Anthropic S-1 Lays Bare AI Money Loop.

Mazen Al Sabbagh

— 5 min read
Share

Today's market movements underline a growing discomfort with the capital dynamics powering the AI surge. Anthropic's S-1 filing, intended to clarify its public offering, instead offers a stark ledger of the industry's intricate, often circular, financial dependencies and accelerating regulatory scrutiny. The core question now is not just growth, but sustainable value creation.

TL;DR

  • Anthropic's Compute Loop: An amended S-1/A filing details Anthropic as SpaceX's "Customer B," generating 34.5% of its Q3 revenue by renting GPU capacity, against ~$17B Q3 revenue, a $23.3B net loss, and $895.4B in compute commitments.
  • FTC Intervention: Days before its IPO roadshow, Anthropic received 50-page civil demands from the FTC regarding a consumer-harm probe, a risk the S-1 explicitly states could necessitate "model design changes."
  • Market Correction: Monday's initial relief rally post-Friday's Nasdaq rout proved fleeting, with tech indices sliding by midday to leave the S&P up a modest 0.3%.
  • OpenAI IPO Shelved: OpenAI's latest funding round was scaled back to ~$20B at a $1.25T valuation, with Sam Altman deeming a public offering "ill-advised" for now.
  • Marvell's Infrastructure Play: Marvell secured Lumentum in an all-stock deal valued at $115B, targeting ownership of critical optical interconnect technology for AI infrastructure.

Lead Story: Anthropic S-1 Lays Bare AI Money Loop

An amended S-1 filed Monday provided clarity on a relationship previously relegated to market speculation: Anthropic is "Customer B," responsible for 34.5% of SpaceX's Q3 revenue. This positions the AI lab as SpaceX's largest client, surpassing even the US government.

The dynamic is straightforward: Anthropic leases Nvidia GPU capacity at SpaceX's Memphis "Colossus" data centers. These commitments amount to up to $120.5B through 2029, with an early 2027 opt-out clause requiring 90 days' notice. This illuminates the circular financing often discussed in AI valuation debates, now explicitly outlined in a prospectus: one prominent AI company's revenue is another's lease obligation.

The amendment also revealed the first hard financials: roughly $17B in Q3 revenue — a tenfold increase YoY, but a sharp deceleration to ~47% sequentially. This growth came against a $23.3B net loss and $895.4B in aggregate purchase and lease commitments, "substantially all" tied to compute. A further $5B of Q3 short-term compute was secured at above-market rates, signaling the acute capacity constraints even for a top-tier lab.

A notable disclosure pertained to Anthropic's Responsible Scaling Policy: the filing warned that future models "may imminently" reach its threshold for "dramatically accelerated progress attributable to AI", effectively indicating AI's capability to significantly self-accelerate research. Internally, Claude usage surged over 250-fold this year, with monthly spend on its own models now exceeding payroll since April.

Concurrently, Anthropic allocated up to ¥552.9B ($3.5B) for Japanese investors, marking the largest-ever Japan raise by a foreign entity, underwritten by Mizuho, Goldman Sachs Japan, MUFG Morgan Stanley, Rakuten, and SBI.

The "ANT" filing, however, still lacks a price range. The roadshow is not expected before November 10, with a listing targeted for the week of November 16, and valuation discussions circulating between $1.5T and $2T.

In Other News

FTC Intervenes Pre-IPO. The Federal Trade Commission issued 50-page civil investigative demands to Anthropic executives, ordering testimony in a burgeoning consumer-harm probe. This inquiry, which also targets OpenAI with an earlier subpoena, seeks records on security, safeguards, and marketing veracity. Anthropic's S-1 acknowledges this could incur "significant costs and model design changes," timing critically with its impending IPO roadshow.

