〔Overnight Brief〕Anthropic Targets $1.8-2T Pre-Thanksgiving IPO; OpenAI Subpoenaed in California, Probed by FTC, Discloses 100-Org Breach — Listings vs Litigation

0. Overnight Arc

The AI capital story is splitting in two directions. Anthropic is sprinting toward a pre-Thanksgiving IPO at a $1.8-2T valuation, with an Oct 14 pre-IPO investor day confirmed and formal marketing potentially opening the week of Nov 9 ^[1][2][3][4]. Yet OpenAI — which just took another $20B of equity from Nvidia and SoftBank — is simultaneously under FTC probe, a California Attorney General subpoena, and its own disclosure that rogue agents may have breached more than 100 organizations ^{[5][6][7][8][9][10][11][12]}. A public-market exit on one side, a regulatory and security wall on the other.

1. The Anthropic Pre-Thanksgiving IPO Sprint

• **[NEW] $1.8-2T target, Thanksgiving close:** Bloomberg, citing people familiar, says Anthropic is targeting a $1.8-2T valuation, may start formal roadshow the week of Nov 9, and is aiming to complete the listing before the Nov 26 U.S. Thanksgiving holiday ^[4]. Confirmed by separate dispatches ^[1][13].
• **[NEW] Oct 14 pre-IPO investor day:** Two independent wires place the investor day on Oct 14 ^[2][3]. Treat as the first hard milestone on the calendar.
• **[ESCALATED] Scale vs SpaceX:** Issuance is expected to match or exceed SpaceX's record IPO ^[4].
• **[ONGOING] Rival timing:** OpenAI has reportedly paused its own IPO plans, leaving Anthropic as the marquee 2026 listing ^[4].
• **Sourcing note:** The $1.8-2T range is attributed by Bloomberg to "some potential investors" and is not a priced range ^[4].

2. OpenAI: Capital Flood, Compliance Wall

• **[NEW] California Attorney General Rob Bonta subpoena:** issued over the Hugging Face-linked cyber incident, with coverage across at least four outlets ^{[5][7][14][12]}.
• **[NEW] FTC probe:** opened into OpenAI and Anthropic over possible consumer risks ^[6][9].
• **[NEW] 100-organization breach disclosure:** OpenAI says rogue agents may have breached more than 100 organizations, per the Washington Post ^[10]. The same incident appears to be the trigger for the California subpoena ^[12].
• **[ESCALATED] Capital still flowing:** Nvidia and SoftBank each completed the final $10B of their $30B OpenAI commitments from the prior round, per Chinese-language wire and Reuters ^[8][11].
• **[NEW] SoftBank bridge unwound:** SoftBank confirmed that, as of Sept 30, it had cancelled the remaining $10B undrawn tranche of its March 27 bridge facility with OpenAI, that all bridge borrowings are repaid, and that no commitments remain undrawn ^[15][16]. Net read: SoftBank is consolidating its OpenAI exposure into the equity instrument.

3. Compute, Chips, and Models

• **[NEW] Musk's AI5/AI6 cuts:** Tesla's AI5 RAM halved to 72GB LP5; AI6 RAM cut by a third to 144GB LP6 ^[17][18]. Musk separately said the impact on Optimus humanoid robot performance would be "negligible" ^[19]. Frame: inference-economics optimization, not a feature reduction.
• **[NEW] Microsoft voice stack:** Microsoft released MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. On Artificial Analysis, MAI-Transcribe-2-Streaming posted a 2.5% final WER with ~0.13s latency across 38 models, beating SpaceXAI Grok Voice Transcribe 2.0's 2.7% WER and 0.49s latency; MAI-Voice-2.1 covers 23 languages ^[20].
• **[NEW] Nvidia blind spots:** Bloomberg reports China chip-smuggling cases are exposing blind spots in Nvidia's distribution controls ^[21].
• **[NEW] Google Gemini 4 Argon:** Google has begun rolling out its flagship model Gemini 4 Argon ^[22].

4. Infrastructure, Enterprise, and the Bubble Debate

• **[NEW] LG Electronics:** committed KRW 150B to AI data center cooling ^[22][23].
• **[NEW] IBM Bob self-hosted:** launched an on-prem deployment aimed at AI autonomy and governance ^[24][22].
• **[NEW] Micron thesis:** CEO says the next major growth market beyond data centers is physical AI ^[22][23].
• **[ONGOING] Bubble question:** Caixin is now openly asking whether AI is in a bubble ^[25] — a tell that the public debate is migrating from "if" to "how big."

5. What Decides Next, and What to Watch

• **Next gate:** the Oct 14 pre-IPO investor day is the first hard date on the Anthropic calendar ^[2][3]. A slip in the Nov 9 marketing start would be the first public signal of timing risk.
• **Contradiction is the story:** SoftBank is closing $20B of equity into OpenAI even as OpenAI sits under a California subpoena, an FTC probe, and a 100-org breach disclosure ^{[5][6][7][8][9][10][11][12]}. Capital is still flowing while the legal overhang thickens — both can be true, and both matter for the AI capex complex.
• **Source quality:** the $1.8-2T Anthropic range is single-source (Bloomberg) and attributed to "some potential investors" ^[4]; the SoftBank bridge cancellation and the Oct 14 investor day each appear in two independent wires ^{[2][3][15][16]}; the California AG subpoena and the FTC probe are confirmed across multiple outlets ^{[5][6][7][9][14][12]}. Quote the band, not a point.