{"schemaVersion":"1.0.0","asOf":"2026-08-23","title":"Frontier thesis map","purpose":"Decision-grade context on the public and revealed AI theses of consequential builders, allocators, researchers and governors.","editorialPolicy":{"principle":"Separate quoted statements from inference. Weight costly, observable commitments above adjectives.","caveats":["Coordinates, forecast-utility scores and revealed positions are editorial judgments.","A revealed position is an inference from capital, product, governance, hiring or career choices—not a quotation.","Utility measures decision value, not honesty or moral worth.","Dates are preserved because thesis movement matters more than a timeless summary."]},"keyJudgments":[{"label":"Capability","confidence":"High","title":"Scaling remains necessary. It is no longer a sufficient thesis.","detail":"Compute commitments keep rising while Sutskever, LeCun, Li and embodied-system builders converge on generalization, world models and interaction as missing pieces.","decision":"Favor teams with a learning loop, not merely model access.","breaksIf":"Brute-force scaling restores broad, reliable transfer across unfamiliar tasks."},{"label":"Power","confidence":"High","title":"The durable contest is stack ownership, not the chatbot leaderboard.","detail":"Alphabet, Microsoft, Meta, Alibaba, NVIDIA and AMD are closing loops across chips, cloud, models, distribution and data.","decision":"Map the control point and switching cost before pricing model advantage.","breaksIf":"Models commoditize faster than distribution and infrastructure can capture value."},{"label":"Access","confidence":"High","title":"Open is a strategy with thresholds—not an ideology.","detail":"DeepSeek and Qwen use openness for ecosystem formation; Meta’s posture moves with product control; Anthropic draws a capability boundary.","decision":"Track what is released, when, and under which constraints—not the word open.","breaksIf":"Frontier open weights consistently destroy the releaser’s economic or security position."},{"label":"Diffusion","confidence":"Moderate","title":"Technical clocks are outrunning institutional clocks.","detail":"Measured task horizons are expanding, but workflow redesign, robotics reliability, regulation and organizational absorption remain binding.","decision":"Separate capability exposure from realized operating leverage.","breaksIf":"Autonomous systems deliver broad, audited ROI without process redesign."}],"decisionQuestions":[{"question":"Does capability generalize?","signal":"Unfamiliar, multi-day tasks completed without human rescue","now":"Improving; still jagged"},{"question":"Who owns the margin?","signal":"Inference economics plus durable workflow switching costs","now":"Infrastructure + distribution"},{"question":"Does openness hold at the frontier?","signal":"Weights released promptly after training, with reproducible evals","now":"Conditional"},{"question":"Has robotics crossed the demo gap?","signal":"Unscripted task success, field uptime and paid repeat use","now":"Not yet"},{"question":"Can restraint be verified?","signal":"Shared thresholds, incident disclosure and observable consequences","now":"Fragmentary"}],"players":[{"id":"amodei","name":"Dario Amodei","role":"CEO","organization":"Anthropic","region":"US","strategicLens":"Lab","currentThesis":"Very capable digital workers arrive soon; biology and software move first; catastrophic misuse and loss of control are live constraints.","timing":"Earliest case already inside the 2026–27 window; core claim is a short runway, not a single AGI date.","statedPosition":"Scale quickly, measure dangerous capability, and raise safeguards at explicit thresholds.","revealedPosition":"Secured up to 5GW and committed more than $100B to AWS technologies while repeatedly tightening a public scaling policy.","editorialReadThrough":"He is not pricing in a voluntary industry slowdown. The bet is that safety can be made compatible with winning the race.","thesisShift":"The capability forecast stayed aggressive; the public emphasis moved from abstract alignment to security, labor and enforceable reporting.","nextResolvingEvidence":"Whether capability thresholds trigger a costly delay when competitors keep shipping.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"Ten-year compute commitments measured in gigawatts and nine figures."},{"label":"Governance","strength":5,"evidence":"A versioned policy with thresholds, risk reports and non-compliance channels."},{"label":"Career","strength":5,"evidence":"Built a frontier lab around this thesis."}],"positionCoordinates":{"timingRisk":{"x":91,"y":82},"openPower":{"x":28,"y":35},"formValue":{"x":28,"y":76}},"positionHistory":[{"year":2023,"x":76,"y":80,"note":"Catastrophic-risk framing becomes operating policy."},{"year":2024,"x":91,"y":84,"note":"Powerful systems could arrive as early as 2026; upside thesis expands."},{"year":2025,"x":94,"y":85,"note":"ASL-3 protections make the risk claim operational."},{"year":2026,"x":91,"y":82,"note":"Still short-timeline; regulation and labor displacement move forward."}],"forecastUtility":{"score":88,"components":{"specificity":22,"falsifiability":20,"proximity":18,"backing":20,"record":8}},"sourceIds":["amodei-loving-grace","anthropic-rsp","anthropic-compute","anthropic-open"]},{"id":"hassabis","name":"Demis Hassabis","role":"CEO","organization":"Google DeepMind","region":"Europe","strategicLens":"Research","currentThesis":"General systems need planning and world models; the highest-value early use is accelerated science, not synthetic office labor alone.","timing":"Near enough to plan for this decade, but usually expressed as a range rather than a date.","statedPosition":"Build general capability responsibly and use it against root scientific problems.","revealedPosition":"AlphaFold is a globally used research tool; Alphabet is pairing DeepMind research with $175–185B in 2026 capex and an automated science lab.","editorialReadThrough":"The strongest evidence is not rhetoric or a benchmark. It is a repeatable research-to-deployment pipeline with one major scientific win already banked.","thesisShift":"Less about proving that scaling works; more about agents, simulation, automated experimentation and useful generality.","nextResolvingEvidence":"Whether the next AlphaFold-like system produces validated discoveries rather than better research workflow.","convictionSignals":[{"label":"Track record","strength":5,"evidence":"AlphaFold has broad documented research use and a Nobel-recognized result."},{"label":"Capital","strength":5,"evidence":"Alphabet is funding compute, distribution and frontier research at hyperscaler scale."},{"label":"Product","strength":5,"evidence":"Gemini is embedded across Search, Cloud and consumer surfaces."}],"positionCoordinates":{"timingRisk":{"x":75,"y":72},"openPower":{"x":42,"y":42},"formValue":{"x":54,"y":69}},"positionHistory":[{"year":2023,"x":62,"y":70,"note":"Google Brain and DeepMind combine to accelerate a responsible AGI program."},{"year":2024,"x":70,"y":74,"note":"AGI described as a few years away, with science as the primary proof path."},{"year":2025,"x":74,"y":73,"note":"AlphaFold’s adoption strengthens the science-first case."},{"year":2026,"x":75,"y":72,"note":"World models, automated labs and product deployment run in parallel."