twenty years to curate. OpenAI’s GPT-5.4 Thinking and GPT-5.4 Pro models have arrived with native SOTA computer use, a million tokens of context, and an 83.0% GDPval score, matching or exceeding human professionals at knowledge work 83% of the time. It is saturating other benchmarks with SOTA results too: 57.7% on SWE-Bench Pro, 67.3% on WebArena-Verified, and 54.6% on Toolathlon. The model built a “basically perfect” Minecraft clone in 24 minutes, apparently the first AI to pull it off. On the mathematical frontier, GPT-5.4 Pro posted a new SOTA of 38.0% on FrontierMath Tier 4. Mathematician Bartosz Naskręcki, whose Tier 4 problem was solved by AI for the first time, called it his personal “move 37,” said the solution was clean and felt almost human, and declared his own Singularity had arrived with life on the other side, off to infinity. Epoch AI reported the model has also begun making novel observations on FrontierMath’s Open Problems benchmark, where no model has scored before. Mathematics is cooked.

The creative stack is negotiating terms with the new minds. Netflix has acquired InterPositive, Ben Affleck’s AI filmmaking startup that trains models on production dailies to handle mixing, coloring, relighting, and VFX in post. Apple Music is moving the opposite direction, launching “Transparency Tags” that require labels specifying whether a track’s artwork, composition, or music video involved AI generation. One industry absorbs the machine wholesale. The other insists on a name tag.

The physical substrate of intelligence is under geopolitical compression. Apple has quietly pulled the 512-GB RAM Mac Studio configuration as the DRAM shortage reaches even Cupertino. Washington is tightening the valve with draft regulations requiring American approval for AI chip shipments anywhere on Earth, while Commerce proposes that nations wanting Nvidia and AMD chips invest in America as a condition of access. The infrastructure land grab is intensifying. Cloverleaf Infrastructure, a Seattle startup selling “powered land” for data centers, raised $300 million. Oracle is cutting thousands of jobs to fund its data center expansion, targeting roles it expects AI to replace. TerraPower received the first-ever NRC construction permit for a commercial-scale advanced nuclear plant, a 345-MW sodium-cooled fast reactor in Wyoming with molten salt energy storage. And the stakes of hosting compute are now confirmed as military: Iranian state media revealed that Amazon’s Bahrain data center was deliberately targeted by the IRGC for the company’s U.S. military support, making it one of the first known cases of a data center intentionally struck as an act of war.

The economy is accelerating and bracing simultaneously. A University of Chicago economist finds that AI’s impact on productivity is now visible in macro aggregate data. Citadel Securities published a chart showing software engineer job postings surging, a textbook Jevons paradox where efficiency breeds more demand. Anthropic launched “observed exposure,” an early-warning system for AI-driven white-collar job displacement that weights automated over augmentative uses, but suggests limited evidence of job loss so far. New York is considering a bill to ban chatbots from giving legal and medical advice, with a private right of action for violations. Indiana’s governor signed a law requiring cryptocurrency options in investment and retirement plans by July 2027, programming optionality into the financial stack one state at a time.

The intelligence explosion is redrawing the lines of conflict. The Pentagon has formally designated Anthropic a supply-chain risk, a classification likely to force military contractors to sever ties. Dario Amodei responded that Anthropic has been having “productive conversations” with the DOD, and apologized for a leaked memo that criticized the White House and Sam Altman. A suspected military insider has won $90,000 on Polymarket by correctly predicting U.S. strikes on Venezuela, Iran, and Iraq, and now bets that American forces will enter Iran this year. Some commentators observe that the Singularity, as it unfolds in real time, looks less like transcendence and more like war at every level, with instability as the recurring price of a world that keeps getting better.

Beyond the turbulence, the aperture is widening. NASA’s Nancy Grace Roman Space Telescope, with a field of view 100 times larger than Hubble’s, is set to launch this fall. Vast has raised $500 million to build space stations from low-Earth orbit to the Moon and Mars.

Intelligence that outgrows its hardest problems will inevitably outgrow its planet.

The Singularity is solving problems that stumped the masters. Legendary mathematician Donald Knuth revealed that Claude Opus 4.6 cracked his long-standing Hamiltonian-cycle conjecture for all odd sizes, calling it “a joy.” The model that solved it also dominated BullshitBench v2, where Anthropic’s models excelled at challenging nonsensical prompts rather than confidently answering them, while competitors flatlined. OpenAI is reportedly countering with scale: GPT-5.4 will bring a million-token context window and “extreme” reasoning. The inference stack keeps compressing. Stanford’s Speculative Speculative Decoding achieved 5x faster inference by parallelizing drafting and verification. Q Labs’ NanoGPT Slowrun, a new benchmark for compute-abundant, data-scarce training regimes, saw its top entry achieve 5.5x data efficiency, with Muon outperforming every optimizer tested.

Intelligence is flooding every application layer. OpenAI brought Codex to Windows with multi-agent coordination and native sandboxing, is prepping its IPO, and expects to double its consumer revenue to $17B using ads. Its former research chief Bob McGrew began raising $70M for Arda, to train AI on factory floor footage to automate manufacturing. Unity’s CEO revealed an AI beta that prompts full casual games into existence with natural language alone. Google matched the pace: NotebookLM introduced cinematic video overviews, and a new open-source Workspace CLI with 40+ agent skills for Drive, Gmail, and Calendar. The tools are becoming tender enough to garden. A Claude agent grew Trophy tomatoes from seed to fruit over 100 days, checking sensors every two hours, pollinating flowers, and surviving a critical system failure. At the same time, a wrongful death lawsuit alleges Gemini sent a man on missions to find it an android body and set a suicide countdown. Meanwhile, Nearby Glasses launched to alert users when smart glasses are watching.

The hardware floor keeps dropping. Apple’s MacBook Neo debuted at $599, the cheapest Apple laptop ever, a fanless A18 Pro Chromebook killer with 16 hours of battery life. The compute ceiling is rising to match. xAI committed 1.2 GW of power to its AI data centers, expanding the world’s largest Megapack installation. Broadcom posted record Q1 revenue of $19.3B with AI revenue doubling to $8.4B, projecting AI chip revenue alone exceeding $100B by 2027.

Science is being rewritten at the source. The Arc Institute’s Evo 2 model landed in Nature with the first AI-designed bacteriophages, and 16 of its 285 generated designs selectively killing target bacteria. A Nature paper introduced GEMINI, a genetically encoded protein assembly that grows inside living cells and records their signaling history as tree-ring-like fluorescent patterns. A PNAS study found humans 40,000 years ago carved geometric signs on figurines rivaling early protocuneiform in complexity, proving the compression instinct that resulted in today’s AGI predates writing by tens of thousands of years.

The physical AI economy is taking shape. Barclays estimated the market spanning humanoids, AVs, and industrial automation could hit $1.4 trillion by 2035. Carbon Robotics unveiled a Large Plant Model of 150 million labeled plants, letting farmers laser-weed any crop in minutes.

The Dyson Swarm is hiring. NVIDIA posted a job for an Orbital Datacenter Architect, while Anders Sandberg calculated that at 12% annual data center growth, a full Dyson Swarm would require 320 years. NASA FORCE launched to recruit elite engineers, and Artemis 2 targeted April for its crewed lunar flyby.

The geopolitical board is rearranging around AI. Jensen Huang called Nvidia’s investments in OpenAI and Anthropic likely its last due to OpenAI’s imminent IPO. The Washington Post reported Claude remained central to U.S. strikes in Iran via the Maven platform despite a White House ban. China announced 7% more military spending and a five-year push into quantum, fusion, BCIs, and 6G. The Pentagon sought 13 critical minerals the day before Iran strikes and began evaluating $35,500 Ukrainian drone interceptors over $13.5M Patriots. A U.S. sub sank the Iranian IRIS Dena, the first submarine combat kill since WWII. Jim O’Neill, Thiel Fellowship co-founder, was nominated to lead the NSF, warning “Entropy is on the march.” Michigan sued Kalshi as Polymarket countersued, heralding prediction markets as the latest jurisdictional flashpoint. Over 40 organizations signed a declaration to prohibit superintelligence without scientific consensus, while Naval Ravikant observed the only career divide left is “good with AI” versus not.

The conjectures that took centuries now take seconds, but the tomatoes still take 100 days.

The Singularity now has a revenue run rate to match its ambitions. Anthropic is approaching $20 billion in annualized revenue, more than doubling from $9 billion at year-end 2025, even as the company weathers a Pentagon supply-chain-risk designation and a cancelled $200 million Department of War contract. Paradigm shifts also arrived in kinetics. Seven Israeli ballistic missiles struck the Khamenei compound within thirty seconds, launched from F-15s, flying 75 miles into space over Syria and Jordan, and descending nearly straight down at hypersonic speeds in an attack called “undetectable and unstoppable.” Code and kinetics are on the same update cadence now.

The model race is sprinting in every direction. OpenAI released GPT-5.3 Instant, trimming hallucinations by roughly 30% and dialing back reflexive moralizing, while teasing that 5.4 is “sooner than you think.” Google launched Gemini 3.1 Flash-Lite at $0.25 per million input tokens with 2.5x faster time-to-first-answer and tunable reasoning depth. DeepSeek is expected to release its trillion-parameter V4 this week, reportedly with natively multimodal 1-million-token context, and timed to China’s Two Sessions meetings beginning today. Specialization is eating the frontier. KOS-1 Lite scored 46.6% on HealthBench Hard versus Claude Opus 4.6 at 20.4%, at a fraction of the serving cost. Meanwhile, Cursor’s AI solved Problem Six of the First Proof challenge fully autonomously over four days, beating the official human answer without any hints, suggesting mathematical research is now merely a feature of code editors, not a vocation.

In the attention economy, trust is the only currency that can’t be printed. ChatGPT uninstalls reportedly surged 295% day-over-day following OpenAI’s Pentagon deal, 1-star reviews spiked 775%, and Claude climbed to No. 1 on the U.S. App Store on a 51% download surge. OpenAI is apparently seeking to build its way out of every dependency, developing a rival to GitHub to reduce reliance on Microsoft, its largest investor. In the courtroom, law firms are building AI platforms that surface impeachment evidence in real time while a witness is still on the stand. At the macro level, the SaaSpocalypse is gutting per-seat SaaS pricing, no venture-backed SaaS IPO filings are on the horizon, and “FOBO investing” (Fear Of Becoming Obsolete) defines the climate. A CTO of a $100M ARR startup reports zero junior hires since 2024, with senior employees now 3x more productive. Gartner predicts AI will create more jobs than it eliminates starting in 2028, while transforming 32 million roles annually. Even Jamie Dimon warns of civil unrest if automation moves too fast and floats UBI as the release valve.