Market Reversal on Monday. Following Friday's substantial downturn, where the Nasdaq declined 2.86% and Intel experienced its worst day since 1974 with a 28.6% drop, Monday opened with a tentative relief rally. However, tech stocks soon pared gains by midday, leaving the S&P up a modest 0.3% despite a 7% dip in oil prices. Nvidia maintained roughly +1%. The market's preference for free cash flow over AI-branded capital expenditure remains evident.

AI's Rising Debt Load. Oracle's shares slid about 6% after the company paired a $10B bond sale with a $10B common-stock offering to fund data center expansion; its Q1 free cash flow was negative $5.4B, with interest expense up 55%. CoreWeave priced a roughly $3.55B junk-bond deal, including the first euro-denominated high-yield note from a US AI infrastructure firm. The AI buildout is increasingly reliant on leverage.

Marvell's $115B Infrastructure Bet. Marvell agreed to acquire Lumentum in an all-stock transaction—4.5 Marvell shares per Lumentum share, translating to ~$102.6B in equity value and an enterprise value of $115B. This strategic move aims to integrate custom silicon with optical interconnect technology, targeting a projected $550B AI networking market. The deal, expected to close in H1 2028, ranks among the largest tech acquisitions on record.

X / Social Pulse

Elon Musk offered a rare, direct observation on inference economics, stating Grok Bot's serving costs were "unsustainably high relative to revenue" after Halloween-weekend usage limits hit their ceiling. Meanwhile, Sam Altman confirmed no OpenAI IPO in 2026, deeming it an "ill-advised moment to go public," as the company's latest funding round tightened to ~$20B, with UAE's MGX emerging as the lead investor.

On the technical front, SemiAnalysis introduced its GDPx benchmark, placing Claude Opus 5.5 marginally ahead of Gemini 4 Carbon and positing that safety classifiers might be artificially suppressing scores for both Claude and GPT-6. Yann LeCun maintained his critique of large language models, advocating for JEPA world models as a more efficient path to "understanding" versus brute-force knowledge accumulation. Conversely, Andrej Karpathy highlighted the practical utility of current models with a weekend "HackerTube" project that condensed 43 million Hacker News comments into 6,454 videos in roughly two hours using Opus 5.5.

One to Watch

Earnings & Regulatory Pressure Points. Post-market closes tonight will feature Palantir (consensus at ~$0.41 EPS on ~$2.18B revenue, up ~85%) and Apple, followed by AMD and Arm tomorrow, providing the next litmus test for market appetite regarding AI-driven capital expenditure versus reward. Simultaneously, the extent of the FTC's demands ahead of Anthropic’s roadshow, and Moonshot’s anticipated Nov. 4 release of Kimi K3.1 Pro open weights, warrant close observation.

Quick Hits

  • Meta's Ad Edge: Q3 ad revenue surged 26% for Meta, outperforming Google Search's 12%, attributed to AI-enhanced targeting.
  • Apple's China AI: Apple Intelligence enters developer testing for China on Nov. 4, with a December public launch via iOS 27.3, integrating Alibaba's Qwen for specific features.
  • Token Quota Consolidation: OpenAI announced a unified ChatGPT + Codex token quota commencing in December; Anthropic concurrently rebranded its Max 20x tier to "Ultra" at $200/month.
  • Copyright Scrutiny: The Concord Music v. Anthropic fair-use case remains unresolved, with Judge Eumi Lee finding the lab's arguments "slippery" during Friday's hearing.
  • Brave Sued: The New York Times and three other publishers filed suit against Brave Software for AI scraping, pushing the count of US AI copyright litigation towards 160.

The market's conviction in AI's demand remains firm, yet increasingly both investors and regulators are scrutinizing its true cost basis and the implicit liabilities. The question extends beyond who pays for the unprecedented compute required; it now includes the safety and control of models whose own creators project "dramatically accelerated progress." Anthropic’s S-1, while illuminating the financial architecture, inadvertently shifted the focus from scale to sustainability and responsible development.

Sources

Lock in. M. mazen@thorterminal.com

Read more