}],"forecastUtility":{"score":85,"components":{"specificity":17,"falsifiability":17,"proximity":20,"backing":20,"record":11}},"sourceIds":["deepmind-about","alphafold-impact","alphabet-capex"]},{"id":"sutskever","name":"Ilya Sutskever","role":"Co-founder","organization":"Safe Superintelligence","region":"Global","strategicLens":"Research","currentThesis":"Current recipes generalize poorly. The next jump is a research breakthrough, and the first true superintelligence must be aligned before release.","timing":"Longer and wider than the frontier-CEO cluster after the 2025 thesis change.","statedPosition":"Do one thing: solve safe superintelligence without product pressure.","revealedPosition":"Left the most valuable research seat in the field and built a company with no near-term product roadmap.","editorialReadThrough":"The costly signal is unusually clean. The evidence signal is not: outsiders cannot yet inspect the claimed technical path.","thesisShift":"The biggest update in the set: from predictable gains through scale to uncertain gains through research.","nextResolvingEvidence":"A public result that improves out-of-distribution generalization, not another benchmark peak.","convictionSignals":[{"label":"Career","strength":5,"evidence":"Exchanged a leading lab position for a single-mission company."},{"label":"Focus","strength":5,"evidence":"No product detour and no broad application portfolio."},{"label":"Disclosure","strength":1,"evidence":"Technical evidence remains deliberately private."}],"positionCoordinates":{"timingRisk":{"x":68,"y":94},"openPower":{"x":10,"y":25},"formValue":{"x":20,"y":82}},"positionHistory":[{"year":2023,"x":88,"y":91,"note":"Superalignment and rapid scaling dominate the frame."},{"year":2024,"x":80,"y":96,"note":"Leaves OpenAI and creates a safety-only lab."},{"year":2025,"x":66,"y":94,"note":"Declares the age of scaling over; weak generalization is the bottleneck."},{"year":2026,"x":68,"y":94,"note":"Research-first path remains; product evidence is intentionally scarce."}],"forecastUtility":{"score":82,"components":{"specificity":17,"falsifiability":14,"proximity":20,"backing":20,"record":11}},"sourceIds":["ilya-interview","ssi-mission"]},{"id":"li","name":"Fei-Fei Li","role":"CEO · Professor","organization":"World Labs · Stanford","region":"US","strategicLens":"Physical","currentThesis":"Language is not enough. Systems need spatial representations to perceive, reason and act in the physical world.","timing":"A staged research and product path, not a declared AGI deadline.","statedPosition":"Build world models that expand human creativity and embodied capability.","revealedPosition":"Founded World Labs, shipped a persistent-world product and expanded toward robotics simulation.","editorialReadThrough":"A narrower claim than AGI, with visible artifacts. That makes it easier to test and harder to inflate.","thesisShift":"Human-centered principles stayed fixed; the technical bet narrowed from broad vision research to spatial systems.","nextResolvingEvidence":"Robots trained in generated worlds transferring reliably into physical environments.","convictionSignals":[{"label":"Career","strength":5,"evidence":"Moved from field advocacy to a company built on one architecture bet."},{"label":"Product","strength":4,"evidence":"Persistent, navigable worlds are already available for inspection."},{"label":"Timeline","strength":2,"evidence":"Few dated forecasts limit calibration."}],"positionCoordinates":{"timingRisk":{"x":48,"y":59},"openPower":{"x":69,"y":72},"formValue":{"x":94,"y":57}},"positionHistory":[{"year":2023,"x":40,"y":66,"note":"Human-centered deployment remains the anchor."},{"year":2024,"x":46,"y":61,"note":"Spatial intelligence becomes the next-frontier thesis."},{"year":2025,"x":48,"y":59,"note":"Marble makes persistent generated worlds inspectable."},{"year":2026,"x":48,"y":59,"note":"Simulation and robotics become the downstream market."}],"forecastUtility":{"score":79,"components":{"specificity":16,"falsifiability":15,"proximity":18,"backing":18,"record":12}},"sourceIds":["worldlabs-frontier","worldlabs-marble"]},{"id":"lecun","name":"Yann LeCun","role":"Executive Chair · Professor","organization":"AMI Labs · NYU","region":"Europe","strategicLens":"Research","currentThesis":"Autoregressive language models will not reach robust machine intelligence; world models, persistent memory and planning are missing.","timing":"Human-level systems are not close on the current path; no hard alternative-path deadline.","statedPosition":"Stop anthropomorphizing fluent text prediction and build systems that learn the physical world.","revealedPosition":"Left Meta and raised a new organization around the architecture he had argued for inside FAIR.","editorialReadThrough":"The separation from Meta removes the largest stated-versus-revealed ambiguity in his profile. The alternative still needs a decisive empirical win.","thesisShift":"The thesis did not move. The institution did.","nextResolvingEvidence":"A world-model result that beats language-model agents on planning, adaptation or robotics with less data.","convictionSignals":[{"label":"Technical","strength":5,"evidence":"A multi-year architecture program with published components."},{"label":"Career","strength":5,"evidence":"Left a scaled lab to pursue the dissenting path independently."},{"label":"Deadline","strength":1,"evidence":"Strong critique, weak date discipline."}],"positionCoordinates":{"timingRisk":{"x":15,"y":25},"openPower":{"x":83,"y":64},"formValue":{"x":96,"y":35}},"positionHistory":[{"year":2023,"x":13,"y":20,"note":"Current language models called insufficient for human-level intelligence."},{"year":2024,"x":14,"y":23,"note":"JEPA and world models become the explicit alternative."},{"year":2025,"x":15,"y":24,"note":"Leaves Meta as its strategy consolidates around scaled generative models."},{"year":2026,"x":15,"y":25,"note":"AMI Labs turns the dissenting architecture into a funded program."}],"forecastUtility":{"score":78,"components":{"specificity":17,"falsifiability":14,"proximity":20,"backing":18,"record":9}},"sourceIds":["lecun-vision","lecun-exit"]},{"id":"huang","name":"Jensen Huang","role":"Founder & CEO","organization":"NVIDIA","region":"US","strategicLens":"Capital","currentThesis":"Intelligence becomes an industrial output measured in tokens per megawatt; agents and robots expand the compute market beyond people.","timing":"Commercial inflection is now; no need to wait for a shared AGI milestone.","statedPosition":"Every industry will operate compute-intensive agents and physical systems.","revealedPosition":"Ships a full stack for training, inference, networking, simulation and robotics; designs each generation around lower token cost.","editorialReadThrough":"The infrastructure signal is excellent. The demand forecast carries the cleanest commercial incentive in the set: NVIDIA sells the capacity it predicts everyone will need.","thesisShift":"From training scarcity to inference economics, agent workloads and the physical world.","nextResolvingEvidence":"Utilization and revenue per deployed megawatt after the current capacity wave lands.","convictionSignals":[{"label":"Product","strength":5,"evidence":"The roadmap spans chips, racks, networking, software and simulation."},{"label":"Ecosystem","strength":5,"evidence":"Cloud, industrial and robotics partners are shipping against it."},{"label":"Conflict","strength":1,"evidence":"More industry compute directly expands NVIDIA’s market."}],"positionCoordinates":{"timingRisk":{"x":83,"y":22},"openPower":{"x":58,"y":38},"formValue":{"x":91,"y":94}},"positionHistory":[{"year":2023,"x":73,"y":20,"note":"Accelerated compute and copilots lead the industrial frame."