Anthropic’s rumored path to a $2 trillion IPO is framed as a potential monster listing built on extreme revenue growth, enormous loss math distorted by a huge non-cash accounting charge, and hundreds of billions in locked-in compute and cloud commitments. The bull case is that AI infrastructure winners, from NVDA and AVGO to datacenter landlords and the next wave of index-fund forced buying, could benefit if the listing comes fast and large enough. The bear case is that Anthropic’s revenue is still far smaller than its obligations, customers can switch, and the stock could reprice hard once earnings, lockups, and real filings replace hype. The core unresolved question is whether Claude’s moat stays strong enough to justify fixed compute bills that are already years ahead of visible cash flow.
Anthropic is expected to go public at up to a $2 trillion valuation, which would be the biggest IPO in history.
Anthropic’s revenue ran from $4.6 billion in 2025 to $4.7 billion in Q1 2026, $11.6 billion in Q2 2026, and $16.8 billion in Q3 2026.
Anthropic reportedly lost $80.7 billion on $16.3 billion of revenue in H1 2026, including a $72 billion non-cash accounting charge tied to financing; stripping that out leaves about an $8.7 billion loss.
Anthropic lost $42 billion on $4.6 billion of revenue in 2025.
Anthropic said Q2 2026 was profitable on an adjusted basis; Q3 2026 was adjusted-profitable with about an 11% adjusted operating margin and positive adjusted cash flow, while adjusted profit excludes stock-based compensation.
Anthropic said it generated about $1.70 of revenue for every $1 spent on compute in Q3 2026, and the S1 is expected to include the first full cash flow statement.
The Wall Street Journal reported about $450 billion of mostly non-cancelable cloud contracts with GOOGL, AMZN, and MSFT, with chip and other deals potentially taking Anthropic’s commitments to as much as $750 billion over the coming years.
At $750 billion of commitments over roughly 10 years, Anthropic would need about $75 billion a year of spend covered; at Q3’s $16.8 billion revenue run rate, about $67 billion a year is needed just to stay ahead of the bills.
About 20% of Anthropic’s revenue comes from just two customers.
AMZN, AAPL, MSFT, GOOGL, META, NFLX, and TSLA are used as IPO analogs: Amazon went public in May 1997 at $18 and did not post a first full-year profit until 2003; Amazon fell about 94% from December 1999 to September 2001; Apple fell 65% from its December 1980 high to June 1982; Microsoft and Alphabet were profitable at IPO; Meta fell 53% from $38 at its May 2012 IPO to $17.73 by Sept. 4, 2012; Meta spent about $13.7 billion in a year on the metaverse; Netflix had about $3.3 billion of negative free cash flow in 2019; Tesla went public in 2010 at $17 and did not post a profitable full year until 2020.
At IPO, Facebook and Amazon were each valued at about 28 times the revenue they did the year before; Anthropic at $2 trillion would be about 30 times its Q3 revenue run rate, or about 22 times at $1.5 trillion.
NASDAQ changed its rules so a brand-new listing big enough can be added to the NASDAQ 100 on its 15th trading day, a rule said to have been done for SpaceX and now relevant to Anthropic; at $1 trillion to $2 trillion, Anthropic could qualify, triggering about $9 billion of forced buying, including by index-tracking retirement accounts.
S&P 500 inclusion is said to require a full year of trading and real standard accounting profits, which is far off for Anthropic.
The IPO pattern cited is first earnings, lockup expiration, and early selling driving post-listing drawdowns before later second chances; examples given include Amazon, Alibaba, Uber, and ARM.
The bull case for the AI stack is that Anthropic’s $450 billion to $750 billion of contracts could benefit GOOGL, AMZN, MSFT, NVDA, and AVGO; for smaller names, a single Anthropic contract can be bigger than the whole business and drive 100%, 200%, or 300% moves.
Data-center and former Bitcoin-miner names are singled out as beneficiaries: Terawulf announced a 20-year lease with Anthropic for a Kentucky campus of about 41 megawatts and about $19 billion of contracted revenue; RIOT signed a 20-year lease for 191 megawatts at Rockdale, Texas, with Bloomberg reporting Anthropic and about $9.1 billion over the first term; HUT announced in December 2025 a partnership with Anthropic and Fluidstack for at least 245 megawatts in Louisiana, with a first lease worth about $7 billion and room to grow to more than 2 gigawatts; IREN reported an unnamed AI-lab customer said to be Anthropic.
CELESTICA, CREDO, and ASTRA are described as riding the same Anthropic-related wave through racks, cables, and connectivity chips used in Google and Amazon AI systems, even though none names Anthropic directly; each is said to be down about 36% to 40% from summer highs.
The stated trading setup is to avoid FOMO, watch the first earnings report and the ~180-day lockup, and expect a cheaper entry within the first year rather than the first week; if the S1 lands in late November, the lockup would fall around May 2027.
The unresolved question is whether Anthropic’s fixed compute bills and sticky cloud contracts are offset by a durable moat, or whether customers can switch next quarter to OpenAI, Google, SpaceX’s AI business, or a cheaper open-source Chinese model.
If Claude stays dominant, the already-committed compute bills could look cheap in hindsight and a $2 trillion valuation could look like a bargain; if models converge and Claude loses pricing power, Anthropic faces direct price competition from Google’s own chips, cloud, and cash.
The missing data problem remains central: there is not enough information yet to make an accurate judgment until more earnings reports and filings arrive.