Anthropic Hits $65 Billion Run Rate and Flips the AI Cash Test

Anthropic’s annualized revenue run rate hit $65 billion by the end of July, the company told investors, up from $47 billion in May and roughly $9 billion at the close of 2025. Preliminary second-quarter revenue topped $11.5 billion and produced the lab’s first positive adjusted operating income.

The figures, first reported by Bloomberg and confirmed by CNBC, arrive as Anthropic prepares a possible public listing this fall that some backers model at $2 trillion or higher. OpenAI’s own run rate sits near $40 billion. The gap is less a scoreboard than a filter: public markets will now demand proof that AI revenue is infrastructure-grade cash flow, not a temporary spike.

From $9 Billion to $65 Billion in Seven Months

Anthropic shared the July run-rate number in a regular investor update. The metric annualizes recent performance rather than reporting audited full-year sales. It still paints a steep curve.

The same arc reads more clearly as a sequence of checkpoints rather than isolated prints.

1. Close of 2025: run rate near $9 billion sets the baseline.
2. Q1 2026: actual revenue of $4.73 billion shows the climb already under way.
3. May 2026: run rate reaches $47 billion as enterprise demand compounds.
4. Q2 2026: preliminary revenue tops $11.5 billion and adjusted operating income turns positive.
5. End of July 2026: run rate hits $65 billion, the figure now circulating with investors.

Q2 revenue rose more than fourteen-fold from about $787 million a year earlier. Sequential growth more than doubled the prior quarter. Documents viewed by Bloomberg also showed positive adjusted operating income for the first time, a milestone that matters more for IPO buyers than any single run-rate print.

Investors modeling the rest of the year still expect the company to finish 2026 between $100 billion and $120 billion in annualized revenue. Separate reporting has pointed to internal 2028 targets near $190 billion to $200 billion. Those remain projections. The gap between the July run rate and those year-end models is the stretch public markets will price first.

Enterprise Seats and Coding Tools Carry the Load

The acceleration is not evenly spread. Enterprise contracts and developer tools have pulled far ahead of pure consumer chat.

• Claude Code and related coding products became a primary growth engine after general availability, with earlier 2026 estimates already placing multi-billion run-rate contribution.
• Business customers spending more than $1 million annually on Claude crossed 1,000 by April, doubling in under two months from the Series G disclosure.
• API and enterprise seats, not the free or Pro consumer tiers, generate the bulk of recognized revenue according to multiple analyst tallies.
• Multi-cloud distribution on AWS Bedrock, Google Vertex AI and Microsoft Azure lets large buyers keep existing procurement and compliance frameworks.

That mix explains why the absolute dollars kept climbing even after temporary model restrictions earlier in the summer. When the Trump administration briefly forced Anthropic to disable advanced Claude variants on national-security grounds, the company restored access within weeks and returned to growth. The demand base proved sticky enough to absorb the pause.

Concentration at the top of the customer list also changes how a slowdown would look. A base built on million-dollar seats and coding workflows tends to renew on contract cycles, not on daily consumer habit. That is why the April count of large accounts, and the speed at which it doubled, still shapes the underwriting case more than any single consumer metric.

Readers tracking how Claude actually behaves inside organizations can see how Claude’s internal reasoning maps across teams in earlier reporting on the company’s own visualization tools.

OpenAI’s $40 Billion Pace Looks Different

Bloomberg reported last week that OpenAI’s annualized revenue had reached more than $40 billion, roughly double its end-2025 pace. Coding tools helped there too. Yet Anthropic’s faster compounding has shifted the narrative.

The two labs may define run rate differently. Anthropic’s figure can reflect a recent high-water month annualized; OpenAI’s may smooth more conservatively. Direct comparison therefore carries a caveat. Even with that discount, Anthropic’s enterprise tilt and earlier profit turn have captivated investors more than OpenAI’s larger consumer footprint.

Both companies have filed confidential IPO paperwork. Market expectation still favors Anthropic listing first, possibly as soon as late September or October. That sequencing alone would force OpenAI to answer the same cash-flow questions in public while its rival already trades.

If Anthropic does go first, the S-1 will set the template. Segment mix, adjusted margins, and cloud concentration will become the default checklist for every lab that follows. OpenAI’s larger consumer base would then be judged against a bar already set by enterprise cash flow rather than by brand reach alone.

The $2 Trillion Bet and the Private Mark It Leaves Behind

In late May Anthropic closed a $65 billion Series H that valued the company at $965 billion post-money, briefly surpassing OpenAI’s last private mark. Backers now discuss a public debut at $2 trillion or even higher, which would eclipse SpaceX’s earlier 2026 listing and set a new IPO record.

Valuation snapshots investors are watching

• $965 billion, May 2026 private round post-money
• $2 trillion+, investor models for autumn IPO discussed with the Financial Times
• ~20x, implied multiple on a $100 billion year-end run rate
• First adjusted profit, Q2 milestone that changes the underwriting conversation

At $2 trillion against $100 billion of run-rate revenue the multiple sits near 20 times sales. Expensive by historical software standards, yet defensible if growth remains in the high double digits and margins expand. The same math looks stretched if the run rate stalls near current levels. That tension is why the July number landed with mixed reactions on X: some investors found the May-to-July step-up less explosive than prior months, while others called any slowdown talk absurd given the absolute scale. Tae Kim, writing as @firstadopter, captured the split: “There are some investors who find this disappointing. But come on people, you can’t extrapolate m/m growth acceleration to infinity. This is still INSANE growth.”