},{"year":2024,"x":77,"y":21,"note":"AI factories replace data centers as the core unit."},{"year":2025,"x":80,"y":22,"note":"Reasoning inference and robotics move to the foreground."},{"year":2026,"x":83,"y":22,"note":"Agent throughput and physical systems define the next platform cycle."}],"forecastUtility":{"score":76,"components":{"specificity":16,"falsifiability":13,"proximity":20,"backing":20,"record":7}},"sourceIds":["nvidia-rubin","nvidia-physical","epoch-trends"]},{"id":"nadella","name":"Satya Nadella","role":"Chairman & CEO","organization":"Microsoft","region":"US","strategicLens":"Distribution","currentThesis":"The winning layer is not one model. It is the enterprise control plane for models, agents, identity, data and distribution.","timing":"Agents are an active software cycle; infrastructure remains constrained through 2026.","statedPosition":"Every organization becomes agent-rich and manages those agents like a digital workforce.","revealedPosition":"Microsoft is committing roughly $190B of 2026 capex while building governance, identity and management around tens of millions of agents.","editorialReadThrough":"The strongest view on value capture: model advantage decays; workflow, control and distribution persist.","thesisShift":"From a copilot beside every user to an operating layer managing many agents per organization.","nextResolvingEvidence":"Paid usage and retained gross margin after depreciation catches up with the buildout.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"A calendar-year infrastructure plan near $190B."},{"label":"Distribution","strength":5,"evidence":"Identity, productivity, developer and cloud surfaces are already owned."},{"label":"Model","strength":3,"evidence":"Increasing in-house work, but the stack remains partner-dependent."}],"positionCoordinates":{"timingRisk":{"x":78,"y":36},"openPower":{"x":53,"y":46},"formValue":{"x":25,"y":93}},"positionHistory":[{"year":2023,"x":66,"y":38,"note":"Copilot becomes the distribution unit."},{"year":2024,"x":70,"y":37,"note":"The platform opens to models, tools and enterprise data."},{"year":2025,"x":74,"y":36,"note":"Agents replace single-assistant framing."},{"year":2026,"x":78,"y":36,"note":"Agent control planes and scarce capacity become the operating model."}],"forecastUtility":{"score":74,"components":{"specificity":15,"falsifiability":13,"proximity":18,"backing":20,"record":8}},"sourceIds":["microsoft-capex","suleyman-humanist"]},{"id":"suleyman","name":"Mustafa Suleyman","role":"CEO","organization":"Microsoft AI","region":"Europe","strategicLens":"Governance","currentThesis":"Very capable systems are coming, but should remain bounded to human-directed applications rather than open-ended autonomy.","timing":"Near-term capability growth; avoids a single universal-AGI event.","statedPosition":"Build superhuman capability in domains while preserving human control, containment and consent.","revealedPosition":"Formed a dedicated superintelligence team inside the company making one of the world’s largest infrastructure bets.","editorialReadThrough":"The tension is deliberate: pursue the capability inside a platform with enough control surface to contain it.","thesisShift":"From warning governments about a coming wave to owning a team expected to deliver it.","nextResolvingEvidence":"Whether bounded-domain language survives pressure to ship a general autonomous product.","convictionSignals":[{"label":"Institution","strength":5,"evidence":"Runs a dedicated model and consumer organization inside Microsoft."},{"label":"Governance","strength":4,"evidence":"Containment has remained consistent across book, policy and product roles."},{"label":"Boundary","strength":2,"evidence":"No public test yet shows what the lab would refuse to build."}],"positionCoordinates":{"timingRisk":{"x":72,"y":79},"openPower":{"x":37,"y":56},"formValue":{"x":32,"y":80}},"positionHistory":[{"year":2023,"x":66,"y":86,"note":"Containment is framed as the central governance problem."},{"year":2024,"x":68,"y":84,"note":"Moves inside Microsoft to build consumer systems."},{"year":2025,"x":72,"y":80,"note":"Creates a bounded ‘humanist superintelligence’ program."},{"year":2026,"x":72,"y":79,"note":"Model and agent programs expand under the humanist frame."}],"forecastUtility":{"score":73,"components":{"specificity":16,"falsifiability":14,"proximity":18,"backing":17,"record":8}},"sourceIds":["suleyman-humanist","microsoft-capex"]},{"id":"altman","name":"Sam Altman","role":"CEO","organization":"OpenAI","region":"US","strategicLens":"Lab","currentThesis":"Capability will keep compounding into superintelligence; the practical task is abundant, cheap distribution through enormous infrastructure.","timing":"Superintelligence is placed by 2035 with meaningful step-changes well before then.","statedPosition":"Concentrated capability should become broadly available, affordable and subject to public oversight.","revealedPosition":"Stargate expands the centralized compute base required to train and serve the systems, while product distribution grows faster than institutional governance.","editorialReadThrough":"He is highly informed and maximally committed, but milestone language moves with the frontier. Treat direction as higher-confidence than dates or definitions.","thesisShift":"From ‘build AGI’ to ‘deploy the infrastructure and institutions for superintelligence.’","nextResolvingEvidence":"A stable, external definition of the milestone and a governance mechanism that constrains deployment under pressure.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"A multi-gigawatt infrastructure program built around the thesis."},{"label":"Distribution","strength":5,"evidence":"Mass-market and enterprise use create direct feedback."},{"label":"Definition","strength":2,"evidence":"The target has moved from AGI toward superintelligence."}],"positionCoordinates":{"timingRisk":{"x":90,"y":58},"openPower":{"x":39,"y":58},"formValue":{"x":22,"y":91}},"positionHistory":[{"year":2023,"x":78,"y":67,"note":"AGI is framed around economically valuable work and safety."},{"year":2024,"x":84,"y":64,"note":"Gradual takeoff and broad access move forward."},{"year":2025,"x":89,"y":60,"note":"Public language moves past AGI toward superintelligence."},{"year":2026,"x":90,"y":58,"note":"Infrastructure and distribution become the public program."}],"forecastUtility":{"score":71,"components":{"specificity":15,"falsifiability":11,"proximity":19,"backing":20,"record":6}},"sourceIds":["openai-ten-years","openai-plan","openai-compute"]},{"id":"zuckerberg","name":"Mark Zuckerberg","role":"Founder & CEO","organization":"Meta","region":"US","strategicLens":"Distribution","currentThesis":"Personal superintelligence should be distributed to billions, with open models limiting institutional concentration.","timing":"Superintelligence is discussed as a next-few-years product cycle.","statedPosition":"Open systems and personal agents are both a growth strategy and a balance-of-power mechanism.","revealedPosition":"The first Muse model launched in private preview, then Meta promised to resume some open releases while retaining capability-dependent safeguards.","editorialReadThrough":"The openness thesis is real but conditional. Distribution and competitive advantage dominate when openness conflicts with frontier lead time.","thesisShift":"From open models as the destination to staged access as the route, followed by a public recommitment to selective openness.","nextResolvingEvidence":"Which frontier weights are actually released, under what license, and how far they trail the private model.