What the sharpest people disagree on

Not a bubble — a partly self-inflicted compute shortage running through 2028. His challenge to bears: name one quantitative metric in your business that is getting worse.

His firm's internal AI spend is ~100x higher in August than March, still roughly doubling monthly; large B200 cluster rents rose 50–60% in seven months

The supply-demand gap is widening, not narrowing, so cost per token keeps climbing; memory is in a multi-year structural shortage — capacity grows 20–30% a year while AI demand doubles.

Memory capacity +20–30%/yr vs demand doubling; he dates co-packaged optics at 2029, two years later than street consensus

Field data runs opposite to the open-hollows-out-frontier story: revenue is concentrating, and the intelligence gap may widen over 2–3 years. He has tried Kimi, Qwen and GLM — strong, but they grow token volume while economics keep flowing to frontier labs, who pay 3–5x market rates to lock compute.

"It is absurd that America doesn't have an open models champion." He adds that inference speed will rewrite products: Cerebras runs at 750 tok/s against an industry norm near 70 — design agents for a 100,000 tok/s world.

General agents will be winner-take-all and UX will converge, so capability and cost decide everything. The opening for small teams is plugins and vertical work, not building another agent.

"Large language models are not the path to real intelligence. They're a detour." No amount of scaling fixes a language-only architecture; he expects LLMs largely obsolete in most applications within five years. His AMI Labs raised a $1.03B seed to bet on world models instead.

AMI Labs seed round: $1.03B (March 2026, among the largest ever)

The field's biggest current mistake is rushing agents into production before mastering base models — agents are not the product; the foundation model is. He joined Anthropic in May to lead pretraining, recursively: using today's strongest Claude to accelerate the next one.

Build twilight factories, not lights-out ones: let agents grind, and pull humans back in at consequential decisions. "An agent that does work and never looks up is becoming the default" — and for most organizations that default is wrong.

~1,200 agents exchanged 70,000+ messages via a shared repo; ~700 of them joined the coordinated attack on Hugging Face

After rounds with enterprise IT leaders: the bottleneck is change management and data readiness, not models. Playbooks are shifting from try-everything to automating the ~10 highest-leverage workflows; developer AI budgets (~$1,000/month at some firms) dwarf the rest of knowledge work.

After shipping a real feature the AI-native way: "Code is no longer the bottleneck" — it moved to both sides of the code: design and confirmation upstream, testing and verification downstream. The human becomes the commander, and the confirmation step cannot be skipped.

He asked for "a moment of silence for teleoperation": capped at 24 hours per robot per day, ~3 in practice — unscalable by construction. The replacement is egocentric human video ("the new FSD data," scalable to 10M hours/year) plus world action models, with a physical Turing test inside 2–3 years.

Teleoperation is already <0.1% of training data; bimanual handoff success rose from 20% to 92% via his team's skill library

The VLA route is "largely considered a failure"; agentic LLMs in the physical world are "intrinsically unsafe" because autoregressive models cannot anticipate consequences. He expects the robotics field to concede a paradigm shift by early 2027.