Daily digest · 9 November 2026
No valuation moved today. The pass looked for what the map does not draw rather than what is new, and OpenAI was the biggest gap: its own website names five silicon suppliers and this map carried two. In went the 10 gigawatts of accelerators it designs with Broadcom and the 750 megawatts of Cerebras hardware already serving two of its fastest models. The same trick on TSMC's side, listing every chip company with no foundry recorded, found eighteen and fixed the first.
Nothing moved. Every deal-shaped item in the feed was a company this map re-marked in the past week, and the generic sweep found no closed round. The one fresh figure in circulation, that OpenAI's round has been cut to about $20bn with Gulf funds leading, is one aggregator relaying one trade outlet.
openai -designed_with→ broadcom - OpenAI designs the accelerators and the systems; Broadcom co-develops them, supplies the Ethernet networking and deploys the racks, 10 GW of them starting in the second half of 2026 and completing by the end of 2029. OpenAI, 2025-10-13. verified. The map already drew the identical Broadcom design relationship for Meta's chips and Alphabet's TPUs, so this was the third leg of a set with two legs drawnopenai -builds_on→ cerebras - 750 MW of low-latency compute added to OpenAI's platform in tranches through 2028, and unlike most announced silicon deals this one is already running: OpenAI says GPT-5.3-Codex-Spark runs on Cerebras hardware as the first milestone of the partnership, and its Ultrafast API tier serves GPT-5.6 Sol there too. OpenAI, 2026-01-14. verifiedmeta -rents→ alphabet - Meta agreed to spend more than $10bn with Google Cloud over six years, mainly on AI infrastructure. CNBC, 2025-08-21. reported, because the terms are confidential and the figure comes from two people familiar with the deal rather than either company. Meta's rental legs to CoreWeave and Nebius were drawn; its largest cloud landlord was notwse -fabricated_by→ tsmc - the Cerebras WSE-3, the four-trillion-transistor wafer-scale chip, is built on TSMC's 5nm process. Cerebras, 2024-03-13. verified. An old date for a current fact, so it lands straight on the stale-tie backlog, which is the right place for ittsmc -fabricated_by→ gf UPGRADED from an aggregator citation with no date at all to GlobalFoundries's own release of 2026-10-08: a multi-year $2bn agreement to make the silicon interposers for TSMC's CoWoS advanced packaging at Malta, New York, on a five-year initial term, with volume production ramping in the first half of 2028deepmind -builds_on→ tpu REFRESHED from the 2023 Gemini launch post to the Gemini Omni Flash model card, last updated August 2026, which states the model was trained on Google's Tensor Processing Unitsglean -routes→ anthropic REFRESHED from a 2024 press release to Glean's own release notes of 2026-09-22, which add Claude Opus 5.5 as a premium model in Glean Assistant and AgentsReviewed: TSMC, SpaceX, Meta, Broadcom, OpenAI, Anthropic. NVIDIA came due too and stays due, with 107 edges and little chance of yield per search.
baidu -rivals→ alibaba, dated 2024, was re-checked and left alone: the 2026 coverage putting them head to head is trade press and comparison pages.Further reading: TLDR AI · The Rundown · Techmeme
Validator: 229 entities, 457 relations, 0 errors, 1 warning (63 ongoing ties over a year old). Commit: 4 new edges, 3 refreshed or upgraded, 3 roles extended, no marks moved.
State of AI World maps every significant AI company, what it is worth and how it is connected. Every number carries a source and an as-of date; see the methodology. Data is free to reuse with credit: open data. · All companies · Daily digest · Corrections and contact