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"New labs, data centers and a company-wide product reset."},{"label":"Distribution","strength":5,"evidence":"Billions of users and a hardware edge through glasses."},{"label":"Openness","strength":3,"evidence":"Strong commitment, but frontier access has become conditional."}],"positionCoordinates":{"timingRisk":{"x":88,"y":30},"openPower":{"x":84,"y":88},"formValue":{"x":58,"y":89}},"positionHistory":[{"year":2023,"x":63,"y":28,"note":"Open model ecosystem is the differentiation strategy."},{"year":2024,"x":71,"y":28,"note":"Open source is framed as the likely standard."},{"year":2025,"x":86,"y":30,"note":"Personal superintelligence and a new lab accelerate the target."},{"year":2026,"x":88,"y":30,"note":"Private previews expose limits; the open commitment is renewed in August."}],"forecastUtility":{"score":68,"components":{"specificity":14,"falsifiability":12,"proximity":17,"backing":20,"record":5}},"sourceIds":["meta-open-2024","meta-muse","meta-future"]},{"id":"musk","name":"Elon Musk","role":"Founder","organization":"xAI · Tesla · SpaceX","region":"US","strategicLens":"Physical","currentThesis":"Digital and physical intelligence arrive extremely soon, produce abundance and pose severe control risk; compute speed decides the race.","timing":"Shortest public deadlines in the set, repeatedly renewed.","statedPosition":"The technology is existentially dangerous and should not be monopolized.","revealedPosition":"xAI raised $20B and assembled more than one million H100-equivalent GPUs while Tesla keeps robotics central to its valuation story.","editorialReadThrough":"Capital conviction is exceptional. Calendar calibration is not. Use his build-rate claims as evidence; heavily discount his dates.","thesisShift":"The worldview barely moved. The deadline did: near-term targets roll forward as the milestone stays undefined.","nextResolvingEvidence":"A third-party test that clearly crosses the prior ‘smarter than any human’ claim, rather than a self-declared label.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"One of the fastest and largest compute buildouts in the field."},{"label":"Integration","strength":5,"evidence":"Models, social data, chips, vehicles, robots, energy and launch capacity."},{"label":"Calibration","strength":1,"evidence":"Short deadlines recur after prior windows pass."}],"positionCoordinates":{"timingRisk":{"x":98,"y":48},"openPower":{"x":45,"y":27},"formValue":{"x":86,"y":98}},"positionHistory":[{"year":2023,"x":84,"y":64,"note":"Warnings remain severe while xAI is formed."},{"year":2024,"x":97,"y":56,"note":"Smarter-than-any-human target moves to 2025 or 2026."},{"year":2025,"x":98,"y":51,"note":"Compute, robotics and abundance claims expand."},{"year":2026,"x":98,"y":48,"note":"The any-human threshold is again described as next year."}],"forecastUtility":{"score":54,"components":{"specificity":18,"falsifiability":17,"proximity":14,"backing":20,"record":-15}},"sourceIds":["xai-round","musk-timing-2024","musk-timing-2026"]},{"id":"ng","name":"Andrew Ng","role":"Founder","organization":"DeepLearning.AI / AI Fund","region":"US","strategicLens":"Distribution","currentThesis":"Treat the technology as general-purpose infrastructure: value appears through workflows, domain data, evals and deployment—not one threshold event.","timing":"Fast application iteration; slower institutional and economic diffusion.","statedPosition":"General-purpose technology with many use cases; agents improve through structured workflows and evaluation.","revealedPosition":"Builds education, venture formation and application infrastructure rather than a frontier pretraining lab.","editorialReadThrough":"A useful counterweight to lab-centric forecasts: adoption mechanics are the unit of analysis.","thesisShift":"From supervised learning evangelism toward agentic workflow engineering.","nextResolvingEvidence":"Whether agent teams generate durable ROI outside coding and narrow enterprise processes.","convictionSignals":[{"label":"Career","strength":5,"evidence":"Education and company-building organized around broad diffusion."},{"label":"Capital","strength":4,"evidence":"AI Fund backs application companies rather than one model bet."},{"label":"Measurement","strength":5,"evidence":"Consistently emphasizes evals and production feedback."}],"positionCoordinates":{"timingRisk":{"x":42,"y":24},"openPower":{"x":76,"y":72},"formValue":{"x":28,"y":78}},"positionHistory":[{"year":2023,"x":30,"y":22,"note":"Diffusion, applications and practical deployment dominate the thesis."},{"year":2024,"x":34,"y":22,"note":"Agentic workflows become the practical scaling path."},{"year":2025,"x":39,"y":23,"note":"Evals and workflow engineering matter more than model spectacle."},{"year":2026,"x":42,"y":24,"note":"Parallel agents extend the workflow thesis without requiring a discontinuity."}],"forecastUtility":{"score":84,"components":{"specificity":20,"falsifiability":18,"proximity":18,"backing":19,"record":9}},"sourceIds":["ng-senate","stanford-index"]},{"id":"liang","name":"Liang Wenfeng","role":"Founder & CEO","organization":"DeepSeek","region":"China","strategicLens":"Lab","currentThesis":"Original model research, open weights and organizational density can offset capital and chip constraints.","timing":"AGI could be two, five or ten years away; no single internal roadmap.","statedPosition":"DeepSeek will remain open and prioritize foundational innovation over applications.","revealedPosition":"High-Flyer funds the lab; DeepSeek ships efficient open models and prices inference aggressively.","editorialReadThrough":"The strongest live challenge to the assumption that frontier advantage equals maximum capital expenditure.","thesisShift":"From a quiet Chinese research bet to a global price and architecture setter.","nextResolvingEvidence":"Whether openness survives frontier costs and whether efficiency gains persist across new regimes.","convictionSignals":[{"label":"Ownership","strength":5,"evidence":"Self-funded through High-Flyer rather than strategic cloud capital."},{"label":"Release","strength":5,"evidence":"Repeated open-weight releases expose the work to replication."},{"label":"Organization","strength":5,"evidence":"Bottom-up access to compute and young domestic research talent."}],"positionCoordinates":{"timingRisk":{"x":72,"y":30},"openPower":{"x":93,"y":86},"formValue":{"x":18,"y":38}},"positionHistory":[{"year":2023,"x":57,"y":25,"note":"Curiosity-led AGI research begins with High-Flyer compute."},{"year":2024,"x":64,"y":26,"note":"Open model efficiency becomes the product and geopolitical thesis."},{"year":2025,"x":69,"y":28,"note":"Reasoning results validate algorithmic efficiency over brute-force spend."},{"year":2026,"x":72,"y":30,"note":"Open foundational research remains the stated moat."}],"forecastUtility":{"score":82,"components":{"specificity":20,"falsifiability":19,"proximity":20,"backing":15,"record":8}},"sourceIds":["liang-interview"]},{"id":"wu","name":"Eddie Wu","role":"CEO","organization":"Alibaba Group","region":"China","strategicLens":"Capital","currentThesis":"Open models are the operating system; cloud and commerce are the monetization and distribution layers.","timing":"Immediate infrastructure cycle, multi-year platform capture.","statedPosition":"Keep Qwen open and make Alibaba Cloud the full-stack service layer.","revealedPosition":"$53B three-year AI/cloud program, trillion-parameter models and enterprise agent infrastructure.","editorialReadThrough":"Openness is an acquisition strategy for cloud demand, not a relinquishing of control.","thesisShift":"From e-commerce transformation to an explicit AI + Cloud identity.","nextResolvingEvidence":"Qwen developer share, overseas cloud reach and utilization of the committed infrastructure.