The private-to-public jump also inherits the structural issues already visible at the earlier $965 billion private mark and its catch for secondary buyers and employee liquidity.

A $2 trillion debut would also lock in the reference price for later sellers. Employees and early backers who missed secondary windows at the Series H mark would face a public tape instead of a negotiated round. That shift is mechanical: once shares trade, the $965 billion private print becomes history and the open market sets the clearing price day by day.

Cloud Partners Locked In for the Scale Phase

Revenue at this velocity only works if the compute arrives on time. Anthropic has spent 2026 locking multi-year, multi-gigawatt deals that double as both cost and strategic commitment.

In April the company expanded its Amazon partnership, committing more than $100 billion over the next ten years to AWS technologies and securing up to 5 gigawatts of new capacity, including Trainium chips. Amazon simultaneously invested another $5 billion, with room for more. Dario Amodei, Anthropic’s CEO, said at the time that users now treat Claude as essential and that infrastructure had to keep pace.

Our users tell us Claude is increasingly essential to how they work, and we need to build the infrastructure to keep pace with rapidly growing demand.

Amodei made the comment in the Amazon announcement. The same month Anthropic deepened its Google and Broadcom relationship for multiple gigawatts of next-generation TPU capacity expected online from 2027. Claude remains available across all three major clouds, a distribution edge OpenAI has not fully matched.

Those deals convert the revenue story into a capital-intensity story. Hyperscalers win guaranteed demand. Anthropic wins capacity certainty. Future gross margins will show how much of the $65 billion run rate survives the power and silicon bill.

The multi-cloud layout also reduces single-vendor risk for buyers who already standardized on AWS, Google Cloud, or Azure. Procurement teams can adopt Claude without opening a fourth cloud relationship. That friction cut is part of why enterprise seats, not consumer tiers, dominate recognized revenue in analyst tallies.

The Summer Pause Tested Demand Stickiness

The brief national-security restriction earlier in the summer remains the clearest stress test in the public record. Advanced Claude variants were disabled under pressure from the Trump administration, then restored within weeks. Growth resumed.

For IPO underwriters that episode is more useful than another upbeat run-rate slide. It showed that enterprise and API customers did not evaporate when access wobbled. Contracted seats and coding workflows waited for restoration rather than switching in mass.

Stickiness of that kind supports the higher end of the valuation models now in circulation. A $2 trillion case depends on revenue that survives shocks, not only on revenue that compounds in calm months. The pause did not last long enough to prove permanence, yet it was long enough to matter.

The same episode also underlines why multi-cloud distribution is strategic rather than cosmetic. When one path is constrained, buyers already wired into Bedrock, Vertex AI, or Azure have somewhere to return. That architecture is now part of the cash-flow story public markets will read beside the July print.

Targets Through 2028 Frame the Upside Case

Investor models do not stop at the July run rate. Expectations for the rest of 2026 still cluster between $100 billion and $120 billion in annualized revenue. Internal targets discussed for 2028 reach near $190 billion to $200 billion. Both bands are projections, not guidance locked into a filing.

The path between those markers is where margin math becomes decisive. Hitting the upper 2026 band from a $65 billion July run rate implies continued high-double-digit compounding. Reaching the 2028 band from there implies years of sustained enterprise expansion after the IPO window.

• 2026 exit models: $100 billion to $120 billion annualized revenue
• 2028 internal targets: near $190 billion to $200 billion
• Implied 2026 multiple at $2 trillion: roughly 20 times on a $100 billion run rate
• Supporting milestone already booked: first positive adjusted operating income in Q2

None of those figures is audited full-year revenue. Run-rate math can overstate a soft month or understate a backlog conversion. That is precisely why the first post-listing quarters will carry more weight than the investor updates that got the company to the door.

If growth tracks the higher band and adjusted profits hold, the 20-times framework discussed around a $2 trillion debut stays in the conversation. If the run rate flattens near current levels, the same framework becomes the argument for a lower clearing price. The July number simply fixed the starting point of that debate.

Public Markets Will Price the Cash, Not the Hype

The second-order effect of the July print is already visible. Every AI lab still private must now demonstrate a path to adjusted profitability and diversified cloud supply, or accept a steeper discount. Traditional software vendors watching Claude Code absorb coding and agent workflows face the same question from their own boards: is the AI line item growing fast enough to offset share loss?

Anthropic has not commented publicly on the $65 billion figure. The company declined to discuss financials in some early IPO meetings, according to CNBC. That silence will end once the S-1 becomes public. Quarterly reports after listing will replace run-rate slides with audited numbers and segment detail.

For now the evidence is the acceleration itself, the enterprise concentration, the first profit, and the compute contracts that make the next leg possible. The largest IPO in history, if it arrives this fall, will test whether public investors assign the same multiple to realized infrastructure cash that private ones assigned to the growth curve.

Listing in late September or October would compress that test into a few trading sessions. The Series H mark, the July run rate, the Q2 profit turn, and the Amazon and Google capacity deals would all appear in one document. Buyers would then decide in public whether the cash-flow path justifies the multiple private backers already underwrote.

That test starts with the $65 billion number already in the market.