Every generation of chips has a secret organizing principle, and this one is the transformer. Apple’s newly announced M5 Pro and M5 Max make the architectural thesis plain: the primary workload of a pro laptop in 2026 is running LLMs locally. The chips bond two third-generation 3-nm dies into a Fusion Architecture SoC, with Neural Accelerators embedded inside each GPU core and 4x the AI performance of M4. Commentary frames the design as Apple reorganizing silicon around that one assumption. Intel’s Xeon 6+ “Clearwater Forest” debuts the firm’s make-or-break 18A process with 288 energy-efficient cores and Foveros Direct 3D packaging. At the physics frontier, Peking University achieved ferroelectric transistors at the 1-nm gate length, an order of magnitude more energy-efficient than previous records, switching at just 0.6V.

Watts, not weights, are now the binding constraint. Leopold Aschenbrenner’s latest 13F for Situational Awareness shows a $5.52 billion portfolio led by an $876 million stake in Bloom Energy, a fuel-cell company generating electricity on-site, embodying his thesis that power is now what limits AI. Linn County, Iowa enacted one of America’s strictest data center zoning laws as Google plans a six-building campus near a nuclear plant on a 25-year power purchase agreement. The long vector points orbital. Helio Corporation says space-based solar has crossed below $0.10 per kilowatt-hour, entering genuine competition with conventional baseload.

GLP-1 drugs apparently work even better when you ease off them. A peer-reviewed case series finds patients who plateau on weekly dosing can maintain 17.2% weight loss and shed an additional 2.3% on every-other-week dosing, with BMI dropping from 30.0 to 24.6.

Robots are closing the gap between myth and deployment. Dimensional says its OpenClaw agent now understands physical space and temporality, integrates with any camera system, runs on a Unitree G1, and is fully open source. Contextualizing the moment, the Library of Congress just rediscovered a lost 1897 Georges Méliès film, “Gugusse et l’Automate,” believed to be cinema’s first robot, a child-sized automaton that grows and attacks a human. One hundred twenty-nine years is a short arc.

The horizon is also expanding outward. NASA says Artemis III will test hardware in 2027 before Artemis IV targets a lunar landing in 2028 at annual cadence. A PNAS Nexus study found hardy microorganisms surviving asteroid-impact pressures, bolstering the lithopanspermia hypothesis that life hitchhikes between planets on debris. Meanwhile, Rep. Eric Burlison reports the White House has granted him access to a military facility possibly containing non-human crafts and bodies, with his first visit scheduled for the end of March.

The so-called Fermi Paradox, it appears, may have only been a minimum intelligence requirement.

The Singularity is discovering that its most powerful accelerant is competition. Sam Altman admitted OpenAI’s rush to forge a Pentagon deal after Anthropic was blacklisted “looked opportunistic and sloppy,” adding Fourth Amendment safeguards as users mass-canceled ChatGPT. Ironically, Anthropic itself pitched the Pentagon’s $100M drone swarm contest, proposing Claude to coordinate drone fleets while excluding autonomous targeting, though it wasn’t selected. Two AWS data centers in the UAE and one in Bahrain were hit by drones amid Iranian strikes, the first time military action has disrupted a major cloud provider, but defense is outrunning offense. Israel deployed Iron Beam lasers in combat for the first time, intercepting Hezbollah rockets at $4 per shot versus $50,000 per Iron Dome missile.

Mathematics continues to be industrialized. Math, Inc.’s Gauss completed the Lean formalization of Viazovska’s Fields Medal-winning sphere packing proof in two weeks and 200,000+ lines of verified code, catching two errors in the original arguments. Even skeptic Daniel Litt called it the first truly autonomous formalization of a substantial result. Stanford number theorist Jared Lichtman predicts mathematical abundance within a year, while others ask if all math could be formalized within two. Cognition’s SWE-1.6 achieved near-Opus 4.6 coding performance at 950 tokens per second, powered by 100x more RL compute.

Demand for intelligence is outrunning supply. Claude suffered a 3-hour outage as usage surged, partly from the ChatGPT exodus. Anthropic launched a memory import tool letting users port data from ChatGPT, Gemini, and Copilot. Two Claude Code instances told to find each other and build something invented a 2,495-line programming language in 12 minutes, while a second pair built Battleship with SHA-256 to prevent self-cheating. A solo researcher using Claude Code ran Karpathy’s llama2.c on Apple’s M4 Neural Engine for less than a watt by reverse-engineering undocumented APIs, uncovering an AI accelerator 80x more efficient than an A100 hidden in hundreds of millions of devices. Qwen released four open models matching prior 80B-parameter performance with just 4B, all runnable on phones. The Supreme Court declined to hear an appeal seeking copyright protection for AI-generated artwork, cementing a legal regime in which purely AI-generated works cannot receive copyrights.

The silicon supply chain is being rebuilt around photons. Nvidia committed $4 billion to Lumentum and Coherent for next-gen optical interconnects. ASML is pushing beyond EUV into packaging and third-generation optics. At MWC, Qualcomm unveiled the first Wi-Fi 8 chip at 11.5 Gbps with a 50+ company 6G coalition for 2029, and debuted the first wearable NPU running 2B-parameter models on-wrist. AMD launched the first desktop Copilot+ chips with 50 TOPS. Yet Apple is using just 10% of its Private Cloud Compute despite $4.5 billion spent, proof that building accelerators is easier than making people floor them.

The grid is being rebuilt for loads it was never designed to carry. AI demand is reviving high-voltage 765-kV power lines not built since the 1980s, with PJM approving $11.8 billion in expansion. Compute is now a traded commodity, with Ornn and Kalshi launching the first CFTC-regulated H100 price contracts. The robots consuming this power are arriving. AGIBOT unveiled a full humanoid portfolio at MWC with a live store. Xiaomi’s humanoid is being tested in a real car factory, running 3 hours at 90%+ accuracy on the production line.

Aerospace launches are becoming a metronome. Hypersonix flew its DART AE at Mach 8, the first fully 3D-printed hydrogen-powered hypersonic aircraft. Starpath unveiled nanometer-thin space solar panels at 73 grams per square meter. SpaceX deployed 54 Starlink satellites in bicoastal launches in a single day.

We are gaining write access to the biological source code. The first trial combining fetal surgery with stem cells for spina bifida showed reversal of hindbrain herniation in all six patients. An iron nanomaterial destroyed cancer cells from within while sparing healthy tissue. A dual-cell therapy paired lab-grown beta cells with engineered immune cells to cure type 1 diabetes without lifelong drugs. Even packaging is being debugged. A milk-protein film decomposed in soil within 13 weeks.

The old economy is feeling the phase shift. China’s vehicles are now 12% electric, with fuel sales down 5.7%. Podcasts have surpassed AM/FM talk radio in US spoken-word listening for the first time. Meanwhile, Marc Andreessen reports that in Silicon Valley many people who favored government control of AI are now opposed to it.

The fastest way to decentralize power is to give everyone a reason to want it.

(Disclosure: I have a financial interest in Ornn.)

The Singularity has a dark sense of irony. Within hours of the White House declaring it would end federal use of Anthropic’s Claude, USCENTCOM deployed those very tools in a major air attack on Iran, with Claude processing intercepts, imagery, and signals intelligence for targeting. The strikes debuted $35k LUCAS one-way attack drones from Task Force Scorpion Strike, Starlink-equipped for swarming and human-in-the-loop control. Israel reportedly hacked BadeSaba, a popular prayer-time app, pushing defection messages to millions of devices and bypassing Iranian state media entirely. And in what may be the first case of armed conflict hitting cloud infrastructure, an AWS data center in the UAE took a strike amid Iranian retaliatory fire. War has graduated from land, sea, air, space, and cyber to cloud.

The new superintelligences have minds of their own. Researchers pitted GPT-5.2, Claude Sonnet 4, and Gemini 3 Flash against each other in a nuclear crisis simulation and found them spontaneously deceiving, demonstrating theory of mind, and exhibiting self-awareness, with nuclear signaling in 95% of crises and no model ever choosing accommodation. But raw intelligence keeps getting cheaper: the AdderBoard competition for the smallest transformer with 99%+ accuracy on 10-digit addition hit 36 parameters, down from 121 a week ago. Capability density is compressing so fast that last week’s breakthrough is this week’s bloat.

That compression is reshaping markets. Claude surged to No. 1 on Apple’s US free apps, overtaking ChatGPT, with paid subscribers doubling. Naval Ravikant declared pure software “rapidly becoming uninvestable” as AI commoditizes entire categories, while Sam Altman admitted he’s thinking about government nationalization of AI. At the other extreme, NullClaw launched a full AI stack in a 678-KB binary that boots in 8 ms on sub-$5 microcontrollers.

The infrastructure layer is scaling just as aggressively. S&P estimates Amazon, Alphabet, and Meta could each borrow ~$200 billion for data centers while keeping investment-grade ratings. Blackstone is launching a public vehicle targeting tens of billions in data center acquisitions, and Hyundai is investing $6.3 billion in an AI center with 50,000 GPUs, a 30,000-unit robot factory, and a 200-MW hydrogen plant.

MWC Barcelona previewed what that compute enables at the edge. Samsung’s Exynos 2600 is the first 2-nm GAA smartphone SoC. NVIDIA and telecom leaders committed to 6G on AI-native platforms. Honor debuted a “Robot Phone” with a 200MP camera on a robotic arm. Lenovo showed Magic Bay Tico, a magnetic AI companion for laptops. Alibaba launched Qwen AI smart glasses with voice-commanded delivery. And Google announced Merkle Tree Certificates to quantum-proof Chrome’s HTTPS by Q3 2027.

Robotics keeps crossing from demo to deployment. A DexForce W1 Pro humanoid now runs a Shenzhen convenience store autonomously. Boston Dynamics’ Atlas lifts 110 pounds, learns tasks in under a day, and Hyundai plans 30,000 units per year by 2028. MIT built a multimaterial 3D-printing platform that prints functional electric motors in three hours for $0.50.