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"$53B three-year infrastructure plan."},{"label":"Distribution","strength":5,"evidence":"Cloud, commerce and a large developer base."},{"label":"Open ecosystem","strength":5,"evidence":"Qwen positioned as an operating system."}],"positionCoordinates":{"timingRisk":{"x":75,"y":25},"openPower":{"x":87,"y":67},"formValue":{"x":24,"y":88}},"positionHistory":[{"year":2023,"x":48,"y":22,"note":"AI + Cloud becomes one of Alibaba’s two core businesses."},{"year":2024,"x":57,"y":22,"note":"Qwen expands as an open developer ecosystem."},{"year":2025,"x":70,"y":24,"note":"$53B infrastructure plan makes the strategic priority legible."},{"year":2026,"x":75,"y":25,"note":"Full-stack cloud and open Qwen reinforce each other."}],"forecastUtility":{"score":77,"components":{"specificity":18,"falsifiability":18,"proximity":20,"backing":18,"record":3}},"sourceIds":["alibaba-apsara"]},{"id":"pichai","name":"Sundar Pichai","role":"CEO","organization":"Alphabet","region":"US","strategicLens":"Distribution","currentThesis":"The winning position is the full stack: custom silicon, frontier research, cloud and products with billions of users.","timing":"A platform shift already underway, with transformation mediated through existing products.","statedPosition":"A once-in-a-lifetime general-purpose shift that should be globally accessible.","revealedPosition":"Record capex, TPUs, Gemini, Search integration, Cloud and Waymo create multiple closed feedback loops.","editorialReadThrough":"Alphabet can be directionally right without winning a single-model leaderboard because distribution captures learning and margin.","thesisShift":"From cautious launch posture to rapid, full-stack deployment.","nextResolvingEvidence":"Search economics, TPU externalization and whether agent use expands queries or cannibalizes them.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"$175–185B 2026 capex guidance."},{"label":"Distribution","strength":5,"evidence":"Search, Android, Workspace, Cloud and YouTube."},{"label":"Vertical stack","strength":5,"evidence":"TPUs through applications under one owner."}],"positionCoordinates":{"timingRisk":{"x":68,"y":42},"openPower":{"x":54,"y":55},"formValue":{"x":42,"y":92}},"positionHistory":[{"year":2023,"x":42,"y":40,"note":"The AI-first pivot becomes a product-defense program."},{"year":2024,"x":51,"y":40,"note":"Gemini consolidates research and product execution."},{"year":2025,"x":61,"y":41,"note":"Search distribution and custom silicon become the revealed thesis."},{"year":2026,"x":68,"y":42,"note":"The full stack is operated as one compounding system."}],"forecastUtility":{"score":80,"components":{"specificity":18,"falsifiability":18,"proximity":20,"backing":20,"record":4}},"sourceIds":["pichai-action","alphabet-capex","deepmind-about"]},{"id":"su","name":"Lisa Su","role":"Chair & CEO","organization":"AMD","region":"US","strategicLens":"Capital","currentThesis":"Demand fragments across training, inference, edge and physical systems; open, heterogeneous compute wins meaningful share.","timing":"Multi-year infrastructure build, resolved quarter by quarter.","statedPosition":"The stack needs diverse compute and open software, not one accelerator monopoly.","revealedPosition":"Annual accelerator roadmap, ROCm, systems acquisitions and hyperscaler partnerships.","editorialReadThrough":"A cleaner infrastructure forecast than model-level claims because revenue and design wins resolve it quickly.","thesisShift":"From GPU challenger to rack-scale platform builder.","nextResolvingEvidence":"Software friction, accelerator share and inference economics.","convictionSignals":[{"label":"Roadmap","strength":5,"evidence":"Annual Instinct cadence with public performance targets."},{"label":"M&A","strength":4,"evidence":"Systems capability acquired to compete above the chip."},{"label":"Open software","strength":4,"evidence":"ROCm is central to the competitive thesis."}],"positionCoordinates":{"timingRisk":{"x":70,"y":20},"openPower":{"x":80,"y":64},"formValue":{"x":58,"y":91}},"positionHistory":[{"year":2023,"x":46,"y":18,"note":"Accelerators become AMD’s next platform cycle."},{"year":2024,"x":55,"y":18,"note":"Annual Instinct cadence and systems capability expand."},{"year":2025,"x":66,"y":19,"note":"Open software and rack-scale systems become the competitive wedge."},{"year":2026,"x":70,"y":20,"note":"Inference diversity strengthens the heterogeneous-compute thesis."}],"forecastUtility":{"score":81,"components":{"specificity":20,"falsifiability":20,"proximity":20,"backing":16,"record":5}},"sourceIds":["amd-scale"]},{"id":"murati","name":"Mira Murati","role":"Cofounder & CEO","organization":"Thinking Machines","region":"US","strategicLens":"Lab","currentThesis":"Frontier systems should be customizable, collaborative and shaped by users rather than merely autonomous.","timing":"Long-term frontier build with incremental open products.","statedPosition":"Extend human will and judgment through customizable, understandable systems.","revealedPosition":"Raised frontier-scale capital, secured at least 1GW of NVIDIA systems and released open weights.","editorialReadThrough":"The human-agency message is backed by a conventional high-capital lab structure—but unusually open product choices.","thesisShift":"From operating OpenAI’s product frontier to founding a lab around customization and plural control.","nextResolvingEvidence":"Whether user-shaped systems produce a distinct capability or distribution advantage.","convictionSignals":[{"label":"Career","strength":5,"evidence":"Left a central role to form a competing institution."},{"label":"Compute","strength":5,"evidence":"At least one gigawatt committed with NVIDIA."},{"label":"Release","strength":4,"evidence":"Open-weight model and customization platform shipped."}],"positionCoordinates":{"timingRisk":{"x":58,"y":56},"openPower":{"x":74,"y":76},"formValue":{"x":32,"y":62}},"positionHistory":[{"year":2023,"x":65,"y":60,"note":"Frontier deployment and governance are managed inside OpenAI."},{"year":2024,"x":60,"y":61,"note":"Leaves to build around a distinct institutional thesis."},{"year":2025,"x":57,"y":58,"note":"Customization, open science and human collaboration define the new lab."},{"year":2026,"x":58,"y":56,"note":"Open weights plus gigawatt-scale compute make the thesis testable."}],"forecastUtility":{"score":72,"components":{"specificity":15,"falsifiability":15,"proximity":20,"backing":18,"record":4}},"sourceIds":["tml-partnership"]},{"id":"alex-wang","name":"Alexandr Wang","role":"Chief AI Officer","organization":"Meta","region":"US","strategicLens":"Lab","currentThesis":"Frontier advantage comes from concentrated talent, proprietary data loops, infrastructure and speed.","timing":"Urgent competitive window rather than a stable public date.","statedPosition":"National and corporate leadership require control of data and frontier capability.","revealedPosition":"Meta’s $14.3B Scale investment, a new chief-AI role and a compact superintelligence team.","editorialReadThrough":"High power and costly backing; public forecasting record remains too thin for a high utility score.","thesisShift":"From neutral data supplier to direct frontier-lab operator.","nextResolvingEvidence":"Whether MSL ships durable model leadership and whether Meta’s open posture survives.","convictionSignals":[{"label":"Capital","strength":5,"evidence":"$14.3B strategic transaction enables the role."