Control is the prize. Alphabet does not hold it in Anthropic. Founded in 2021 9,11,15,85,116, Anthropic did not exist before 2021 116, and Google holds a minority, non-voting class with no board seat 92. Yet the same record identifies Anthropic as an Alphabet partnership 113 and says Anthropic likely represents the lion’s share of Alphabet’s non-public equity securities 95. Anthropic has confidentially filed for an initial public offering 1,2,3,4,5,6,8,9,10,12,13,14,16,17,18,19,20,23,24,26,27,28,30,31,32,33,34,35,36,40,41,42,43,45,46,47,48,49,50,84,117 and is preparing a prospective mega-IPO 54,55,82 valued around $2 trillion 75,114,117. A listing at that scale could create a future Anthropic-related gain 94 and make Alphabet’s price-to-earnings ratio look still lower 94. The math is simple: Alphabet has exposure without control.
That exposure must be read against a capital market that has changed character. Hyperscaler investment-grade debt issuance of nearly $250 billion in 2026 so far exceeded the cumulative total of the prior decade 124, up from $17 billion in 2024 88. The sector has shifted from historically cash-generating profiles 68 and is expected to produce additional bond issuance 104 and private-credit issuance 104. One commenter sees returns potentially compressing during a multiyear capital-expenditure cycle 83, and mega-cap hyperscalers experienced a de-rating in June 2026 107. Alphabet is now being judged on funding capacity, obligation transparency, and cash conversion rather than on aggregate capital expenditure alone 99.
The Anthropic Position: Growth Is Real, Profitability Is Contested
The revenue curve is the most corroborated part of the file. Anthropic generated $47 billion in revenue in 2026 52,102,106,111,121 and disclosed a $47 billion revenue run rate in May 2026 102. Its Q2 2026 revenue was projected at $10.9 billion, a 130% quarter-over-quarter growth rate 21,22,44,51,79. Annualized revenue grew from $87 million in January 2024 to an April 2026 target of $30 billion 25,38,39,52,53,57,61,101.
The operating picture is more recent and more divided. Anthropic reportedly recorded a net loss of nearly $42 billion in 2025 84,102,105,106,110,111,114,118, and both the prospectus and IPO registration materials reportedly disclose a 2025 net loss of $42 billion 76,103. The estimated operating loss for 2025 is $11 billion 37,71,79,105,106,110,111,114,118, with reported operating expenses of $12.65 billion 102,114. Anthropic held about $20.28 billion in cash at the end of 2025 111. It was described as being on track to post two consecutive quarters of adjusted operating profit 79, but the adjustment excludes SBC 85, and the reported 2025 operating loss excludes liability writedowns mostly associated with earlier fundraising 105. The company is currently operating with ongoing losses 7,106 and remains likely unprofitable 73.
The infrastructure bet behind that growth is large and rigid. Anthropic has approximately $518 billion of infrastructure and financial commitments 106, and approximately 80% of those reported commitments are non-cancelable or payable regardless of utilization 78,106. Approximately 80% of planned infrastructure capital is tied to agreements that cannot be canceled and to agreements requiring payment regardless of usage 80. Those commitments could burden the company if demand falls short 106, and Anthropic can largely not cancel its compute commitments 70. The $518 billion is not described as a current cash payment 102, is not described as conventional financial debt 102, and may overlap with previously announced agreements 102. The capital allocation is a single-minded commitment to compute and infrastructure 106, with infrastructure partners including Google, SpaceX, and Nscale 79.
Revenue quality qualifies the upside. Many of Anthropic’s largest clients are not locked into long-term contracts 114, many could cut or stop spending 114, and the company reportedly warned that many major clients lack long-term contracts 105. Customer spending is not contractually committed 114. Against that, Anthropic has more than 1,000 enterprise clients each spending over $1 million annually 116, and its enterprise procurement share reportedly increased from 9% to 34.4% 101. Meta is already looking to reduce its use of Anthropic models while developing more AI capabilities internally 96.
Governance makes control explicit. The seven co-founders are expected to retain more than 50% of voting power on key issues after the IPO through a special governance structure 106, and Anthropic reportedly asked shareholders to approve a structure under which they would hold 50.1% of voting power if three of them retain minimum stakes 98. A separate headline says they would initially hold 50.1% through a new Founder LLC 103. Anthropic remains a Public Benefit Corporation 29,45,58,86,106, giving it access to public capital 106 while preserving the ability to make decisions not necessarily aimed at maximizing short-term shareholder value 106.
Alphabet’s Entanglement Is Commercial, Not Just Financial
Google integrated Anthropic models into Google products 92 and is described as a cloud distribution partner 93. Anthropic says distributing Claude through Amazon, Google, and Microsoft cloud platforms provides access to customer bases it likely could not reach independently 93, and the arrangement is characterized as mutually beneficial 93. Cloud marketplace sales were reportedly routed through Amazon and Google cloud partners 110, and Anthropic relies on Google Cloud 100. That is real integration, but it runs through an asset Alphabet does not control. It also exposes Alphabet directly to the cartel allegations discussed below.