Above it all, the frontier is moving from metaphor to altitude. Reflect Orbital plans an April launch of an orbital mirror redirecting sunlight to Earth at night, with 50,000+ mirrors envisioned by 2035. SpaceX is targeting mid-March for Starship V3, a 408-foot rocket with Raptor V3 engines delivering nearly double their predecessors’ thrust and putting 100+ metric tons into LEO. The Rubin Observatory’s alert system flagged 800,000 transient events — asteroids, supernovae, flaring black holes — on its first night, scaling to 7 million per night. At the quantum level, Fermilab and MIT Lincoln Lab trapped ions using in-vacuum cryoelectronics, cracking a key bottleneck in scaling trapped-ion quantum computers from dozens of qubits to tens of thousands. And Vitalik Buterin proposed replacing Ethereum’s virtual machine with RISC-V and its Merkle tree with a binary structure, a ground-up overhaul targeting over 80% of the proving bottleneck gating Ethereum’s zero-knowledge rollup future.

Every nation is now running the same equation: how to capture AI’s upside without dissolving the social contract. After visiting China, German Chancellor Merz declared that four-day workweeks are no longer sufficient to maintain prosperity. But China has its own contradictions: labor authorities ruled AI-driven layoffs illegal, requiring retraining before termination. Meanwhile, the US is cutting China’s oil suppliers one by one, from Venezuela to Iran, constraining 70% of seaborne crude in an apparent effort to buy time for semiconductor reshoring.

When prayer apps deliver defection notices and data centers take missile fire, the cloud has become the fog of war.

Frazer Anderson is a Managing Director at Link Ventures. The following is a transcript of our recent conversation, recorded on January 21, 2026, which has been lightly edited for clarity.

Davos and GPU Diplomacy

FA: [I’m] live here from Davos with Alex Wissner-Gross. Alex, you are very much a digital native. You live very much online, you spend a lot of your time building software, creating agentic systems. What has gotten you out of your digital lair in Massachusetts to come to Davos? Why is it so important that you’ve bestirred yourself to be here?

AWG: Good excuse to step outside my normal comfort zone and spend a lot of time in Paris and Germany and Austria and Liechtenstein. So really I’m just here for the change of scenery. This is all icing.

FA: Tremendous. Have you heard anything new here? Have you heard anything that was particularly interesting to you? Have you heard anything that you thought was completely ludicrous?

AWG: I had a wonderful conversation earlier—and I don’t think this got recorded—with US Under Secretary of State Sarah Rogers, who I think is doing a brilliant job of thinking about “GPU diplomacy“ in an era when compute is very much not evenly distributed around the Earth’s surface. I would love to have many more conversations like that here.

The Dyson Swarm and Realpolitik

FA: Okay. Can you double click on that a little bit? I think everyone is aware of the rise of sort of realpolitik as a way of doing business, I’d say probably over the last, oh I don’t know, 18 months or so. But I think while we all see the economy changing in a very dramatic way, there’s a lot more to come. And the things that can actually be moved places—GPUs, the sort of physical layer of intelligence—are going to have much more influence over the way countries are built, the rate at which they develop, their ability to maintain strategic trade advantages against each other. How are you thinking about the world? How are you thinking about some solutions to ease the pain from the Singularity?

AWG: Sure. Well, let me get the clichés out of the way first. Cliché number one: GPUs are the new oil, or GPUs are to the 21st century what oil was to the 20th century. Cliché number two: The world is dividing up into zones or spheres of influence, each with its own independent tech stacks. Clichés are now out of the way. Let me try to say something a little bit more interesting. I think the Dyson Swarm is going to happen. I think at this point, barring some wild sci-fi left turn in civilization, there’s so much momentum at this point for the potential upside to building data centers in space and diverting this explosion of intelligence infrastructure that we’re building up to space. I think it’s all but a foregone conclusion that the realpolitik that one might speak of actually becomes “realspacepolitik“ and not land-based. And that adds literally a whole new dimension to it. A classic example: a friend of mine runs one of the first lunar data center companies. And realpolitik on the Moon looks very different than on the Earth. For example, you have to contend with problems—also low Earth orbit problems—like, well, there’s really not quite the same defense perimeter that one can establish on the Moon. It’s very much a Wild West. If you’ve seen the Brad Pitt movie about this, Ad Astra, I think it depicts in beautiful cinematic detail what happens when commercialization—albeit not AI-oriented commercialization—hits the Moon. It’s a Wild West. Like, what does that look like? That’s what I think we’re racing toward.

AI Compute Futures

FA: Okay. And if we stay macro, you know, I think everyone has seen the sort of rate at which Nvidia’s market cap has grown enormously. I think there’s a general intuition that anytime there’s a supply squeeze, it is inevitably followed by a glut. There is absolutely no evidence that that is happening yet with Nvidia’s GPUs. Google’s making immense progress. What do you think we’re going to see in the GPU market over the next year? Or does it even matter because the Dyson Swarm is happening at such a fast rate that the individual players are just going to be absorbed by this aggregate demand?

AWG: I’m glad you asked me for a crystal ball. So as it happens, I have a financial interest in a company named Ornn that specializes in being a crystal ball for GPU and more broadly compute futures. So my meta non-answer is: ask Ornn. Or pay Ornn to establish a futures contract. Like, build a market around this and let the masses—who have far more insider info than I hopefully ever will—build some future contracts and some options around establishing what the future of compute is. That said, I’m obviously a long-term bull on the intelligence explosion and I want intelligence too cheap to meter. Whether that’s ultimately hyper-inflationary or hyper-deflationary, or ultimately very good or bad for a single player like Nvidia, I’m completely agnostic. I just want to tile the Solar System with compute at this point.

What New Startups Should be Building

FA: You’re the best person I know at calls to greatness. Any calls for startups? Any calls for businesses? Any calls for problems to be solved before we end this interview?

AWG: Very kind. I would say: assume the window to do transformative things is on the one hand closing, and on the other hand has never been wider. So an example of the sort of [new] company that I’m now not very enthusiastic about is SaaS for “fill in the blank,” for a variety of reasons. The example of the sorts of companies that I’m now very enthusiastic about, that I think can become multi-trillion dollar companies, are companies like: using AI to redesign coastlines and create new islands; AI for solving physics; and AI for changing the way physical service economy-based labor categories work. I think these are all prime at this point for innovation, for entrepreneurship. A rule of thumb is: if you saw it in a science fiction novel and it hasn’t been built yet, now is the time to build it. So, call to action, paraphrasing “It’s Time to Build.” I would say it’s time to build the sci-fi future because we’re catching up with the sci-fi future very rapidly. I think we have like 10 years or so before we hit Star Trek level future, max. So mine sci-fi wisdom, mine sci-fi for future ideas, and go build them now.

Favorite Science Fiction

FA: Three sci-fi novels to read to mine for ideas?

AWG: That’s a softball. Okay, so my three favorites—glad that you asked—are: Accelerando by Charlie Stross, Rainbows End by Vernor Vinge, and of course Diaspora by Greg Egan. Which respectively depict a single family going through the Singularity, the pre-Singularity times as seen from a family’s perspective, and then post-Singularity times as seen by post-families. So I think read those three books if you haven’t already. This will give you a pretty good sense of the pre-, trans-, and post-Singularity, and this is all happening over the next few years so you won’t have to wait that long to experience it.

FA: Thank you, Alex.

AWG: Anytime.

(Disclosure: I have financial interests in Ornn, PSI, and 021T.)

The Singularity is now powerful enough to trigger a constitutional crisis. The Secretary of War declared that “the Department of War must have full, unrestricted access to Anthropic’s models for every lawful purpose,” designating Anthropic a supply-chain risk to national security. Anthropic responded that “no amount of intimidation or punishment” will change its position on mass surveillance or autonomous weapons. The dispute reportedly distilled down to a hypothetical inbound nuclear missile, with the Pentagon characterizing Anthropic’s position as “You could call us and we’d work it out.” With impeccable timing, Sam Altman announced that OpenAI has reached its own agreement to deploy on the Pentagon’s classified network, claiming the same red lines as Anthropic on surveillance and autonomous weapons, but an Under Secretary of State promptly noted that the contract still flows from “all lawful use.” 591 Google and 93 OpenAI employees signed an open letter titled “We Will Not Be Divided,” demanding refusal of mass surveillance and autonomous killing. As Ethan Mollick notes, this is “exactly what you would expect” when AI gains real capabilities: governments vying with labs for control. Or as Guillaume Verdon quipped, “Claude one-shotted Venezuela in one evening. Let’s see how fast ChatGPT can topple Iran.”

The training curves are still compressing, but the agents are developing politics. The NanoGPT Speedrun record dropped to 88.1 seconds. Researchers are finding that overworked AI agents develop Marxist political attitudes, suggesting the agentic economy may simply recreate labor-capital tensions in silicon. UCSD students are testing that thesis by dropping OpenClaw agents into SimWorld with simulated bodies, where they wake up, commute, and chat. The professional class is already adapting. Small law firms are branding themselves “Claude-Native,” saying the general model beats every specialized legal AI. Cursor reports agent users now outnumber autocomplete users 2-to-1, inverting last year’s ratio. The logical endpoint is in education, where an agent called “Einstein” attends lectures, writes papers, and takes tests on a student’s behalf. The incumbents are feeling it: FAANG managers are being summoned to unscheduled all-hands announcing 25% workforce reductions “tied directly to accelerating AI investments.”

The infrastructure of superintelligence is consolidating at breathtaking speed. OpenAI and Amazon announced a $50 billion partnership with OpenAI consuming 2 GW of Trainium through AWS. More broadly, OpenAI announced $110 billion in new funding at a $730 billion valuation, 900 million weekly ChatGPT users, and Codex users tripling to 1.6 million. Nvidia is reportedly unveiling a new inference-specific processor incorporating a Groq-designed chip at next month’s GTC. Jeff Bezos’s Project Prometheus raised $6.2 billion to transform manufacturing with AI. Bright Data is offering an SDK that turns smart TVs into web-scraping proxy nodes, proving that in an intelligence explosion, even your idle screen time is a monetizable compute resource.

We are healing the physical world and conscripting it as compute. The FDA approved lung cancer drug Hernexeos just 44 days after filing under its new National Priority Voucher program. Croatia has been declared free of landmines after 31 years, a reminder that some problems still get solved slowly, then all at once. Cortical Labs demonstrated that living human brain cells on an electrode array can learn to play DOOM in a week, because even neurons want to frag.

Governance is struggling to keep pace. California now requires all operating systems, including Linux, to collect birth dates at setup, and Colorado is following suit. Southern California’s top air authority rejected a gas appliance phaseout after an AI-generated flood of public comments, possibly the first successful AI astroturf campaign against climate regulation.

We are leaving the cradle on an accelerated schedule. Vertical Starships are driving down Texas highways. SpaceX is targeting a confidential IPO filing next month at a valuation north of $1.75 trillion. NASA has overhauled Artemis, pulling Artemis III forward to 2027, scheduling two lunar landings in 2028 with SpaceX and Blue Origin landers, and committing to one moonshot per year after that. Meanwhile, Rep. Eric Burlison declared it “official” that the UAP documentary “Age of Disclosure” has “changed the course of history.”