},{"label":"Authority","strength":5,"evidence":"Direct control of Meta’s superintelligence program."},{"label":"Public record","strength":2,"evidence":"Few dated, falsifiable capability forecasts."}],"positionCoordinates":{"timingRisk":{"x":79,"y":34},"openPower":{"x":42,"y":42},"formValue":{"x":26,"y":91}},"positionHistory":[{"year":2023,"x":50,"y":30,"note":"Data infrastructure and national competition dominate the frame."},{"year":2024,"x":59,"y":31,"note":"Scale expands from labeling to enterprise and government infrastructure."},{"year":2025,"x":75,"y":33,"note":"Meta buys strategic access and gives Wang operating authority."},{"year":2026,"x":79,"y":34,"note":"A smaller, concentrated lab becomes the execution model."}],"forecastUtility":{"score":65,"components":{"specificity":12,"falsifiability":12,"proximity":20,"backing":17,"record":4}},"sourceIds":["meta-scale","meta-future"]},{"id":"bengio","name":"Yoshua Bengio","role":"Founder","organization":"LawZero","region":"Global","strategicLens":"Governance","currentThesis":"Do not grant powerful systems persistent agency; build a non-agentic Scientist AI that predicts without pursuing goals.","timing":"Risk rises with agency before any agreed AGI threshold.","statedPosition":"Safety should take priority over commercial incentives.","revealedPosition":"Redirected his research program, chaired an international report and launched a funded nonprofit lab.","editorialReadThrough":"One of the clearest costly belief revisions in the cohort.","thesisShift":"From capability pioneer to institution-builder for safe-by-design alternatives.","nextResolvingEvidence":"Whether Scientist AI yields useful guarantees and competitive scientific performance.","convictionSignals":[{"label":"Career","strength":5,"evidence":"Research program pivoted in 2023."},{"label":"Institution","strength":5,"evidence":"LawZero formed with dedicated funding."},{"label":"Coordination","strength":5,"evidence":"Led the international safety report."}],"positionCoordinates":{"timingRisk":{"x":72,"y":96},"openPower":{"x":54,"y":70},"formValue":{"x":16,"y":28}},"positionHistory":[{"year":2023,"x":56,"y":83,"note":"Public pivot toward catastrophic-risk governance."},{"year":2024,"x":62,"y":88,"note":"International scientific assessment replaces general warning."},{"year":2025,"x":69,"y":94,"note":"LawZero turns critique into a competing technical program."},{"year":2026,"x":72,"y":96,"note":"Non-agentic Scientist AI remains the concrete alternative."}],"forecastUtility":{"score":86,"components":{"specificity":20,"falsifiability":20,"proximity":18,"backing":19,"record":9}},"sourceIds":["bengio-lawzero"]},{"id":"wang-xingxing","name":"Wang Xingxing","role":"Founder & CEO","organization":"Unitree","region":"China","strategicLens":"Physical","currentThesis":"China can scale capable robot bodies; general-purpose software and real-world learning remain the binding constraints.","timing":"A robotics model inflection within years; broad household use later.","statedPosition":"Language models are not the robotics breakthrough; software and embodied data are the missing layer.","revealedPosition":"Unitree ships low-cost quadrupeds and humanoids at scale while continuing full-stack development.","editorialReadThrough":"The most decision-useful physical-world correction to purely digital forecasts.","thesisShift":"From hardware productization to an explicit embodied-model bottleneck thesis.","nextResolvingEvidence":"Unscripted task completion, field reliability and gross margin—not demonstration videos.","convictionSignals":[{"label":"Product","strength":5,"evidence":"Commercial units ship into research and industry."},{"label":"Manufacturing","strength":5,"evidence":"Cost and iteration advantages are observable."},{"label":"Forecast","strength":3,"evidence":"General-policy deadlines remain unresolved."}],"positionCoordinates":{"timingRisk":{"x":48,"y":18},"openPower":{"x":64,"y":62},"formValue":{"x":96,"y":86}},"positionHistory":[{"year":2023,"x":30,"y":14,"note":"Humanoids enter the product roadmap from a quadruped base."},{"year":2024,"x":39,"y":15,"note":"The missing robotics model—not hardware—is identified as the bottleneck."},{"year":2025,"x":45,"y":17,"note":"Shipments scale while the general-policy claim remains unresolved."},{"year":2026,"x":48,"y":18,"note":"Commercial hardware runs ahead of embodied intelligence."}],"forecastUtility":{"score":76,"components":{"specificity":20,"falsifiability":18,"proximity":20,"backing":14,"record":4}},"sourceIds":["unitree-interview","nvidia-physical"]}],"exactQuoteTimeline":[{"date":"2023.11","year":2023,"lane":"diffusion","player":"ng","quote":"AI is a General Purpose Technology with numerous use cases.","revealed":"Builds education, application companies and deployment practice—not a frontier lab.","source":"ng-senate"},{"date":"2024.04","year":2024,"lane":"physical","player":"wang-xingxing","quote":"For humanoid robots, the biggest question is how to integrate with AI models.","revealed":"Ships robot bodies while treating general-purpose software as unresolved.","source":"unitree-interview"},{"date":"2024.07","year":2024,"lane":"open","player":"liang","quote":"We will not change to closed source.","revealed":"DeepSeek releases efficient open models and uses price as ecosystem pressure.","source":"liang-interview"},{"date":"2024.10","year":2024,"lane":"timing","player":"amodei","quote":"I think it could come as early as 2026, though there are also ways it could take much longer.","revealed":"Threshold governance and a multi-gigawatt compute plan back the short-run case.","source":"amodei-loving-grace"},{"date":"2025.02","year":2025,"lane":"diffusion","player":"pichai","quote":"AI is a once in a lifetime technology.","revealed":"Alphabet integrates Gemini across Search, Cloud, Workspace, Android and custom silicon.","source":"pichai-action"},{"date":"2025.06","year":2025,"lane":"safety","player":"bengio","quote":"I am launching a new non-profit AI safety research organization called LawZero.","revealed":"A research and institution-level pivot toward non-agentic Scientist AI.","source":"bengio-lawzero"},{"date":"2025.07","year":2025,"lane":"open","player":"zuckerberg","quote":"We believe in putting this power in people’s hands to direct it towards what they value.","revealed":"Meta spends for a concentrated superintelligence team while making openness conditional.","source":"meta-personal"},{"date":"2025.09","year":2025,"lane":"open","player":"wu","quote":"We remain committed to open-sourcing Qwen and shaping it into the operating system of the AI era.","revealed":"$53B of infrastructure turns ecosystem language into a cloud-distribution bet.","source":"alibaba-apsara"},{"date":"2025.11","year":2025,"lane":"architecture","player":"sutskever","quote":"We are squarely an ‘age of research’ company.","revealed":"SSI declines near-term products and organizes around reliable generalization.","source":"ilya-interview"},{"date":"2025.11","year":2025,"lane":"safety","player":"suleyman","quote":"We are not building an ill-defined and ethereal superintelligence.","revealed":"Microsoft