Analysis
As of October 28, 2026, Anthropic's S-1 puts a number on a question VCs have been asking privately for months: how does a public market price hundreds of billions in compute commitments against a company that lost tens of billions in the first half of the year, tracked in full on our live stats page?
The liability public investors haven't priced yet
Anthropic's S-1 compute disclosures:
- New minimum payment obligations — $417 billion, under compute agreements signed or amended after December 31, 2025.
- Estimated expansion capacity, 2028-2031 — $125 billion to $150 billion.
- Fortune's combined total — $542 billion to $567 billion.
- Anthropic's Series H valuation, May 2026 — $965 billion.
“- Estimated expansion capacity, 2028-2031 — $125 billion to $150 billion.”
Private investors who backed that Series H were underwriting a growth story; public shareholders inheriting the compute liability at IPO will be underwriting a company whose biggest expense line is locked in years before the revenue to cover it fully materializes. That mismatch between when the spending commits and when the revenue to match it actually shows up is the core pricing problem for whoever underwrites Anthropic's roadshow, and it's a mismatch the private round never had to resolve.
Revenue against the liability
- H1 2026 revenue — $16.3 billion.
- H1 2026 net loss — $80.7 billion, mostly a non-cash charge.
- Cash and short-term investments — $131.7 billion as of June 30, 2026.
- Large customers — more than 11,900 generating at least $100,000 in trailing-twelve-month revenue as of September 30, 2026.
That customer count is real enterprise traction, but a base that would need to grow by an order of magnitude to organically cover compute commitments in the hundreds of billions. The revenue is growing fast; the liability is growing faster, and public filings now make both numbers visible side by side for the first time.
The SpaceX precedent
Anthropic is not the first AI-infrastructure-heavy company to test how public markets price this kind of spending. Pulse has tracked how 2026's AI-infrastructure IPOs cooled from their debut-day pops into calmer secondary-market ranges once investors had time to model committed spending against actual revenue, SpaceX's own post-listing trading among them. If that pattern holds, Anthropic's eventual listing price may matter less than where it settles three to six months after its debut, once the compute-liability math gets fully priced in.
What the VC math misses
Private-market valuations like Anthropic's $965 billion Series H were set by investors who could exit through a later round or an IPO before compute liabilities came fully due. Public shareholders don't have that luxury — they hold the stock through whatever quarter the bill actually lands. That's a structurally different risk than the one VCs priced, however bullish they were on Anthropic's growth, and it's a risk no revenue multiple alone captures.
What to watch next
Whether Anthropic's roadshow materials break out compute spending by year, and whether the eventual IPO price embeds any discount for the liability relative to where comparable, profitable software companies trade on revenue multiples alone.