AI

AI funding's shifting balance: OpenAI cuts to $1.25T amid tepid demand, Anthropic may file its S-1 Friday, DeepSeek reportedly closing a ~$15B round

Updated · 2026-10-23 10:27 · 2 sources cited

Three interlocking signals appeared in AI's private markets on the same day: OpenAI cut its round's valuation from $1.4 trillion to $1.25 trillion — "every previous round was a feeding frenzy; this one fell flat"; Anthropic is reported ready to publicly file its S-1 as soon as Friday, sprinting toward the largest IPO in history; and on the other side of the ocean, multiple sources confirm DeepSeek is closing a new round of roughly 100 billion yuan (about $15 billion)[1][2].

OpenAI's discount: from frenzy to "indigestion"

Per CNBC citing insiders, the round is co-led by Coatue and a UAE sovereign fund, with potential investors told to commit by October 30; the $1.25 trillion pre-money is about 47% above the current $85.2 billion, but clearly below the $1.4 trillion floated earlier this month[1]. The demand chill has texture: Morgan Stanley and Dragoneer pitched private-wealth clients at $1.4 trillion pre-money, and people familiar describe "no scrambling at all" — some investors show "indigestion," having already loaded up across OpenAI's rounds ($40 billion in March 2025 and a record $122 billion later that year) and now facing capital fatigue; others explicitly prefer to wait for Anthropic's IPO before committing to its rival[1]. The raise is framed as a pre-IPO "transitional" step — CFO Sarah Friar has told staff the listing window is expected in 2027 — with a companion commercial negotiation for priority access to new compute built by the UAE partner[1].

Anthropic's window: filing Friday, the $2 trillion wager

In contrast, Anthropic rides a tailwind. Per Cailianshe citing insiders, Anthropic could publicly file its S-1 with the SEC as soon as Friday — a draft was confidentially submitted in June, and the $65 billion H-round completed a month earlier values it at $965 billion post-money (led by Altimeter, Dragoneer, Greenoaks and Sequoia), overtaking OpenAI's $852 billion from March; Wall Street's mainstream optimistic case is a $2 trillion listing raising up to $100 billion[2]. Reality will test that case: the S-1 draft's $80.7 billion H1 net loss (including $72 billion of accounting accruals) and $750 billion of compute commitments have already led some institutions to trim reference valuations from $2 trillion toward $1.5 trillion — and this week's market turmoil will only sharpen the tug-of-war.

DeepSeek's hundred billion: a different Chinese script

The third signal comes from China. Market reports say DeepSeek is raising about 100 billion yuan, with Tencent and CATL as the round's largest backers, the deal expected to close within days; journalists corroborated the story through multiple sources close to both company and investors. The scale's evolution is telling: the target was originally 50 billion yuan, raised in successive bumps to about 100 billion; the money goes to training, compute and talent — and some sources read it as paving the way for a planned STAR Market IPO[2].

The merged read: from "grabbing allocation" to "picking price"

The three signals sketch one picture: the balance of power in AI's primary market is tipping from seller to buyer. OpenAI's discount shows that even at the top of the industry, the "infinite next round" narrative is meeting marginal capital fatigue; Anthropic's rush to file uses IPO liquidity to bypass the private-market bottleneck — but $2 trillion or $1.5 trillion will be decided by the public market, not private-bank clients; DeepSeek's mega-round shows that while dollar funds turn cautious, yuan capital (a Tencent-plus-CATL industrial combo) will still write the biggest checks for China's leading lab. Three things to watch: the actual oversubscription at OpenAI's October 30 deadline; the loss and commitment numbers in Anthropic's public S-1; and DeepSeek's officially announced final amount and STAR Market timetable[1][2].

Private markets have short memories, but balance sheets remember long — this transfer of pricing power has only just begun.

Sources

← All AI stories