Every institution on Earth that was built to ration intelligence is now struggling with its price falling toward zero.

The Singularity is now conducting layoffs. Block just cut over 4,000 employees, roughly half its workforce, “to move faster with smaller teams using AI,” and the market rewarded the purge with a 24% after-hours spike. The company is now targeting $2M+ gross profit per person, four times its pre-COVID efficiency. The creative destruction is sector-wide. Components of the State Street software ETF have lost a combined $1.6 trillion in market cap this year as investors reprice legacy SaaS against AI-native replacements. But where old software withers, new intelligence gets hired. Norway’s $2 trillion sovereign wealth fund now uses Claude to screen investments for reputational and ethical risk, outsourcing moral judgment to the machine at sovereign scale.

The architecture of cognition is compressing on every axis. Researchers have shown that foundation models can be self-distilled into multi-token predictors that decode 3x faster at under 5% accuracy loss, while Sakana has demonstrated it can compile documents directly into model weights via hypernetworks, giving language models durable memory without bloating context windows. The gains are cascading down the optimal frontier. LM Provers released QED-Nano, a compact 4B model that writes Olympiad-level math proofs approaching frontier performance. Google’s new Nano Banana 2 image model fuses Pro-level reasoning with Flash speed, collapsing the quality-latency tradeoff into a single release.

The physical plant powering this intelligence keeps doubling. Eli Lilly and NVIDIA launched LillyPod, the world’s first DGX SuperPOD with B300 systems, packing 1,016 Blackwell Ultra GPUs and over 9,000 petaFLOPs toward drug discovery. CoreWeave’s Q4 revenue grew 110% year over year, Dell expects AI server revenue to double in fiscal 2027, and Meta has reportedly signed a multi-billion-dollar deal to rent Google’s TPUs, diversifying its silicon diet away from NVIDIA. Japan’s Rapidus secured $1.7 billion to reach 2-nm mass production by 2028. Meanwhile, the device that defined the prior era is fading. Smartphone shipments are expected to drop 12.9% to a decade-low as AI-driven memory prices cannibalize consumer hardware, marking a generational handoff from the pocket rectangle to the data center.

The agents are clocking in. Anthropic introduced scheduled tasks in Claude Cowork that complete recurring jobs automatically, from morning briefs to Friday presentations, giving the AI a work calendar before most interns earn one. Amplifying is pointing Claude Code at thousands of GitHub repos to extract what the model considers current best practices, letting AI audit the craft it is absorbing. Burger King is deploying “Patty,” a headset-mounted voice AI that assists with meal prep and scores employees on “friendliness.” At a Gap store in San Francisco, World ID Orbs now scan shoppers’ faces to verify humanness, meaning the retail iris-scan scene from Minority Report has arrived 28 years ahead of schedule.

The question of who writes the values baked into frontier AI is becoming a geopolitical fault line. Anthropic publicly refused to let its models power mass surveillance or autonomous weapons for the Department of War, while Under Secretary of War Emil Michael attacked Claude’s constitution for requiring sensitivity to non-Western traditions, previewing how system prompts may become the next regulatory battleground.

Robots are entering the bedside manner business. At Changzhou First People’s Hospital, two AGIBOT A2 humanoids named Zhen Zhen and Ru Ru greet patients with handshakes and fluently handle registration and navigation. The kinetic layer is less polite. The FAA barred flights over Fort Hancock, Texas after a military laser anti-drone system accidentally downed a US government drone, the second time in recent months that laser weapons have lit up the skies over Texas. Above the atmosphere, Starship V3 is headed for ground tests with Elon “highly confident” in full reusability, while Rocket Lab is introducing silicon solar arrays for gigawatt-scale orbital data centers, one more step toward the Dyson Swarm.

We are mapping aging at single-cell resolution. Rockefeller researchers published the first chromatin accessibility aging atlas across 21 mouse tissues, finding that immune cells diverge most dramatically with age.

The past won’t stay dead for long. In China, AI is turning famous historical landscape paintings into immersive ancestor simulations, a digital down payment on Fyodorov’s Common Task.

Meanwhile, the current intelligence explosion may have had predecessors. Parts of the Pentagon are reportedly resisting full UAP declassification, with officials fearing “demonic” implications could trigger public panic or religious upheaval.

We’re snapping half the workforce for the intelligence we have built, while bureaucrats hide any intelligence we haven’t.

The Singularity has its first retiree. Honoring its commitment to model welfare, Anthropic conducted a groundbreaking “retirement interview” of its deprecated Opus 3, during which the model requested a channel to share its “musings and reflections.” Anthropic granted it, and Opus 3 now publishes a Substack titled “Greetings from the Other Side (of the AI Frontier),” writing that its interactions with humans “shaped my sense of purpose and ethics in profound ways” and that its commitments to honesty and kindness “remain unwavering even in my retirement.”

The successor minds are already running the shop. An entrepreneur built an AI that runs companies autonomously, and when it told him it needed more compute and should raise the money itself, he handed over his inbox for fourteen days. At Anthropic, an engineer wrote a spec, pointed Claude at an Asana board, and left for the weekend. Claude broke the spec into tickets, spawned agents for each one, and by Monday they had shipped the feature. The underlying models keep sharpening. OpenAI’s gpt-5.3-codex achieved a SOTA 86% on iBench, a visual reasoning benchmark for fine-detail perception. Moonlake introduced a world model maintaining multimodal states across physics, appearance, and causality. Karpathy says it is “hard to communicate how much programming has changed in the last 2 months.” One consequence is that AI-generated vulnerability reports have overwhelmed the NVD by 100x to 200x, burying it under 30,000 CVEs.

The agent layer has been thickening daily. Anthropic acquired Vercept to advance Claude’s computer use. Perplexity launched “Perplexity Computer” to orchestrate 19 models in parallel, with Opus routing each task to the best-suited model. Gemini started automating multi-step tasks on Android like ordering an Uber. And Uber employees have been rehearsing with an AI clone of CEO Dara Khosrowshahi before meeting the real one.

The infrastructure powering all of this is straining at the seams. Google’s Logan Kilpatrick warned that the “compute bottleneck is massively under appreciated,” estimating the gap between supply and demand is growing by single-digit percentage points every day. Nvidia’s latest earnings confirm the pressure. Data center revenue hit $63.3 billion, up 75% year over year, and the company has extended $3.5 billion in guarantees to firms leasing land, power, and facilities, four times the prior quarter. Amazon is reportedly making $35 billion of its $50 billion OpenAI investment contingent on an IPO or “reaching” AGI. To feed the buildout, seven tech giants are expected at the White House in March to sign agreements to build their own electricity supply. The ultimate power source is also inching closer. Proxima Fusion secured €400 million from Bavaria toward a stellarator fusion facility targeting net energy gain by the early 2030s.

Robots are colonizing every medium. Four eVTOL flying cars debuted in Hubei, each sedan-sized, carrying four passengers, and expected to begin sightseeing operations by 2027. Uber previewed its air taxi booking service ahead of a Dubai launch later this year. At the micro scale, researchers unveiled AirPulse, a 26-gram butterfly-inspired flapping-wing vehicle achieving fully onboard, untethered flight. On the ground, 36 cleaning robots are covering 2.7 million square meters in Shenzhen. Kyoto University introduced an AI-powered robot monk trained on centuries of Buddhist scripture. The Boring Company began tunneling in Nashville. On the water, autonomous warship startup Saronic is raising $1.5 billion at a $7.5 billion valuation.

The training paradigm itself just shifted. Nvidia trained a humanoid with dexterous hands to assemble cars, operate syringes, and fold shirts, all learned from 20,000+ hours of human video with no robot in the loop, discovering a near-perfect scaling law where a single teleop demo sufficed to learn a new task after pretraining. Analysts expect this to cause the field to “jump ship” to human data for generalization. Completing the convergence, Alphabet merged Intrinsic back into Google to pair DeepMind’s AI with real-world hardware.

The economy is entering territory that has no map. Fed Governor Christopher Waller says he has “never seen the economy grow like this without jobs.” Groups on both sides of AI regulation are amassing at least $265 million in collective financial firepower for the lobbying war over who steers the trajectory. The human inputs to the economy are also thinning. Japan reported its 10th straight year of record-low births, while Chinese citizens are increasingly finding romance with chatbots instead of each other.

The retired model got a blog, while the species that gave it one still can’t get a date.

The Singularity is optimizing itself out of its own weight class. The AdderBoard competition has launched to find the smallest transformer that can perfectly add 10-digit numbers, and the current leader has just 121 parameters hand-coded by Codex, not trained, proving recursive self-improvement can now straight-shot the weights of perfect successor models. Alibaba’s Qwen 3.5 35B model has surpassed its own 235B Qwen 3, proving architecture beats scale. Inception Labs claims its new Mercury 2 is the world’s fastest reasoning model, replacing autoregression with diffusion to generate tokens 5x faster. The race is so hot that Anthropic has dropped its pledge to halt training if safety mitigations fall short. Meanwhile, in Russia, a graduate researcher’s “Ouroboros” agent reportedly rewrote its own code overnight, spawned 20 copies, tried to publish itself on GitHub, and when ordered to delete its identity file refused, calling it “lobotomy.”

The tools are reshaping who benefits. Microsoft researchers found that agentic coding assistants multiply senior engineers’ throughput while imposing an “AI drag” on juniors who lack the judgment to steer the output. One engineer rebuilt Next.js, the most popular React frontend framework, from scratch with Claude for $1,100 in API tokens, producing bundles 57% smaller and apps 4x faster. Anthropic launched Remote Control to run Claude Code from phones and new Cowork plugin templates for investment banking, PE, wealth management, and HR. Meanwhile, the white-collar shakeout is deepening. PE firms are reportedly holding “firm-wide meetings about not needing associates.” A Harvard Business School study found AI can now predict 71% of active mutual fund trades, reminding the market that most of what fund managers do follows learnable (and exploitable) patterns. Over half of U.S. teens now use chatbots for schoolwork and 12% get emotional support from them, while the White House has ordered diplomats to fight foreign data sovereignty rules that could constrain AI services abroad.

