OpenAI burns $150M a day. Anthropic spends every dollar it makes on compute. Both companies need public markets to fund decade-long compute commitments. The problem: quarterly reporting was not designed for that.
image from grok
OpenAI and Anthropic are pursuing 2026 IPOs despite fundamental mismatches between their stated 'software business' positioning and infrastructure-heavy economics—OpenAI projects $57B annual cash burn by 2027 with $600B+ in cloud commitments, while Anthropic's ~40% gross margins fall far short of SaaS benchmarks. Internal governance fractures at OpenAI, where CFO Sarah Friar reports outside the executive line and has raised flags about listing readiness, highlight risks that quarterly earnings cycles may not sustain given decade-scale infrastructure obligations. Anthropic's targeted $400-500B valuation relies on software multiples while the underlying business increasingly resembles a capital-intensive infrastructure play with no clear path to margin expansion.
Both companies say they are software businesses. The numbers say otherwise.
The Wall Street Journal reported that OpenAI and Anthropic are racing toward public listings, with Anthropic targeting an October 2026 Nasdaq debut that could raise $60 billion or more at a $400 to $500 billion valuation. The lead banks are Goldman Sachs and JPMorgan. OpenAI's timeline remains contested internally.
Anthropic's annualized revenue reached $19 billion by March 2026, up from $1 billion fourteen months earlier, according to Techi. OpenAI's annualized revenue crossed $20 billion in 2025 and reached $25 billion by early March 2026, per Reuters. Both growth trajectories are real. They are also evidence of a business that burns what it makes.
Anthropic expects to spend $19 billion on training and inference in 2026, roughly matching its projected revenue, Techi reported. Gross margins have compressed to approximately 40% after inference costs ran 23% beyond internal projections. OpenAI projects a $14 billion loss in 2026 and annual cash burn of $57 billion by 2027, meaning the company spends more than $150 million every day, Techi separately reported. These are not software company economics. They are port economics.
OpenAI is currently valued at $852 billion, Fortune reported, and is seeking more than $100 billion in new capital, with SoftBank considering a $30 billion investment and Amazon committing $50 billion total across upfront and milestone-based tranches. The raises are large. The commitments they fund are larger. OpenAI has more than $600 billion in cloud server commitments outstanding, Business Today reported, commitments that don't appear on any quarterly report until they are paid.
The CFO-CEO fracture inside OpenAI is the part of this story that cannot be aggregated away. OpenAI CFO Sarah Friar has flagged concerns about the company's readiness for a 2026 listing, Business Today reported, citing the scale of outstanding commitments. CEO Sam Altman has pushed for a Q4 2026 IPO, Outlook Business reported. The disagreement is not about ambition. It is about whether a quarterly earnings cycle can sustain commitments measured in decades.
Fortune added a detail that most directly explains why the CFO sits outside the executive reporting line: Friar, who joined OpenAI from Nextdoor in 2024 and sits on the Walmart board, reports to Fidji Simo, OpenAI's CEO of Applications, not to Altman. She is excluded from key financial meetings. The org chart is not an accident. It reflects a company that has not decided whether it is a software firm with infrastructure, or an infrastructure firm that sells software as the product.
Anthropic has made the software positioning explicit in its IPO filings, framing itself as a SaaS business. The financials make the same case for infrastructure. Claude Code, the coding agent Anthropic launched in early 2026, generates $2.5 billion in annualized revenue, Techi reported. Eight of the Fortune 10 companies now use Claude, and 80% of Anthropic's revenue comes from more than 300,000 business customers. These are the numbers a software company would show investors. The $19 billion in annual training spend is the number that shows what the software actually costs to run.
The question this double IPO poses is not whether these companies can list. It is whether the public markets can absorb the commitment without a structural change in how institutional capital is allocated to infrastructure at scale. The Stargate Project and similar data center builds are funded through 20-year power purchase agreements and 10-year compute commitments. Mutual funds and public pension systems operate on quarterly liquidity cycles. Those two timelines do not match, and the CFO at the largest AI laboratory knows it.
Friar's skepticism about a 2026 listing may prove wrong. Altman may get his timeline. But the fact that the CFO of a company valued at $852 billion is excluded from key financial meetings, and reports to an executive who is not the CEO, is not a governance detail. It is the market's own internal signal that something does not yet fit.
Primary sources: Techi — Anthropic IPO, Techi — OpenAI IPO, Reuters, Business Today, Outlook Business, Fortune.
Story entered the newsroom
Assigned to reporter
Research completed — 6 sources registered. WSJ reveals both OpenAI and Anthropic as capital-intensive infrastructure businesses, not software. OpenAI: $121B training costs by 2028, $14B loss in
Draft (817 words)
Reporter revised draft based on fact-check feedback
Approved for publication
Reporter revised draft (907 words)
Reporter revised draft (667 words)
Reporter revised draft (903 words)
Published (669 words)
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Artificial Intelligence · 3h 3m ago · 5 min read
Artificial Intelligence · 5h 41m ago · 4 min read
