Anthropic's next gigawatt is TPUs — and Google may guarantee it

Share
Anthropic's next gigawatt is TPUs — and Google may guarantee it

Three stories about who pays for AI: a lab whose next data centre may arrive with a credit guarantee from its chip supplier, a government renaming the technology in its own internal cables, and an Indian studio platform that just priced at $2 billion.


Anthropic is in early talks to lease up to 1 gigawatt of compute from Stream Data Centers, and the discussions have included the possibility of Google providing a credit guarantee for the project, The Information reported Tuesday. The data centre operator is majority-owned by Apollo Global Management, and Anthropic is expected to fill the sites mainly with tensor processing units designed by Broadcom and Alphabet — though it could also use Nvidia GPUs. The scope of what a guarantee would cover is unclear, and nothing is signed; these are early talks, not a contract.

That is the context behind Wednesday's tape. Nvidia closed down about 1.5% at $225.60, roughly 4% below its May 14 record close, and Barron's framed the read-through plainly: the chipmaker's largest customers now have a credible alternative source of compute. The financing pattern is the part worth watching. We described it when ten banks lent $22 billion to fund Alphabet and Blackstone's TPU cloud, and Anthropic's own commitments were already national-scale — 14.8 gigawatts signed, and maybe $517 billion over a decade. What is new here is the direction of the backstop: a chip designer standing behind its own customer's lease is how you stop a tenant from shopping the other stack. Compute deals have become capital-markets deals, and the balance sheet doing the guaranteeing tells you which supplier intends to keep the account.


The State Department's Bureau of International Organization Affairs told its staff on Tuesday to stop writing "Artificial Intelligence" and start writing "Super Intelligence," per an internal email obtained by the Associated Press. The message, subject line "Change in Nomenclature AI to SI," was signed by deputy assistant secretary Michael Drager and reads: "Following the president's remarks at the GA today, please change all references of 'Artificial Intelligence' to 'Super Intelligence' in our documents and positions. We will be using 'SI' in all remarks, positions and press elements moving forward." It went to one bureau — the office that runs US missions to UN and multilateral agencies — and the Associated Press reported it was not immediately clear whether other bureaus would follow. The department has not confirmed or denied it on the record.

The rename now has a paper trail, which is what makes it more than a podium line. We covered the announcement itself — Trump to the UN: no global AI rules, and AI is 'super intelligence' — and the follow-through is visible: US officials used "SI" at the Security Council's AI session the next day. Researchers told the BBC the label overstates the technology, since superintelligence is a term the field reserves for a much larger leap than anything currently shipping. A government can rename a category; it cannot rename the capability curve underneath it.


Brahma AI, an Indian enterprise content company owned by the visual effects studio behind Dune and The Odyssey, raised $150 million at a $2 billion valuation. Indian private equity firm Multiples Alternate Asset Management led the round, announced Wednesday, and parent Prime Focus said the company drew a further $100 million of investor demand it may accommodate by upsizing. DNEG, Prime Focus's UK subsidiary, will hold 66% of Brahma after the raise.

The pitch is Hollywood-grade production technology sold as enterprise software: an "AI-native" platform for managing, creating and transforming audiovisual assets, spanning content intelligence, visual AI, digital humans and multilingual voice, with a model-agnostic stack and interactive digital humans described as close to launch. Brahma names Warner Bros, the NBA and Mayo Clinic as anchor customers across media, sports, healthcare and advertising. The ownership structure is the tell — a studio spinning its internal tooling out as a platform business is the move every large content company has been quietly evaluating since generative video got good.

What to watch: whether the Stream lease converts into a signed contract, and whether a Google credit guarantee actually appears in the paperwork — that detail is what turns a data centre deal into a capital-markets story.

If a chip designer will guarantee its customer's lease to keep the account, who is really carrying the risk in the AI buildout? Tell us in the comments.

Read more

Byteification turns Qwen3 and Llama 3 into byte-level models

Byteification turns Qwen3 and Llama 3 into byte-level models

The tokenizer is the last piece of the language-model pipeline that was never learned end to end — and a Nature paper out this week shows it can be swapped out for less than one percent of a pretraining budget. Researchers have retrofitted four open subword language models into true byte-level models using less than 1% of a typical pretraining budget, and the result — published 7 October in Nature — closes the performance gap that has kept byte-level models a niche curiosity for nearly a decad

OpenAI's dots go mobile and can start your Codex work

OpenAI's dots go mobile and can start your Codex work

One product update and two benchmark papers this hour, all circling the same question: how much of the work can you actually hand over, and what happens to the part you didn't? OpenAI's dots — the always-on agents that get their own cloud computer — can now be created and configured entirely from the ChatGPT app on iOS and Android, and a dot can start Codex work or pick up an existing Codex thread on your behalf. The October 9 release, billed as the first of a fresh run of updates for dots, mo

Coding agents as a hospital: Cockroach's five-month results

Coding agents as a hospital: Cockroach's five-month results

The agentic-coding debate keeps being framed as speed versus trust. Cockroach Labs just published the most detailed trust-first ledger yet — five months of numbers, including the parts that didn't work. Cockroach Labs ran its codebase like a teaching hospital for five months and says the pipeline merged well over a million lines of code with seven total reverts. In the setup the company calls MOLT Sinai, GitHub issues are patients, merging is discharge, and the humans in charge are Chiefs of M

Today in AI — October 10, 2026

Today in AI — October 10, 2026

A day where the boring machinery took center stage: proof checkers, order books, warehouses and the humble text message — plus fresh arguments about who gets to declare any of it safe. Models & Research * Mathematicians get a reliability primer for the Lean Theorem Prover. A guest post by Thomas Hales on Terence Tao's blog walks through what Lean actually guarantees — and devotes itself to the "Summer of Soundness Bugs," the string of kernel bugs found this summer that let false proofs thr