AI forms a bounded, domain-oriented superintelligence team.","source":"suleyman-humanist"},{"date":"2025.12","year":2025,"lane":"timing","player":"altman","quote":"In ten more years, I believe we are almost certain to build superintelligence.","revealed":"OpenAI’s practical program is infrastructure, distribution and iterative deployment.","source":"openai-ten-years"},{"date":"2026.03","year":2026,"lane":"capital","player":"huang","quote":"The agentic AI inflection point has arrived.","revealed":"Seven production chips and rack-scale systems target every phase of the workload.","source":"nvidia-rubin"},{"date":"2026.03","year":2026,"lane":"capital","player":"murati","quote":"This partnership accelerates our capacity to build AI that people can shape and make their own.","revealed":"Thinking Machines secures at least one gigawatt and later releases open weights.","source":"tml-partnership"},{"date":"2026.08","year":2026,"lane":"open","player":"zuckerberg","quote":"We will resume releasing some open source models soon.","revealed":"The recommitment follows a private-preview phase, showing access policy remains strategic.","source":"meta-future"}],"games":[{"id":"race","title":"The capability race","form":"Repeated prisoner’s dilemma","pressure":"High","players":"Frontier labs · cloud sponsors · states","equilibrium":"Every player invests because unilateral restraint transfers option value to rivals. Public safety language rises with capability, but spending does not slow.","evidence":"Multi-gigawatt commitments are arriving simultaneously across competing stacks.","escape":"Shared evaluations, incident disclosure, verifiable thresholds and liability that make unsafe speed expensive."},{"id":"open","title":"Open versus closed","form":"Public-goods game with asymmetric downside","pressure":"Rising","players":"Meta · Anthropic · open ecosystems · states","equilibrium":"Open lower-risk systems; delay or restrict frontier weights; use openness as distribution and geopolitical leverage.","evidence":"Both Meta and Anthropic now describe capability-dependent release boundaries.","escape":"Portable evaluations and release tiers that separate useful diffusion from irreversible proliferation."},{"id":"stack","title":"Who captures the margin","form":"Vertical foreclosure + complements","pressure":"Very high","players":"Chip vendors · clouds · model labs · application owners","equilibrium":"Each layer commoditizes the layer beside it while bundling its own control point. Distribution and scarce compute retain bargaining power.","evidence":"NVIDIA expands up the stack; clouds fund models; model labs build applications; applications route across models.","escape":"Interoperability, portable state and procurement that prevents any one layer from owning identity, memory and execution together."},{"id":"labor","title":"The adoption race","form":"Coordination game under measurement uncertainty","pressure":"Medium","players":"Firms · workers · software vendors","equilibrium":"Firms deploy before ROI is clean because waiting risks a capability gap. Work changes first at entry-level and decomposable tasks.","evidence":"Broad reported adoption coexists with low agent penetration and uneven labor effects.","escape":"Task-level measurement, redesign budgets and transition mechanisms tied to observed productivity rather than vendor claims."}],"sources":[{"id":"openai-ten-years","title":"Ten years","publisher":"OpenAI","date":"Dec 2025","url":"https://openai.com/index/ten-years/","note":"Altman puts superintelligence inside a ten-year window."},{"id":"openai-plan","title":"Built to benefit everyone: our plan","publisher":"OpenAI","date":"Jun 2026","url":"https://openai.com/index/built-to-benefit-everyone-our-plan/","note":"Distribution, affordability and public oversight become the public frame."},{"id":"openai-compute","title":"Building the compute infrastructure for the Intelligence Age","publisher":"OpenAI","date":"Apr 2026","url":"https://openai.com/index/building-the-compute-infrastructure-for-the-intelligence-age/","note":"Stargate is described as the long-run compute base for broad deployment."},{"id":"amodei-loving-grace","title":"Machines of Loving Grace","publisher":"Dario Amodei","date":"Oct 2024","url":"https://darioamodei.com/essay/machines-of-loving-grace","note":"Defines powerful systems, gives an earliest-arrival case and maps scientific upside."},{"id":"anthropic-rsp","title":"Responsible Scaling Policy v3.4","publisher":"Anthropic","date":"Jul 2026","url":"https://www.anthropic.com/responsible-scaling-policy","note":"Capability thresholds are tied to security, evaluation and reporting requirements."},{"id":"anthropic-compute","title":"Anthropic and Amazon expand compute collaboration","publisher":"Anthropic","date":"Apr 2026","url":"https://www.anthropic.com/news/anthropic-amazon-compute","note":"A ten-year, $100B commitment secures up to 5GW of capacity."},{"id":"anthropic-open","title":"Our position on open-weights models","publisher":"Anthropic","date":"Jul 2026","url":"https://www.anthropic.com/news/position-open-weights-models","note":"Supports lower-risk open weights while drawing a line at dangerous capabilities."},{"id":"deepmind-about","title":"About Google DeepMind","publisher":"Google DeepMind","date":"Aug 2026","url":"https://deepmind.google/about/","note":"The lab states a responsibility-led path to general systems and scientific benefit."},{"id":"alphafold-impact","title":"AlphaFold: five years of impact","publisher":"Google DeepMind","date":"Nov 2025","url":"https://deepmind.google/blog/alphafold-five-years-of-impact/","note":"A concrete record of the science-first thesis, including global research use."},{"id":"alphabet-capex","title":"2025 Q4 earnings call","publisher":"Alphabet Investor Relations","date":"Feb 2026","url":"https://abc.xyz/investor/events/event-details/2026/2025-Q4-Earnings-Call-2026-Dr_C033hS6/default.aspx","note":"Alphabet guides to $175–185B of 2026 capital expenditure and describes model, product and cloud demand."},{"id":"ilya-interview","title":"We’re moving from the age of scaling to the age of research","publisher":"Dwarkesh Patel","date":"Nov 2025","url":"https://www.dwarkesh.com/p/ilya-sutskever-2","note":"Sutskever identifies weak generalization as the bottleneck and changes the scaling thesis."},{"id":"ssi-mission","title":"Safe Superintelligence","publisher":"SSI","date":"Jun 2024","url":"https://ssi.inc/","note":"A single-product company organized around safe superintelligence rather than near-term releases."},{"id":"lecun-vision","title":"A path towards autonomous machine intelligence","publisher":"Yann LeCun","date":"Jun 2022","url":"https://openreview.net/forum?id=BZ5a1r-kVsf","note":"The technical program behind learned world models, memory, planning and hierarchical control."},{"id":"lecun-exit","title":"LeCun leaves Meta to pursue advanced machine intelligence","publisher":"Associated Press","date":"Nov 2025","url":"https://apnews.com/article/313159512bb9961f324e0c93bccf4cf5","note":"A career-level commitment to the non-LLM path after twelve years at Meta."},{"id":"meta-open-2024","title":"Open source AI is the path forward","publisher":"Meta","date":"Jul 2024","url":"https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/","note":"Zuckerberg argues that open models will become the industry standard."