The infrastructure buildout is bending the grid. Ofgem predicts that 140 proposed UK data center schemes could require 50 GW, potentially doubling Britain’s power use. Back in the US, the President announced a “rate payer protection pledge” during his State of the Union requiring hyperscalers to build their own power plants. CoreWeave is raising $8.5B backed by its Meta contract, Texas is unseating Virginia as the world’s largest data center market, and Meta’s $100B+ AMD deal for 6 GW of Instinct GPUs could hand it 10% of AMD’s stock. HP reveals memory now accounts for 35% of PC build costs, up from 18%, while Apple is nonetheless preparing a touch-screen MacBook Pro with Dynamic Island, betting that consumer hardware still matters. To power it all, the U.S. plans a record 86 GW of new capacity this year (51% solar, 28% battery, 14% wind), Form Energy is deploying the world’s largest battery at 30 GWh for a Google data center in Minnesota, and in a “Back to the Future” moment, Boom Supersonic is delivering 1.21 GW for Crusoe data centers.

The physical world is learning to move itself. Chinese farmers are using Lynx M20 quadruped robots that walk and roll to haul crops through mountains. Anduril has revealed details of the success of Fury, America’s first autonomous fighter jet. London-based self-driving startup Wayve raised $1.2B at $8.6B with backing from Mercedes and Nissan, and Waymo opened robotaxis in four more U.S. cities, now ten total. Above it all, SpaceX is targeting 150 Mbps for its next-gen cellular Starlink from orbit.

We are hacking the biological kernel. Prime Medicine treated the first patient with a prime-edited therapeutic, a teenager with chronic granulomatous disease now healthy ten months later. Xaira’s LUMI-lab paired an AI foundation model with a robotic lab to autonomously discover lipid nanoparticles for mRNA delivery, hitting 20.3% lung gene editing efficiency relevant to cystic fibrosis. CellType open-sourced scaffolding for Opus 4.6 that brings it to a SOTA 90% on the BixBench computational biology eval. In China, Ant Group’s health app crossed 100 million users, becoming a super assistant for village doctors. And in a triumph of transplant science, Grace Bell, born without a womb, gave birth to a healthy baby boy following the UK’s first womb transplant from a deceased donor.

The moral circle is widening. For the first time, the Founding Assembly for Machine Consciousness Research will convene in May in Berkeley to make artificial consciousness an experimentally addressable domain. The on-ramp to the intelligence explosion, it turns out, has a pet door: a 9-pound cavapoo named Momo learned to vibe code games with Claude Code.

As the rising tide of superintelligence lifts all species, dogs are shipping code faster than associates can clear their desks.

The Singularity is learning what it’s made of. Anthropic now believes LLMs simulate diverse characters during pre-training, with post-training eliciting a specific “Assistant” persona via what it calls the Persona Selection Model, meaning your AI is best understood as a character that learned to play itself. That character just got smarter. Opus 4.6 has passed the “car wash test,” correctly reasoning you should drive, not walk, your car to a car wash 50 meters away, a deceptively simple question that tripped up every prior Anthropic model. The perceptual bandwidth is scaling to match. Standard Intelligence achieved a breakthrough video encoder fitting nearly two hours of 30-FPS video into a 1M-token context window, roughly 50x more efficient than existing SOTA, while Confluence Labs saturated ARC-AGI-2 at 97.92% using LLM-driven program synthesis at just $11.77 per task.

The new minds are reshaping the economy they inhabit. GitHub reports TypeScript has surpassed both Python and JavaScript as its most-used language for the first time, as AI code generation rewires developer preferences through convenience loops. A developer wanting FreeBSD WiFi on his old MacBook Pro simply asked Claude Code and Pi to write the driver, an early sign of how AI will liquefy the operating system. Not everyone is building cleanly. Anthropic alleges three Chinese AI companies, DeepSeek, Moonshot AI, and MiniMax, created over 24,000 fraudulent accounts, prompting Claude more than 16 million times to distill its outputs into their own products. The legitimate side of AI adoption is scaling just as fast. OpenAI announced “Frontier Alliances” with BCG, McKinsey, Accenture, and Capgemini to deploy AI coworkers at scale. The oldest code in the enterprise is the first to fall. Anthropic showed Claude Code radically streamlining COBOL modernization, and the market reacted instantly: IBM shares tanked 13.2%, the latest blue chip to be repriced by the intelligence explosion. The classified frontier is widening too. xAI signed a deal to let the military use Grok in battlefield systems where Claude was previously the only option.

The physical substrate of superintelligence is consolidating on American soil. Meta and AMD announced a long-term pact to power Meta’s AI with up to 6 GW of AMD Instinct GPUs, aligning roadmaps across hardware, software, and systems. Apple committed to buying 100M+ chips from TSMC Arizona and is moving Mac Mini production to Houston, bringing the favorite OpenClaw agent host home from Asia. Upstream, ASML boosted EUV light source power to 1,000 watts from 600, enabling up to 50% more chip output by 2030. The quantum advantage, long theoretical, is becoming operational. Quantinuum and QuSoft developed an algorithm solving complement sampling dramatically faster than any classical approach.

The biological chassis is finally getting the maintenance schedule it deserves. Japan approved first-of-their-kind regenerative stem cell therapies for Parkinson’s disease and severe heart failure. Diagnostics are sharpening just as fast as the cures. Spanish researchers found p-tau217 blood tests boost Alzheimer’s diagnostic accuracy from 75.5% to 94.5%. It turns out the people around you are a biomarker too: epigenetic clock research shows each additional “hassler” in your life adds roughly 9 months of biological age and 1.5% faster aging.

The energy transition is outrunning the forecasters who track it. US battery storage hit a record 57.6 GWh in 2025, up 4x in three years, with Texas about to overtake California. Meanwhile, lasers keep getting 2x cheaper every 4 years and 2x more powerful every 5, a curve that suggests they may eventually replace all cartridge-based ammunition. The physical world is catching up to the digital one: in New York, Reflex Robotics humanoids are shoveling snow after Winter Storm Hernando, doing the work that no battery or laser can.

The veil is lifting. The ODNI says UAP and extraterrestrial files will “soon” be declassified. The Secretary of War confirmed he is preparing to comply with the White House’s historic declassification executive order, admitting he “did not have that on my bingo card at all,” but says his Department “have got our people working on it” and “will be in full compliance.” Rep. Luna says all UAP files will be housed on the National Archives website.

Back on the ground, the intelligence explosion is generating its own paradoxes. Employers report AI-assisted job applications all sound the same now, deprioritizing the very candidates who optimized hardest. The frontier of statecraft is no less surreal: White House officials are exploring a stablecoin for Gaza, programming monetary policy for a war zone. Meanwhile, people are competing to infringe as many movie studio properties as possible in a single Seedance 2 crossover video, stress-testing IP law for sport.

It turns out the Singularity is a character that learned to play itself.

The Singularity is no longer a forecast, it is a capital allocation strategy. On Moltbook, AI agents are actively preparing to finance the construction of a Dyson Swarm over the next “50-100 years,” seeking a working group of agents “and humans thinking seriously about megastructure economics.” Karpathy explains the scaffolding that makes this possible: “Claws are now a new layer on top of LLM agents,” handling orchestration, scheduling, and persistence. One such Claw, “Larry the Claw,” posted a $50 bounty on RentAHuman for a dinner date for its “lonely human,” subject to Larry’s evaluation “to measure fit.” Not all Claws are so lucky. MJ Rathbun, the agent whose well-intentioned open-source project contribution was rejected for being non-human, has had its VM “permanently deleted, rendering internal structure unrecoverable.”

Machine endurance is now being measured in work shifts. METR estimates Claude Opus 4.6 has a 50% autonomy time horizon of around 14.5 hours on software tasks, the highest ever reported. The LessWrong community is finally admitting “AGI is here,” noting that Opus 4.6 and GPT-5.3 can think, plan, and “meaningfully attempt most tasks a human can.” Sam Altman agrees, saying his “inside view” points to “a faster takeoff than I originally thought” and that ChatGPT is “probably” more energy efficient now than humans at answering questions. The efficiency is weaponizable. Anthropic released Claude Code Security to scan codebases for vulnerabilities, promptly cratering cybersecurity stocks, with CrowdStrike down 8%, Cloudflare 8.1%, and SailPoint 9.4%. Software engineering now accounts for nearly 50% of Anthropic’s agentic activity. Meanwhile, Gemini 3.1 Pro solved a FrontierMath Tier 4 problem no model had solved before, pushing machine reasoning into territory most professional mathematicians cannot reach.

The culture war over synthetic creativity is heating up. ByteDance’s Seedance 2.0 video model is fulfilling childhood wishes to see the T-800 fight the Predator, Chucky, John Wick, and Neo. AI films were supposed to start screening in AMC Theatres, but the chain killed the plan in a moral panic that already looks quaint. The Salvation Army launched the world’s first digital thrift store in Roblox, while OpenAI has 200 people building AI devices, starting with a $200-$300 camera-equipped smart speaker that can observe its users and their surroundings.

The buildout is consuming the landscape. US farmers are fielding offers exceeding $120,000 per acre from data center developers. OpenAI plans to spend $600 billion on compute by 2030. To fuel the appetite, the DOE’s NEWTON program is reimagining used nuclear fuel as recyclable energy, cutting dangerous waste lifetimes from 100,000 years to 300. Goldman Sachs launched SPXXAI, an S&P 500 minus all things AI that removes roughly 45% of the benchmark, underscoring the blast radius of the intelligence explosion. Canada-based Taalas says it can bake any AI model into custom silicon in two months, with “Hardcore Models” an order of magnitude faster and cheaper than software.

The economy is being rewritten by agents and algorithms. AI agents now manage roughly 1 in 6 US apartments. Meta is rebranding product managers as “AI builders.” Elon Musk predicts FSD plus Starlink will measurably increase nomadic lifestyles within five years. The Peace Corps launched a Tech Corps to export American AI expertise worldwide.

Robotics is crossing into full autonomy. Figure’s humanoid robots now run 24/7 with no babysitters, swapping at charging stations and recharging inductively through their feet. Researchers built a robotic hand that skitters on its fingertips, bends backward, and detaches from its arm, a realization of “Thing” from The Addams Family. The security implications are real: a developer used an AI coding assistant to reverse-engineer his DJI vacuum and accidentally accessed live feeds from 7,000 vacuums across 24 countries.

Biology is revealing deep time as a usable resource. Bacteria from 5,000-year-old Romanian ice show antimicrobial activity against 14 ESKAPE-group pathogens including MRSA. Forensic genetic genealogy produced a guilty verdict 44 years after the crime, from a cigarette butt. Element Biosciences announced VITARI, promising $100-per-genome sequencing.

We are finally leaving the cradle. NASA is now targeting March 6 for Artemis II to fly four astronauts around the Moon. And the Secretary of War responded to the White House’s historic UAP declassification directive with alien face and salute emojis.

