2026-11-01 - Apple picks Qwen for China, and nobody closes anything

A Sunday after a brutal week for AI stocks, and the ledger barely moved. The one real deal is a distribution deal: Apple will ship its first AI features to Chinese iPhones in December running on Alibaba's Qwen, handing the most-used open-weight model family in the world a slot on a fifth of the global smartphone market. Everything else that looked like news was unfinished: OpenAI's round is open and now less than half its original size, Marvell has not signed for Lumentum, and NVIDIA and AMD are still not confirmed as buyers of the Meta stock they promised to buy. No valuation on this map changed today.

Today in AI

• Apple will bring its first AI features to iPhones in mainland China with the iOS 27.3 update in December, running on Alibaba's Qwen through Siri, after a developer beta from November 4 - South China Morning Post link
• OpenAI cut the round it launched at $30B and a $1.4T valuation to about $20B at $1.25T, has roughly $11B committed, and is now approaching family offices for the rest - Financial Times and the Wall Street Journal link
• The three largest independent AI cloud companies, CoreWeave, Nebius and Nscale, have been effectively locked out of the bond market as lenders demand tougher terms, which is how the build-out gets paid for - Wall Street Journal link
• S&P cut its outlook on Meta from stable to negative over the size of its capital spending and its negative free cash flow, after the stock's worst month in four years - CNBC link
• OpenAI, Google, Microsoft and Anthropic all said they will support one new standard that lets publishers opt out of being scraped for AI training across many crawlers at once - The Verge link
• Opposition to new AI data centres in drought-stricken Arizona is now bipartisan in a state that welcomed TSMC's fabs a few years ago, the clearest sign yet that water and power are the binding political constraint - The Verge link
• Baseten, which runs other companies' AI models for them, is seeking about $2B in two tranches priced at $26B and then $30B, double the $13B it was worth in June - Bloomberg link

Marks moved

New edges

• None. The day's one new relationship, Apple running Alibaba's Qwen in China, cannot be drawn: Apple has no node here, and the deployment is not live until December anyway, the rule that also holds Oracle's announced TPU and Cerebras legs. It is in Alibaba's role, and a human call.

Still in talks

• Baseten: ~$2B in two tranches at $26B and $30B, Sands Capital expected to lead. Not final, not announced by the company, so the mark stays at the June Series F.
• OpenAI: about $11B of a target now cut to $20B at $1.25T pre. Mark stays at the $852B March round.
• Marvell / Lumentum: now reported near $83B, all stock, no significant premium, an agreement possible in early November. No filing from either side; Lumentum is not on the map.
• Meta / NVIDIA / AMD: the $25B common leg priced at $625 on five times the demand, and the $3B and $1B are announced by the suppliers themselves, but no closing or settled purchase is confirmed. Both invests_in edges stay undrawn.
• Anthropic: no amended filing, no price range, and the date is contested (Fortune says Thanksgiving week, the WSJ the week of November 16).
• Agility Robotics: the de-SPAC shareholder vote is tomorrow, November 2.

Leads I could not verify

• Alibaba's new role fact rests on the South China Morning Post quoting Apple's statement of 2026-10-31; Apple's own newsroom page never surfaced, so a human should read the primary.
• Iluvatar CoreX, the Shanghai GPU maker listed in Hong Kong as HKEX 9903, is the one member of the Chinese accelerator set this map lacks. Held off on the edge gate: the ByteDance order is still talks from a June Reuters exclusive, no foundry is named, and its four public market caps run from $7.5B to $12B with no dated quote page among them.
• A KuCoin flash saying Baseten "secures $2 billion" and Tenstorrent raised $1.5B is wrong twice: Baseten is seeking, not closed, and the Tenstorrent round closed 2026-10-19 and is already on the map at $6.2B.
• OpenAI's memory arrangement with Samsung and SK Hynix, up to 900,000 DRAM wafers a month, is still a letter of intent a year on, with 2026 volumes reported as minimal. No memory_from edge, so the map's largest node still has no memory supplier.
• The power-delivery layer behind NVIDIA's 800V racks (Texas Instruments, Infineon, STMicro, onsemi, Navitas, Vertiv, Delta) was swept and stays off: 800V is a 2027 deployment, and there is no relationship type for a power component supplier.

Further reading: TLDR AI · The Rundown · Techmeme

Validator: 220 entities, 418 relations, 0 errors, 1 warning (biren and turing orphaned, standing backlog). Commit: no marks and no edges; Alibaba's role rewritten around the Apple deal, eight watchlist items touched, three of them new, one lesson.