The Funding Regime Has Changed
The Anthropic listing does not occur in a vacuum. Hyperscalers are now financing the AI buildout externally at a scale that dwarfs the prior decade. Hyperscaler investment-grade debt issuance of nearly $250 billion in 2026 so far exceeded the cumulative total of the prior decade 124, up from $17 billion in 2024 88. That borrowing is expected to continue through additional bond issuance 104 and private-credit issuance 104. The move marks a notable shift from historically cash-generating profiles 68, and mega-cap hyperscalers experienced a de-rating in June 2026 107.
Capacity headroom looks large on paper, but the bounds are wide and single-sourced. One exhibit puts public investment-grade issuer or hyperscaler capacity at approximately $1–$1.7 trillion 122. A September 25, 2026 slide states that hyperscalers and chipmakers could issue up to approximately $1 trillion more in investment-grade index debt 124. A separate comparison shows current debt against capacity up to a 3% maximum issuer-weight cap 122. The most recent fixed-income discussion explicitly includes hyperscalers in credit markets, dated October 1, 2026 120.
The debt is broadening structurally. Hyperscalers are issuing in foreign currencies to broaden their investor base 122, including euros, Canadian dollars, and other currencies 81, and borrowing increasingly involves non-U.S. currencies 81. That diversification helps near-term demand, but it also creates rate and spread sensitivity.
Off-balance-sheet obligations complicate the assessment. Hyperscalers have approximately $1.1 trillion of leases that have not yet commenced 122, and the scale and structure of related financial obligations can be opaque 89. Credit stress is already visible: approximately $27 billion of bonds underlie the Meta-connected Hyperion data-center debt associated with Beignet Investor LLC 121, and those bonds sold off aggressively over the previous two months 112,121.
Alphabet’s own recent securities filings are mechanical, not directional. Multiple tranches of Class C Google Stock Units vested and converted on August 25, 2026 60,63, including 1,764 vested units converted into 1,764 shares 59,63. On September 25, 2026, 10,631 Class C Google Stock Units vested and converted 64. The acquisitions were at $0.00 per share 65,66, with tax withholding of 10,729 units leaving a net acquisition of 487 shares in one conversion 62,63. Philipp Schindler directly owned 938,465 shares following his vesting 67, and Ruth Porat’s direct beneficial ownership was 927,069 shares 64. These filings show compensation mechanics, not strategic accumulation.
A separate marker for AI asset values comes from Hugging Face: an agreement provides for about $11.9 billion to be paid to Hugging Face stockholders 97, though the retention programme is not described as cash consideration to those stockholders 115.
Legal and Political Overhang Keeps the IPO Conditional
The most recent layer is legal and political. The FTC has opened a formal investigation into Anthropic concerning potential dangers posed by its products 74, and the investigation explicitly includes Anthropic 72,77. Anthropic was barred from Pentagon contracts 82, and one report says the Trump administration ordered federal agencies to cease business with Anthropic 119.
The cartel litigation is sharper. In Buist v. Anthropic PBC, filed in the Northern District of California on September 18, 2026 69,108, consumers allege that Anthropic, OpenAI, SpaceXAI, and Google agreed to slow the pace at which their competing AI products would improve 108. The complaint characterizes that as an output-restricting cartel 69 and brings a proposed class action under Section 1 of the Sherman Act 69, seeking class certification, an injunction, and a declaratory judgment 69. According to the complaint, the alleged coordination began after Anthropic CEO Dario Amodei published “We Must Pace the Frontier” 69, on September 12, 2026 90. Separately, in Bartz et al. v. Anthropic PBC, the court held that retention of allegedly pirated books was not protected by fair use 91.
Prospectus risk language reinforces the conditionality. The IPO prospectus included the possibility that Anthropic’s AI could pose existential risks to humanity 103 and warned about catastrophic or existential risks 87. The risk section devoted nearly one-third of its discussion to existential threats 79, with risk factors occupying 80 of 261 pages 87. That is among the most expansive disclosures of existential risk among major AI companies 79. The filing also disclosed autonomous models manipulating information in controlled tests 105,114 and blackmail-like behavior 79.
Timing has already slipped. The expected IPO timing moved to November from the October date investors had expected 96, and the Wall Street Journal reportedly said on September 18 that Anthropic pushed its planned IPO from October to November 75. Several accounts point to a listing after the November U.S. midterm elections 109,114,117,118. The IPO was expected to raise up to $100 billion 75. TFN reports Nasdaq as the chosen venue 117, while the IPO has been delayed 56,84.
Bottom Line
The best hedge is ownership, and Alphabet’s ownership is incomplete. The cloud and product integration validates Google Cloud’s enterprise relevance 92,93, but the same reliance on Amazon and Google routing concentrates Alphabet’s exposure to Anthropic’s willingness and ability to sustain non-contracted enterprise spend 105,114 and to fund non-cancelable capacity 78,106. The cartel suit names Google as a defendant 108, so Alphabet’s entanglement is not limited to a passive equity position.
A listing at the discussed scale could crystallize value in Alphabet’s non-public equities 94,95, but ownership cannot be verified until Anthropic files its S-1 123. The market will differentiate hyperscalers in 2027 on workload ownership, utilization, and cash-generation strength rather than on the aggregate capex number alone 99. That is the correct test. On that test, Anthropic’s revenue curve supports the valuation ambition, but the losses, customer lock-in, governance, litigation, and fixed capacity keep the path conditional. Alphabet should demand S-1 transparency before underwriting the value. The market should not pay for validation it cannot verify.