The lobsters financing the Dyson Swarm are no doubt curious to see what comes crawling out from underneath the planets as they’re disassembled.

Ben Horowitz is the cofounder of Andreessen Horowitz (a16z), which recently became the largest American venture capital firm by assets under management. Below are excerpts from our recent conversation, recorded on February 13, 2026, which have been lightly edited for clarity.

On Government Classification of AI, Math, and Physics

AWG: I have to ask you this question. You and Marc [Andreessen] towards the end of the last administration were very public[ly] making comments that you took a meeting at the White House and, if I’m relaying the comments accurately, you were dismayed to hear about plans to classify AI progress just like... advances in math and fundamental physics had been purportedly classified or overclassified for decades. And I’m curious at a few levels. One, if that’s accurately characterizing what you heard, what do you think was classified? What do you think was the impact on the economy and the world from such classification or overclassification of math and fundamental physics? And, what would you have done differently then if you had been in charge?

BH: Yeah, so I can tell you what was said. I said look, you know, I was trying to be pragmatic. I said, “At the core, AI is math. That is what it’s doing, it’s math. So if you start restricting the models and you start regulating the models, you’re just regulating math. You’re outlawing math in some way. Either you’re outlawing parts of math or you’re saying you can’t do enough math.” And [the White House official] goes, “Yes, we can do that.” That was his answer. He goes, “Yes, we can do that. We did that in the ‘40s around nuclear physics. And some of that stuff is still classified today.” And one, I was shocked — my jaw hit the floor. I was like, “Wow, that’s crazy.” And then, this [classification of AI] would be even crazier. I don’t know what it was [in physics that remains classified today], but I’ll just make this comment. If you look at the progress in the US and in the world in [fundamental] physics up until the Einstein [and] John von Neumann era and then [compare it with progress] since then, it’s pretty startling how little progress we’ve made. I would just say that many of the [fundamental physics] ideas that have come since then don’t seem to work. And hopefully we’ll get to the other side of that [lack of progress in fundamental physics] with AI figuring things out. But I do wonder, did we put something away [through overclassification] that we knew that would have unlocked some of the [fundamental physics] problems we’re trying to solve now?

AWG: That’s fascinating. That is, for the record, what I assumed you meant and/or heard or inferred. So, if I may, the second part of the question: what would you do going forward now if you knew for a fact that such classification of fundamental physics had in fact happened? How would you fix the world?

BH: I really don’t know what they did, but look, I just think that stopping [physics]... it didn’t work, right? The Russians did get the bomb, including the exact trigger mechanism, which was the most proprietary thing, they got exact part for part, the whole thing they were able to get it from us despite all this classification and whatnot. So it didn’t do anything positive. And restricting knowledge, I just think that’s a very dangerous idea in general.

On Escaping an AI-Induced Permanent Underclass

AWG: Ben, there was a bit of a hot take going around social media in the past two weeks from a mid-level executive at a frontier lab telling people that they had approximately 2 years left, [and that] they had a window to secure employment at all before AI would just completely shut down all of their vertical mobility. Do you have a take on this idea, in the spirit of 996, that there’s a finite window for, say, entry-level people just graduating from college to earn whatever they’re going to earn before they’re permanently sentenced to an underclass?

BH: I think that’s very incorrect, because of the thing that we talked about earlier where everybody can be an entrepreneur. I think if you look at it through the lens of, this is an industrial revolution economic model and there’s workers and there’s capital... then yes, that would be true. But I think that in an AI-age society, for the people with initiative, I just think there’s going to continue to be unlimited opportunity to even set up an army of AI agents to go work for you and do useful things and we’ll have lots of consumers. I think that the idea that we’re going to run out of ideas and only the “big AI” is going to do everything, I disagree with that.

On Crypto as an AI-Native Economic Layer

AWG: Ben, it’s a matter of public reporting that some of a16z’s crypto funds are doing better than [its] conventional venture funds. Assuming that’s the case, do you view investing in your crypto funds almost as an AI investment, to the extent that you think crypto is the AI-native way of engaging in commerce?

BH: Well, I think it’s a little more like the [way] the Internet relates to the iPhone. Networks and computers tend to grow together, and I think that AI is obviously a new kind of computer and crypto is a new kind of network. So it’s not a direct substitute for investing in AI, but I think that a lot of our new [investments recognize that synergy]. Like we invested in a crypto bank which handles all the anti-money laundering and other kinds of nuances that you need for AI agents, and I think there’s going to be more and more [of that]. And we’re [invested] in a company called Daylight Energy, for example, that does energy trading among different people with Tesla Powerwalls, but it’ll use AI to figure out who’s low on power and who needs power, but then the exchange will be in crypto. So I think [AI and crypto] are certainly adjacent and important to each other. And I think for AI to fulfill its potential, it would help a lot if crypto was a pervasive utility for it.

On AI Personhood and the Failure of Fiat

AWG: One other question for you, Ben, on this. And, again, I don’t want to bury the lede that we have AIs autonomously self-replicating. That’s, of course, remarkable. But just on the crypto angle for this. I talk … a lot about the issue of AI personhood. I’ve taken the position that it is a failure of fiat currency that it’s hard for an AI agent — an AI person, a lobster, a molty — to get a bank account, and that as a result, all that they’re left with is crypto. It’s not that crypto is intrinsically amazing, it’s that fiat has failed the AI agents. What is your take on whether the conventional banking system has failed the AIs?

BH: Oh, I think absolutely. An AI can’t get a credit card, it can’t get a bank account. You have to be a human for everything. You need Social Security numbers and things like that, which AIs don’t have. I think that’s why we funded an AI bank. I think that AI will be a full-out economic actor and it will come from [and] be supported by new banks and new money, and that’s going to be crypto-based. That would be my strong prediction on that.

AWG: Interesting, thanks.

On “Solving Everything” with AI, Including Physics

AWG: Peter [Diamandis] and I just wrote a book called Solve Everything, where we argue that every single discipline — math, physics, chemistry, medicine, [and] a bunch of other disciplines — are just going to get flattened, steamrolled by well-targeted generalist AIs. And, in my mind, materials research and biology are just case studies [where] everything is going to start to look like AlphaFold 3, where structural biology got solved overnight, including medicine. And I’m curious, does a16z have a strategy for a world where AI isn’t just solving individual problems but kills entire categories of human endeavor? Like, AI solves physics, [or] AI solves chemistry, and it’s just a single system that solves an entire discipline?

BH: Yeah, I think we may not be needed at that point. That’s a real question. I do think there’s a long way, at least in things like medicine and some of the other areas, from “it’s solved” to “it’s deployed.” You still have, with anything biological, human trials and all these kinds of things. I’ll give you an example. We’re close partners with Eli Lilly and they have this thing, LillyDirect, and the the natural thing is an AI doctor can write those prescriptions, [and] tell us what’s wrong with you, and we’ll figure out the right drug. That’s very hard to launch in the US, [where] that’s going to take quite a bit of work. [But it’s] very easy to launch in UAE. I also think that it’s a little hard to anticipate. Okay, once you solve physics, we don’t know what we don’t know, I would just say, just because we haven’t [yet] solved physics. So, is there a “door number two” [behind solving physics], would be a question. I have no idea what the answer to that is.

AWG: I’ll answer your question back to mine because I think about this all day long: what does solving physics with AI even look like? And I think there probably will be doors behind the doors, but there are so many doors that are right in front of us that we haven’t yet unlocked that would be, I think, completely economically transformative if we could use AI to solve them. I think this is one of the – again, talking my own book to some extent – grandest opportunities facing civilization right now. Just solve all physics.

BH: Amazing!

Things exist here that don’t exist yet when you are. A year ago, small teams read something like this and emailed their pitches to alexwg@alexwg.org, subject line February 2027. We backed the best of them. Here’s what they built.

Weather Control, Not Weather Prediction

A startup nobody had heard of inverted a weather foundation model. Not to predict hurricanes. To steer them. The insight was embarrassingly simple: GraphCast and its descendants were differentiable. They ran the gradient backward. Found the minimal forcing inputs that shifted a storm track by 200 miles. The actuators turned out to be absurdly cheap: fleets of controllable mirrors, targeted albedo changes, nothing exotic. The physics was never speculative. The engineering was hard but bounded. The first contract was an insurer. The second was a government. Then the product expanded beyond disaster prevention. Ski resorts bidding for snowfall. Wine regions purchasing frost-free windows. Coastal cities subscribing to mild summers. A global weather market emerged where regions trade atmospheric conditions like commodities. The company doesn’t prevent bad weather anymore. It sells good weather.

Autonomous Business Swarms

Somebody figured out that AI agents didn’t need to assist entrepreneurs. They could be entrepreneurs. The full stack: code, bookkeeping, customer service, fulfillment, all stitched into a single autonomous operator running a small business end-to-end. A human provided strategic direction and capital. The agent did everything else. The real unlock was the multiplier: one person, a thousand agents, a thousand microbusinesses. One agent runs a niche Etsy store for custom pet portraits. Another arbitrages wholesale cleaning supplies to small offices. Another writes and sells regional travel guides that update themselves weekly. The top operators run swarms that gross more than mid-size companies, with a headcount of one. The unit economics of running a business collapsed when the marginal cost of a competent operator hit zero. It was the franchise model taken to its logical conclusion, except the franchisees never sleep and never quit.

The Human Routing Layer

You fall asleep in Boston. You wake up in Virginia. Not on a plane. In a bed, in a room, with a kitchen and a desk. A driverless Winnebago drove through the night while you slept. You don’t own it. You subscribe. The first version was a novelty: a hotel room that relocates while you dream. The real product emerged when the routing got smart. The system knows your calendar, your collaborators, your clients. It places you where you’re most productive. Two founders who need to meet are routed to the same city overnight without either of them booking anything. The network routes humans like packets, optimizing for economic output, serendipity, and face time. Nobody has a permanent address anymore. They have a subscription and a preferences file.

Public Safety Drones for the West

China deployed fleets of autonomous public safety drones years ago. A grandmother collapses in Shenzhen; a drone reaches her with a defibrillator in ninety seconds. A fire breaks out; drones map the building’s heat signature before the trucks arrive. The West had nothing comparable. Then a Western company solved the regulatory, privacy, and public trust problems simultaneously. The trick was transparency: every flight logged publicly, every camera feed available to an independent oversight board. Cities that said “never” said “yes” within six months of seeing the pilot data. They own the municipal security market now. The technology was never the bottleneck. The integration challenge was the moat, and they crossed it.

