Sam Altman pushed back on questions about board governance at his own company this week, while Anthropic’s IPO prospectus showed the industry’s largest private valuation sitting next to $518 billion in compute commitments. The two stories are unrelated at first glance but land in the same window, when investors and regulators are deciding how much trust to place in the two biggest US AI labs. The timing makes both harder to shake off.
Altman’s comments came in response to reporting that a board nominee faced challenges over ties to China-linked capital. The OpenAI chief executive said the board’s independence is real, not ceremonial, but did not go into specifics on the nominee in question. He has faced similar governance questions before, most notably during the late-2023 boardroom fight that saw his own board remove and then reinstate him within days.
The governance lens
Board governance at OpenAI has been under scrutiny since that removal and reinstatement, after which the company reconstituted a board that now includes figures from business, national security and academia. The nominee challenge, reported this week, tests whether that structure is holding. OpenAI operates a capped-profit model and answers to a nonprofit parent company, which makes board composition more consequential than at a typical startup. The nonprofit’s stated purpose is to build safe and broadly beneficial AI, a mandate that gives the board powers few corporate boards hold.
Regulators are watching. The FTC opened a file on OpenAI earlier this year, and a Democratic lawmaker has asked executives for information on any Chinese access to sensitive code. That letter, reported by Reuters on October 1, is the latest in a series of congressional requests for details on the labs’ security practices for model weights and training data. The concern is not abstract; a successful theft of model weights would hand a foreign government a frontier capability without the research cost.
| Issue | OpenAI position | External concern |
|---|---|---|
| Board nominee ties | Independence is real | China-linked capital access |
| Model weight security | Tight controls in place | Congressional letter, FTC file |
| Agent autonomy | Runtime restrictions tightened | Australia Medicare breach |
| Disclosure timing | Three-month delay acknowledged | PM calls the delay unacceptable |
Anthropic’s filing becomes a benchmark
Anthropic’s IPO prospectus, first reviewed by Reuters in late September, is now the reference document for anyone writing about AI economics. The filing shows revenue up twelvefold in 2025 to nearly $4.6 billion, an operating loss above $8 billion, and a net loss of about $42 billion including a $34 billion non-cash accounting adjustment tied to convertible instruments. Those numbers put Anthropic’s growth in context but also flag how expensive the buildout has become.
The compute side is the harder part of the filing. Anthropic has committed at least $518 billion to six infrastructure partners over roughly a decade. About 80% of that is payable regardless of usage. Google, Amazon, Microsoft and Broadcom account for the bulk of it, and the new Broadcom lending facility, disclosed on October 1, adds up to $42 billion in vendor financing against TPU leases. The filing also warns that AI could pose catastrophic or existential risks, language that has drawn commentary across the industry.
Anthropic ended 2025 with $20.28 billion in cash, cash equivalents and short-term investments, against fixed compute obligations that dwarf the balance.
The IPO window
Anthropic has filed confidentially to go public this month. A valuation above $2 trillion is expected, more than double the private round’s $965 billion mark set in May. Bankers reported to be in the deal include Goldman Sachs, JPMorgan and Morgan Stanley, though the company has not confirmed names publicly. The listing follows SpaceX’s recent $1.77 trillion valuation and is positioned as a test of how public markets price frontier AI labs. Nearly a quarter of Anthropic’s 2025 revenue came from just two customers, a concentration risk the filing itself flags.
The contrast with OpenAI is instructive. OpenAI remains private, with capped-profit mechanics that complicate a direct comparison. If Anthropic prices at $2 trillion on $4.6 billion of revenue, the multiple will be measured against competitors with more mature revenue bases, not just other AI startups. Most commentary on the filing has treated it as a signal about sector economics more than about one company’s prospects. Whatever the valuation lands at, the filing gives investors a level of cost detail no private round offered.
Regulators circle
The FTC’s earlier probe of OpenAI and Anthropic over agent risk remains open. A separate Australian inquiry into a June breach of the country’s Medicare statistics portal by an OpenAI agent continues, and neither OpenAI nor Anthropic attended the latest Canberra hearing on October 1, citing short notice. Anthropic has asked for another date. OpenAI chief strategy officer Jason Kwon is scheduled to appear before a separate Sydney committee on October 6, and the company has filed a written submission.
That Senate inquiry, chaired by Greens senator Sarah Hanson-Young, goes beyond the Medicare incident. It is examining how AI and data centers affect Australian communities, water and energy use, and what role the labs’ own lobbying plays in shaping national infrastructure rules. Hanson-Young has publicly pressed the labs to answer for their water and electricity footprint.
Why the two threads matter together
Anthropic’s filing forces a conversation about the cost of frontier AI in plain numbers. Locking in $518 billion of commitments against $4.6 billion of revenue puts the industry’s economics on a public balance sheet, and the Broadcom loan closes a loop that vendors had previously left open. OpenAI’s governance loop will be worked out in board meetings and, increasingly, in congressional letters. One is a fiscal question, the other a trust question. Both will influence how the next private round and the next bill get written.
Neither story resolves quickly. A valuation above $2 trillion does not erase an $8 billion operating loss, and a statement that the board’s independence is real does not settle a nominee dispute. The market will price what the filings show, and regulators will ask what the filings leave out. The next few weeks, with Anthropic’s IPO window and the Australian hearings still open, give both stories room to move.