},{"id":"meta-muse","title":"Introducing Muse Spark","publisher":"Meta","date":"Apr 2026","url":"https://about.fb.com/news/2026/04/introducing-muse-spark-meta-superintelligence-labs/","note":"The first new superintelligence-lab model launches in private preview, with future openness left conditional."},{"id":"meta-future","title":"The future is for everyone","publisher":"Meta","date":"Aug 2026","url":"https://about.fb.com/news/2026/08/the-future-is-for-everyone/","note":"Meta renews its case for personal, distributed and partly open superintelligence."},{"id":"microsoft-capex","title":"FY2026 Q3 earnings call","publisher":"Microsoft Investor Relations","date":"Apr 2026","url":"https://www.microsoft.com/en-us/investor/events/fy-2026/earnings-fy-2026-q3","note":"Microsoft expects roughly $190B of calendar-2026 capital expenditure and reports large-scale agent management."},{"id":"suleyman-humanist","title":"Towards Humanist Superintelligence","publisher":"Microsoft AI","date":"Nov 2025","url":"https://microsoft.ai/news/towards-humanist-superintelligence/","note":"Suleyman forms a superintelligence team around bounded, human-directed capability."},{"id":"nvidia-rubin","title":"Vera Rubin opens the agentic frontier","publisher":"NVIDIA","date":"Mar 2026","url":"https://nvidianews.nvidia.com/news/nvidia-vera-rubin-platform","note":"Seven production chips span training, post-training and agent inference."},{"id":"nvidia-physical","title":"Global robotics leaders take physical systems to production","publisher":"NVIDIA","date":"Mar 2026","url":"https://nvidianews.nvidia.com/news/nvidia-and-global-robotics-leaders-take-physical-ai-to-the-real-world","note":"A full-stack robotics ecosystem backs the physical-world thesis."},{"id":"xai-round","title":"$20B Series E","publisher":"xAI","date":"Jan 2026","url":"https://x.ai/news/series-e","note":"xAI reports more than one million H100-equivalent GPUs at year-end 2025."},{"id":"musk-timing-2024","title":"Musk predicts systems smarter than the smartest human by 2025 or 2026","publisher":"Reuters","date":"Apr 2024","url":"https://www.reuters.com/technology/teslas-musk-predicts-ai-will-be-smarter-than-smartest-human-next-year-2024-04-08/","note":"A short, testable deadline that can be compared with later statements."},{"id":"musk-timing-2026","title":"Musk says systems may exceed any human as soon as next year","publisher":"Associated Press","date":"May 2026","url":"https://apnews.com/article/4f8810743d6ef9a72f91f8721a3f4027","note":"The near-term deadline is renewed after the earlier window passes without a shared milestone."},{"id":"worldlabs-frontier","title":"A new frontier: spatial intelligence","publisher":"World Labs","date":"Sep 2024","url":"https://www.worldlabs.ai/blog/a-new-frontier","note":"Li’s move from language-centric systems toward persistent, controllable 3D worlds."},{"id":"worldlabs-marble","title":"Generating bigger and better worlds","publisher":"World Labs","date":"Sep 2025","url":"https://www.worldlabs.ai/blog/bigger-better-worlds","note":"Marble turns the spatial thesis into a public, navigable product."},{"id":"normal-tech","title":"AI as normal technology","publisher":"Arvind Narayanan & Sayash Kapoor","date":"Apr 2025","url":"https://www.aisnakeoil.com/p/ai-as-normal-technology","note":"A slow-diffusion counter-thesis centered on institutions rather than autonomous systems."},{"id":"agi-not-milestone","title":"AGI is not a milestone","publisher":"AI Snake Oil","date":"May 2025","url":"https://www.aisnakeoil.com/p/agi-is-not-a-milestone","note":"Argues that capability thresholds do not imply sudden economic or institutional change."},{"id":"stanford-index","title":"2026 AI Index: Economy","publisher":"Stanford HAI","date":"Apr 2026","url":"https://hai.stanford.edu/ai-index/2026-ai-index-report/economy","note":"Adoption is broad, agent deployment remains early, and infrastructure spending is at record levels."},{"id":"metr-horizon","title":"Frontier risk report: Feb–Mar 2026","publisher":"METR","date":"May 2026","url":"https://metr.org/blog/2026-05-19-frontier-risk-report/","note":"Coding agents reach multi-day task horizons on the measured suite, with saturation caveats."},{"id":"epoch-trends","title":"Trends in artificial intelligence","publisher":"Epoch AI","date":"Feb 2026","url":"https://epoch.ai/trends","note":"Tracks compute stock, training compute and data-center build constraints."},{"id":"ng-senate","title":"Statement to the U.S. Senate AI Insight Forum","publisher":"Andrew Ng / U.S. Senate","date":"Nov 2023","url":"https://www.schumer.senate.gov/imo/media/doc/Andrew%20Ng%20-%20Statement.pdf","note":"Frames the technology as general-purpose and argues for application-led diffusion."},{"id":"liang-interview","title":"DeepSeek CEO interview: The Quiet Giant","publisher":"Waves / ChinaTalk translation","date":"Jul 2024","url":"https://www.chinatalk.media/p/deepseek-ceo-interview-with-chinas","note":"Liang Wenfeng on open weights, original research, organizational design and DeepSeek’s AGI objective."},{"id":"alibaba-apsara","title":"Alibaba Cloud’s Apsara Conference 2025","publisher":"Alibaba Group","date":"Sep 2025","url":"https://home.alibabagroup.com/en-US/document-1911884625546838016","note":"Eddie Wu ties Qwen openness to a $53B infrastructure commitment and full-stack distribution."},{"id":"pichai-action","title":"The AI Action Summit: A golden age of innovation","publisher":"Google","date":"Feb 2025","url":"https://blog.google/innovation-and-ai/products/sundar-pichai-ai-action-summit/","note":"Pichai frames the shift as general-purpose, global and distribution-led."},{"id":"amd-scale","title":"AI at Scale Starts Here","publisher":"AMD","date":"Jun 2025","url":"https://www.amd.com/en/solutions/data-center/insights/ai-at-scale-starts-here.html","note":"AMD’s open, heterogeneous-compute thesis, backed by an annual accelerator roadmap."},{"id":"tml-partnership","title":"Thinking Machines and NVIDIA strategic partnership","publisher":"Thinking Machines Lab","date":"Mar 2026","url":"https://thinkingmachines.ai/news/nvidia-partnership/","note":"Murati backs customizable systems with a multi-year, gigawatt-scale compute commitment."},{"id":"meta-scale","title":"Meta invests $14.3B in Scale AI and recruits its CEO","publisher":"Associated Press","date":"Jun 2025","url":"https://apnews.com/article/4b55aabf7ea018e38ffdccb66e37cf26","note":"The revealed bet behind Alexandr Wang’s authority inside Meta Superintelligence Labs."},{"id":"meta-personal","title":"Personal Superintelligence for Everyone","publisher":"Meta","date":"Jul 2025","url":"https://about.fb.com/news/2025/07/personal-superintelligence-for-everyone/","note":"Zuckerberg’s first-party statement of the personal-superintelligence and access thesis."},{"id":"bengio-lawzero","title":"Introducing LawZero","publisher":"Yoshua Bengio","date":"Jun 2025","url":"https://yoshuabengio.org/en/blog/introducing-lawzero","note":"A career and institution-level pivot toward non-agentic, safe-by-design systems."},{"id":"unitree-interview","title":"Unitree CEO on China’s robot revolution","publisher":"TMT / ChinaTalk translation","date":"Apr 2024","url":"https://www.chinatalk.media/p/unitree-ceo-on-chinas-robot-revolution","note":"Wang Xingxing separates robotics hardware progress from the unresolved general-policy bottleneck."},{"id":"li-ted","title":"With spatial intelligence, AI will understand the real world","publisher":"TED","date":"Apr 2024","url":"https://www.ted.com/talks/fei_fei_li_with_spatial_intelligence_ai_will_understand_the_real_world","note":"Li’s public case for spatial intelligence as the bridge from seeing to acting."}]}