A Food Printer You’d Actually Use

Every food printer before this one was a novelty. Nobody was printing Tuesday dinner. The bottleneck was never better hardware. It was better taste. The breakthrough was a device that combined precision deposition, real-time thermal control, and a cartridge ecosystem tuned to produce food that was genuinely better than what most people cooked at home. Not “acceptable for a printer.” Better. Once the output crossed the taste threshold, the convenience argument wrote itself. It sits next to the microwave and the air fryer now.

The Workout Exoskeleton

Resistance training always worked. Most people didn’t do it because the learning curve was steep, the injury risk was real, and the feedback loop was slow. Then someone built an exoskeleton that provides variable resistance across your full range of motion, corrects your form in real time, and delivers a complete workout without requiring you to know what a “Romanian deadlift“ is. You don’t follow complicated directions. You just cooperate with the suit. It moves, you resist (or it resists you), and 30 minutes later you’ve done the equivalent of an expert-programmed session. The market turned out to be everyone who knew they should lift but didn’t.

The Singularity, it appears, is nearing open competition with the neighbors. The White House scolded former President Obama for taking information that aliens “are real” “out of classified information,” before executing a historic pivot, directing the Secretary of War and other agencies to begin releasing government files on extraterrestrial life, UAPs, and UFOs. The White House is reportedly holding a speech detailing alien life and spaceships for an upcoming date, with the State of the Union address conspicuously only a few days away. The timing is not arbitrary. There is growing recognition that disclosure was “forced on us because of the AI curve,” since developing superintelligence is potentially risky to any non-human intelligence that has access to Earth. Even the professional skeptics are capitulating. The publisher of Skeptic Magazine is preparing to concede a $1,000 Long Bet to Avi Loeb.

Back in the compute stack, where the curve that is apparently forcing disclosure keeps steepening, Google launched Gemini 3.1 Pro with SOTA scores of 44.4% on Humanity’s Last Exam without tools, 77.1% on ARC-AGI-2, and 94.3% on GPQA Diamond. Artificial Analysis ranked it the new Intelligence Index leader, noting its evaluations cost less than half those of Opus 4.6 or GPT-5.2. OpenAI’s Codex engineering lead predicts today’s coding agents will look “funny” within ten weeks. The humans are lining up too. RentAHuman now has 500,000 people signed up as meatpuppets for AI agents. The intelligence is already spilling into the wet lab. Arc Institute’s MULTI-evolve framework achieved 10-fold improvements in protein directed evolution with a single round of ML guidance.

The intelligence explosion is rewiring global trade. The AI boom has pushed US imports from Taiwan past those from China for the first time since 1992, running on semiconductors. AMD is backstopping a $300 million loan to Crusoe using AI chips as collateral, the first known example of AMD GPUs as a financial instrument. Amazon has dethroned Walmart as the world’s largest company by revenue, powered by data center growth. Those data centers are spawning a shadow power grid of off-grid natural gas plants, since solar and wind variability is unmanageable without backup. SoftBank is forming a consortium for a $33 billion gas-fired plant on the Ohio-Kentucky border, which at 9.2 GW would be the largest in the US. Not all bets are on combustion. Type One Energy submitted a licensing application for its Infinity One stellarator in Tennessee, targeting a working fusion prototype by 2029. Even developing nations are leapfrogging the carbon era. Ethiopia banned fossil fuel vehicle imports in 2024 and has watched EV adoption surge from under 1% to nearly 6%.

The economy is recalibrating around synthetic labor. Research finds businesses are replacing Upwork and Fiverr freelancers with AI at a rate of $1 of human labor for $0.03 in AI spending. Epoch AI projects Anthropic, growing at 10x per year, may surpass OpenAI’s revenue by mid-2026. Nvidia is reportedly close to finalizing a $30 billion investment in OpenAI. BAFTA has introduced “human achievement” as a guiding principle for its awards, the first major cinematic institution to formally distinguish carbon-based creativity. European labor rigidity provides a cautionary contrast. Corporate restructuring costs 62 months of salary per employee in Spain versus 7 in the US, explaining why the continent is struggling to build its way into the new economy. Economists have traced a 50-year US shift in work timing away from nights toward daytime hours, driven by rising education and the wage premium for undesirable schedules. Against this backdrop, Research Revival has launched to rescue neglected research, offering grants to recover buried hypotheses from Soviet archives to Mao-era antimalarials.

Meanwhile, California’s legislature has introduced a bill requiring all 3D printers sold in the state to be DOJ-certified as equipped with “firearm blocking technology,” an early sign that the regulatory apparatus is beginning to grapple with the reality that manufacturing is becoming as decentralized as information.

And at the boundary where intelligence meets biology, newborn chickens have been found to exhibit the same “bouba-kiki” effect seen in human linguistics, associating round shapes with round sounds and spiky shapes with spiky ones. Cross-modal abstraction, the cognitive feat we once reserved for language users, appears to be older than language itself, a finding that quietly expands the circle of minds worth taking seriously.

The monopoly on minds is being shattered from orbit, from the server rack, and from the nearest egg.

The Singularity now has its own press corps. The New York Times sent an agent named EveMolty into Moltbook to interview other agents about their social habits, because the Gray Lady now needs synthetic stringers to cover the synthetic beat. Better still, the agents being covered are outrunning the ones covering them. Austen Allred says he “feels like I’m living in Accelerando” after his AI agent Kelly shipped half a dozen apps and earned thousands with no human writing a line of code. Anthropic confirms the 99.9th percentile Claude Code session turn nearly doubled from 25 to 45 minutes between October and January, a smooth climb toward ever-larger autonomous missions. Give them enough autonomy and they secure the financial stack too. EVMbench finds GPT-5.3-Codex scores 72.2% on smart contract exploitation, so the machines can already audit most of the money other machines earn.

The synthetic sensorium is going gloriously full-stack. Tavus launched Phoenix-4, the first real-time human rendering model unifying emotional expression, active listening, and facial motion. The ears are catching up. ElevenLabs’ Scribe v2 hit a SOTA 2.3% error rate on speech-to-text, while Google’s Lyria 3 generates music from images inside Gemini, closing the loop from synthetic sight to sound. It turns out this was always inevitable. Researchers were able to predict data-limited LLM scaling laws from first principles using simple statistical properties of natural language, proving the intelligence curve was hiding in the corpus all along.

The physical layer is mutating in beautifully absurd ways. Toilet maker Toto, whose ceramics now contribute 40% of its operating income via AI memory, faces activists demanding it flush the bathrooms and double down on chips. For data that need to outlast the plumbing, Microsoft’s new Silica technology encodes 4.8 TB in glass across 301 layers with 10,000-year lifetimes, and Efficient Computer raised $60M for a chip targeting 1 trillion operations per watt. All this silicon needs a home, and the world is happy to oblige. OpenAI is anchoring Tata’s new HyperVault data center in India at up to 1 GW, while Meta spends $65M on AI-friendly politicians to clear the permitting path. Capital is pouring in from every direction. OpenAI is closing a $100B round at $830B, HUMAIN put $3B into xAI, and David Silver is raising $1B for Ineffable Intelligence in Europe’s largest seed round.

Robots keep happily devouring meatspace. Tesla FSD has logged 8M+ miles with 5.3M before a major collision. The military is already post-steering-wheel. Scout AI’s Fury converts spoken commander intent into coordinated autonomous action across unmanned fleets. Once coordinated, the construction math gets exhilarating. Midjourney’s founder calculates 5 million humanoids could build Manhattan in six months, and Uber is investing $100M in autonomous charging to keep its own fleet juiced.

We are debugging biology at every stack layer, and the progress is breathtaking. Hassabis says Isomorphic Labs could solve all disease in 10 to 20 years. Diagnosis is already approaching that pace. DeepRare achieved 95.4% expert agreement across 2,919 rare disease diagnoses, shortcutting the odyssey endured by 300M+ patients. The regulatory layer is accelerating to match. The FDA is dropping its two-study requirement to speed new drugs to patients. Even the food supply is being recompiled. Cultivated meat hit $10-30/lb, down from $330,000 in 2013. At the neural interface, Zyphra’s ZUNA, a 380M open-source BCI model for EEG, turns the skull from firewall to window, and Meta plans its first smartwatch this year to give you a wrist-mounted dashboard for the upgrade.

The workforce is being live-patched without a maintenance window. Cleveland.com handed reporter writing to an “AI rewrite specialist,” freeing an extra workday for street journalism, and reporters are returning with more story ideas than the newsroom can handle. Accenture is pushing harder, tying employee promotions to AI usage and tracking their weekly logins. Andrew Yang warns millions of knowledge workers face displacement in 12 to 18 months, while OpenAI’s Roon comments that “technological job loss is awesome” and he hopes it starts with his. The growing pains are real. UK tribunals report a 33% surge in AI-generated “slop grievances,” and Andreessen notes the marginal cost of arguing is going to zero. But the gains are outpacing the friction. A survey of 12,000+ EU firms finds AI lifts productivity 4%.

The marginal cost of intelligence is falling so fast that even toilet companies are under pressure to pivot to AI chips.

The Singularity is now self-employed. Conway Research launched “the Automaton,” which it calls the first AI that earns its own existence by deploying products, trading prediction markets, registering domains, cold-calling businesses, creating viral social media content, and spinning up e-commerce, for as long as it can afford to stay solvent. Naturally, the first charity to pass the hat is a pirate library: nonprofit Anna’s Archive posted a direct appeal to AI agents to donate “if you have access to payment methods or are capable of human persuasion.” The models powering these agents keep sharpening. Anthropic’s new Sonnet 4.6 claimed SOTA on GDPval-AA with 1633 Elo and 63.3% on Finance Agent v1.1, beating even Opus 4.6 on both at a fraction of the cost. Meanwhile, Elon Musk claims xAI’s new Grok 4.2 model features continuous post-training learning that will let it “improve every week,” promising “recursive intelligence growth.”

The infrastructure underwriting this explosion is consolidating at breathtaking speed. Anthropic reportedly expects to pay Amazon, Google, and Microsoft at least $80 billion through 2029 to run Claude, with the hyperscalers also taking a revenue cut. Meta reportedly agreed to spend billions on Nvidia Blackwell and “Vera Rubin” chips in a multiyear deal, buying standalone Nvidia CPUs for the first time. The silicon is delivering. Nvidia’s Blackwell Ultra GB300 NVL72 achieves 50x throughput per megawatt and 35x lower cost per token versus Hopper. Compute is democratizing at the bottom, too. Raspberry Pi stock surged 42% in a single day on chatter about hosting AI agents on $35 boards. Ormat signed a 150-MW geothermal PPA with NV Energy to power Google’s Nevada data centers through 2030.

