Anthropic IPO Valuation: What the S-1 Numbers Imply

Chart illustrating the Anthropic IPO valuation scenarios after the public listing

Anthropic sold shares on November 18 at $83, above the $70 to $78 range it had marketed, and brought in $81.3 billion. No IPO has raised more. The Anthropic IPO valuation will now be tested against the company’s own filing, and one detailed cash-flow model suggests the bankers priced it close to fair.

A day-one pop that already faded

Trading began at $111.50, 34% over the offer and good for about $2 trillion in market value. By afternoon the stock had fallen back into the mid-$90s.

Only about 5% of the company was sold, and demand for the first public frontier AI lab was heavy. A pop on a thin slice of stock says more about supply and demand than about what the business is worth.

Anthropic IPO valuation image 1
Image: alexoppenheimer.substack.com

The 85% margin that is really 28%

Subtract the cost of serving paid inference from revenue and the first nine months of 2026 show an 85.5% gross margin, which looks like software. The analyst behind the model disagrees. Anthropic dropped gross profit from its income statement in 2025, and several costs that rise with revenue are booked in other lines:

  • Training compute is booked as research and development.
  • Cloud marketplace fees from AWS, Google Cloud and Azure sales are counted under sales and marketing.
  • Free-tier and internal compute never reach the cost of serving paying customers.

Add back the channel and free-tier costs and the contribution margin is about 70%. Include training compute and the adjusted margin falls to 28%. Anthropic follows GAAP, so none of this is improper. It just changes what the number means.

Anthropic IPO valuation image 2
Image: alexoppenheimer.substack.com

Cash flow tells a harder story

Nine-month operating cash flow was a positive $6.6 billion, flattered by interest income and by earlier compute prepayments being consumed. Chip prepayments and capital spending sit in investing, and data center lease principal sits in financing.

Once every compute cost, channel fee, cash salary, stock award, tax payment, capex item, chip prepayment and working capital change is counted, free cash flow for the period is negative $4.1 billion. That is about negative 12% of revenue. The third quarter alone was roughly negative 24%.

Anthropic IPO valuation image 3
Image: alexoppenheimer.substack.com

Two variables decide the outcome

All-in compute costs about 57% of revenue today. For a mature free cash flow margin, that has to fall to roughly 35%. The filing says Opus 5.5 cut the cost per task by about 40%, but Anthropic has also signed $895 billion of compute commitments, about $70 billion of it due within twelve months. Those payments are largely fixed. If revenue disappoints, the bill doesn’t shrink.

Prices are the other side. The filing says price cuts reduced third-quarter revenue by roughly $600 million. Compute has to get cheaper faster than prices fall.

What the model assumes

In the base case revenue climbs from about $55 billion this year to $182 billion in 2028 and roughly $577 billion in 2036, with 2027 growth at 100% and fading afterward. Long-term growth is 3.5%. Free cash flow margin starts near 5% in 2027 and climbs to 27% over eight years. The discount rate starts at 11.8% and drifts to 9.5%.

Roughly 70% of enterprise value sits in cash flows beyond 2036, which puts heavy weight on a margin the company has never reached.

Where the Anthropic IPO valuation lands

The base case gives about $86 a share, or $1.57 trillion. Twelve scenarios then move the main inputs together. Weighted by the analyst’s own subjective probabilities, they give about $70 a share, or $1.27 trillion, roughly 16% below the offer price. About 55% of the probability weight sits under $83.

  • Around $86: growth fades on schedule and compute costs come down.
  • Around $105: a compute efficiency breakthrough or a soft landing with lower rates.
  • Around $120: the company’s own plan delivered.
  • $167: Anthropic becomes the dominant AI platform.
  • Below $61: rates stay high, cost cuts stall or demand slips.

Upside cases need several things to go right at once. Downside cases need one thing to break, because the commitments are fixed.

Rates matter as much as operations

Next year’s growth rate is the biggest lever, worth a $1.4 trillion swing. The equity risk premium moves value by about $0.6 trillion and the risk-free rate by about $0.5 trillion. Shift both rates 100 basis points together and the value runs from $1.1 trillion to $2.3 trillion with no change in the forecasts. Capital structure barely matters, since Anthropic carries almost no debt.

Capacity may matter more than demand

The model’s author agrees with a view that supply is the bigger risk. Data centers became a campaign issue this month, Democrats won the House and the Senate, and New York has passed a state moratorium. Maryland’s governor says he would sign one. Because Anthropic’s S-1 flags this risk and most of its capacity is in the US, local permits, power hookups and zoning are the nearer threat than a national ban.

Part of the capacity comes from neoclouds such as Fluidstack and CoreWeave, which borrow against GPUs and contracts. CoreWeave’s bonds have traded around 15% yields. Anthropic buys compute 12 to 18 months ahead, so 2027 is mostly covered, but a squeeze would hit 2028, when the base case needs revenue to rise from about $110 billion to $182 billion.

What to watch after the pop

Earnings and supply now set the price. Watch whether 2027 revenue growth stays close to a doubling, whether compute falls from 57% of revenue toward 35%, whether price cuts eat the efficiency gains, and whether new data centers open on time. Lock-up releases will also come in stages, and each adds shares to the market. A December slowdown in annualized revenue could signal either supply limits or weak demand, and the difference shows in rate limits and pricing.

Source: alexoppenheimer.substack.com

Written by

Saksham Bhargava

Saksham Bhargava runs Alteroid and has been writing about consumer technology for over five years, covering iPhone, Android, Windows, AI tools and gaming. With an MCA and six years of hands-on web development experience, he tests the features he writes about on his own devices and focuses on clear, practical guides that solve real problems. Away from the screen, he enjoys travelling to places most tourists skip.