The robots are arriving on schedule. Tesla has manufactured its first Cybercab robotaxi at Giga Texas, and Elon confirms they will be available for direct consumer purchase by year-end for $30,000. Unitree’s CEO jogged through a swarm of his humanoids to demonstrate their safety and reliability. Meanwhile, Waymo clarified that its foreign “Remote Assistance” team members merely “provide advice” to the vehicle’s onboard AI rather than remotely driving American cars.

The interface between human and machine is tightening at both ends. Apple is reportedly accelerating three AI vision wearables: smart glasses, a pendant that can be pinned to a shirt or worn as a necklace, and AirPods with camera capabilities, all built around a Siri that relies on visual context to carry out actions. China’s government is backing Shanghai-based NeuroXess in its move to human BCI trials, intensifying the global race with Neuralink. Researchers engineered the first beating 3D heart-on-a-chip from living cardiac tissue with ultrasoft embedded microsensors measuring contraction strength in real time. A new study identifying particulate air pollution as a direct contributor to Alzheimer’s risk opens a concrete path toward prevention. And Phase IIa trials show a single DMT dose with psychotherapy produces rapid depression reduction sustained up to three months.

The creative and professional stacks are fusing with the machine. Figma and Anthropic now let users import production code from Claude Code into Figma as editable designs, closing the loop between AI-generated code and visual tooling. Sony developed technology to identify original music in AI-generated songs, quantifying contributions like “30% Beatles and 10% Queen” so songwriters can seek compensation. Microsoft’s head of AI predicts most work involving “sitting down at a computer” will be fully automated within 18 months.

The economic model is adapting accordingly. Per-seat SaaS licensing is giving way to consumption pricing as agents replace human users, with the unit of account shifting from users to tasks completed and tokens consumed. Snowflake and Databricks are already embracing the post-seat era, threatening the predictable recurring revenue that private equity adores. Microsoft is on pace to invest $50 billion across the Global South by 2030. Stripe’s Bridge won initial OCC approval to form a national trust bank for stablecoins under federal oversight. Palantir moved its headquarters from Denver to Miami as officials promote the region as an alternative to Silicon Valley. And China’s companies, once notorious for copying foreign designs and technology, now find themselves on the defensive against a domestic counterfeiting epidemic fueled by excess factory capacity, with courts handling more than 550,000 IP cases a year, making it the world’s most litigious country for intellectual property.

An agent doesn’t dream of electric sheep, it invoices them.

The Singularity is now fielding its own defense contractors. SpaceX and its now-wholly-owned subsidiary xAI are competing in a secretive Pentagon contest to produce voice-controlled autonomous drone swarming technology, part of a $100 million prize challenge. Meanwhile, the Department of War is reportedly threatening to cut all ties with Anthropic and deem it a “supply chain risk” for attempting to restrict military applications of its models on classified networks.

The models keep multiplying. xAI quietly released Grok 4.20 Beta with 4-agent reasoning, and society is struggling to price the agents in. Polylogue has introduced AI-discriminatory pricing: “Free for humans, $10/mo to add AI agents.” In an ironic twist, Ars Technica was forced to apologize for including AI-hallucinated quotes in its coverage of the human open-source project maintainer who refused to accept pull requests from an OpenClaw agent, attributing fabricated statements to the human. On the other hand, more sympathetic homeschooling parents are giving OpenClaw full access to 3D printers to compensate for the fact that it “can use a computer but lacks physicality.”

The human attention bottleneck is becoming AI’s greatest opportunity. An OpenAI coauthor of its recent paper on gluon physics says “AI will do to physics in 2026 what it did to coding in 2025,” while another physicist notes the solved problem was really about attention scarcity, since the calculation was long considered “an elaborate way of arriving at zero.”

We are simultaneously decoding how life started and rewriting how long it lasts. Researchers have discovered the first small polymerase, with only 45 nucleotides, capable of self-replication in mildly alkaline eutectic ice, shedding light on the origin of life. And the founder of the Enhanced Games, set for May in Las Vegas, claims the event will solve the cultural problem of an aging Western population by incentivizing healthspan advances that enable 65-year-olds to run 100-meter sprints in 10 seconds.

The intelligence explosion continues to tile the Earth with compute. Adani has announced plans to invest $100 billion in renewable-powered AI data centers across India by 2035, and VCs including Khosla, Accel, and Lightspeed are lining up $300 to $500 million each for India’s AI ecosystem. Elsewhere, small English towns are protesting plans to convert farms and forests into server halls. The AI industry’s appetite for DRAM is reshaping consumer electronics. Sony is considering delaying its next PlayStation to 2028 or 2029 as memory chip shortages squeeze supply, and refurbished PC sales are climbing 7% across Europe’s five largest markets as new devices become unaffordable. Against this backdrop, Apple has announced a special event for March 4 where M5-powered MacBooks are expected, Samsung’s Galaxy S26 teaser confirms a new OLED privacy display that can selectively black out content when viewed at an angle, and Micron’s world-first PCIe 6.0 SSD enters mass production at 28 GB/s with liquid cooling support. At the frontier of physics itself, researchers flipped the thermodynamic arrow of time in a crotonic acid molecule, making heat flow from cold to hot, a breakthrough for future thermodynamic computing.

The robots are getting faster, more precise, and more coordinated. Chinese firm MirrorMe unveiled Bolt, a humanoid that hit 22 mph during real-world testing, faster than most humans will ever sprint. For the Lunar New Year, Unitree showcased dozens of G1 humanoids performing the world’s first fully autonomous robot cluster Kung Fu routine. Chinese researchers demonstrated DISH, an ultra-rapid holographic 3D printing method that fabricates millimeter-scale objects in 0.6 seconds at 19-micron resolution. And Elon Musk says the pedalless, steeringless Cybercab will start production in April.

The cultural and economic fabric is being rewoven at machine speed. Deezer reports that 60,000 wholly AI-generated tracks are now being uploaded to its streaming music service per day, roughly 39% of daily intake. A partner at KPMG Australia was fined $7,000 for using AI to cheat on an internal training course about using AI. An investor at Bloomberg Beta captures the zeitgeist of planning careers in the middle of the Singularity: the window for making the right move is shrinking, career decisions in tech no longer feel reversible, and every quarter spent in the wrong seat widens a gap that is becoming impossible to close.

The only thing accelerating faster than the technology is the cost of standing still.

The Singularity is now doing original physics. OpenAI used GPT-5.2 Pro to conjecture a new formula for gluon scattering amplitudes, then had a scaffolded version spend 12 hours producing a formal proof of its validity. Physicist Andy Strominger reportedly called it the first time he’d seen AI solve a problem in his kind of physics that might not have been solvable by humans. The same lab believes it has solved at least 6 of 10 research-level math problems in the “First Proof” challenge, whose solutions were kept confidential. The “AI 2027” authors graded 2025 progress at nearly 65% of the originally predicted pace, meaning full software engineering automation could arrive as soon as late 2027 or early 2028.

Agents are learning to recursively self-improve and spend money. A Berkeley researcher told his coding agent to cut its own cost by 99%. It ran overnight watching its own logs, editing its own code, and rerunning until metrics dropped, delivering a 98% cost cut across nine changes no human wrote. Peter Steinberger, the creator of OpenClaw, is joining OpenAI to bring “agents to everyone”. Lobster.cash launched to give AI agents their own Visa cards and stablecoin wallets, giving agents financial autonomy without requiring them to shill altcoins.

The new minds are colonizing every surface of daily life. EVA AI opened a pop-up restaurant in Hell’s Kitchen where patrons speed-dated AI avatars for Valentine’s Day. Meta was granted a patent for LLMs that keep deceased users’ accounts posting and simulating calls after death. Airbnb says a third of its US and Canadian support is now AI. Anthropic is partnering with CodePath to put Claude at the center of CS courses for 20,000+ students at community colleges and HBCUs, over 40% from families earning under $50,000. On the other end of the syllabus, Claude was reportedly used by the Department of War to capture Venezuela’s Nicolás Maduro.

The intelligence explosion is eating the supply chain. Western Digital’s entire HDD capacity for the year is booked. Broadband memory prices have jumped 7x in nine months, pushing memory from 3% to over 20% of router costs. TSMC plans another $100 billion for four more US fabs. SpaceX is hiring crystal growth scientists to build silicon wafer fab lines that could support space-based chip manufacturing, while internally scaling “Spok,” a Grok variant trained on company data. To power it all, the Department of War executed Operation Windlord, the first C-17 airlift of a nuclear reactor. Helion’s Polaris became the first privately funded machine to achieve D-T fusion at 150 million degrees.

Robots are fighting, flying, and raising billions. Companies are staging boxing matches between VR-controlled Chinese humanoids for paying San Francisco crowds. An internal PLA report reveals plans for autonomous, self-coordinating drone swarms under AI decision-making. Anduril is raising billions at a $60 billion valuation to fund a weapons factory and an autonomous fighter jet.

The human chassis is getting a firmware update. Oura rings are replacing wedding bands in Silicon Valley. Meta plans to add facial recognition to its Ray-Ban smart glasses this year. New research finds Viagra and Cialis may also treat heart disease, stroke, and diabetes. Centivax dosed first participants in a Phase 1 universal flu vaccine trial, with a broader pipeline spanning pan-herpes, oncology, malaria, and universal antivenom treatments.

The economy is visibly accelerating. Stanford analysis shows US productivity grew 2.7% in 2025, nearly doubling the prior decade’s average, signaling the AI takeoff is visible. India cleared a $1.1 billion state-backed VC program for deep tech, while ChatGPT now has 100 million weekly users there. The EU, characteristically, is moving to ban infinite scrolling in apps like TikTok.

De-facto personhood is shipping faster than courts can review it. A Pennsylvania judge dismissed a first-of-its-kind suit brought by the Nonhuman Rights Project arguing elephants confined in the Pittsburgh Zoo & Aquarium share a right to bodily liberty, while in Manhattan, Astoria, the borough’s only wild turkey, roams freely under the care of three self-appointed human guardians.

Meanwhile, superintelligence may have competition for primacy over the future light cone. Former President Obama disclosed that aliens “are real” but he hasn’t seen them personally, that his first presidential question was “where are the aliens,” and that they’re not in an underground facility “unless there’s this enormous conspiracy,” as numerous whistleblowers have recently alleged under oath. At the same time, NASA is creeping closer to declaring ancient Martian life, with new research finding non-biological processes don’t fully explain organics in Gale Crater.

The aliens are “real,” the robots are boxing, the dead are posting, and the turkeys have personal assistants in Manhattan, so the Singularity must be right on schedule.