Anthropic Revenue, Valuation, and Funding Rounds

Anthropic has turned a capital-intensive model business into something closer to a high-margin software company on a cash basis, while still signing multi-hundred-billion-dollar compute contracts it mostly cannot cancel. Its October 26, 2026 S-1 is the first audited window into a frontier lab. OpenAI remains private, larger on a consumer footprint, and still burning cash against a plan that does not turn free-cash-flow positive until around 2030.

Anthropic’s funding path: from a $124 million seed-stage lab to a $965 billion Series H

Anthropic’s valuation did not compound smoothly. It jumped in steps that tracked the cost of the next Claude training run and the willingness of cloud vendors to fund a customer that would also buy their chips.

Round	Date	Amount	Post-money valuation	Lead / notable investors
Series A	May 2021	$124 million	~$550 million	Jaan Tallinn; Dustin Moskovitz, Eric Schmidt
Series B	Apr 2022	$580 million	~$4.1 billion	Sam Bankman-Fried / FTX-Alameda
Series C	May 2023	$450 million	~$4.1 billion	Spark Capital; Google, Salesforce Ventures
Series D	Early 2024	$750 million	$18.4 billion	Menlo Ventures
Series E	Mar 3, 2025	$3.5 billion	$61.5 billion	Lightspeed
Series F	Sep 2, 2025	$13 billion	$183 billion	ICONIQ; Fidelity, Lightspeed co-leads
Series G	Feb 12, 2026	$30 billion	$380 billion	GIC and Coatue
Series H	May 28, 2026	$65 billion	$965 billion	Altimeter, Dragoneer, Greenoaks, Sequoia

Sources for the lettered rounds include company announcements as compiled by secondary trackers and contemporaneous reporting. [1] [2] [3] [4]

Strategic capital sits beside those rounds and is easy to double-count:

• Amazon has invested at least $13 billion and in April 2026 agreed to invest up to another $20 billion. In return Anthropic committed to spend more than $100 billion on Amazon cloud services over 10 years. [5] [6]
• Google has committed up to $40 billion, including an initial $10 billion announced in April 2026, tied to cloud capacity and Google’s TPU chips. A separate analysis of the S-1 puts Google’s 2026 equity purchases at about $10 billion of Series G-2 and $10 billion of Series H. [5] [7]
• Microsoft and Nvidia announced in November 2025 commitments of up to $5 billion and $10 billion respectively. [3] [8]

Other recurring names on the later rounds include Fidelity, Blackstone, T. Rowe Price, Temasek, QIA, MGX, Founders Fund, General Catalyst, Jane Street, Capital Group, D1, and Baillie Gifford. Spark’s Yasmin Razavi has been described as the only traditional VC on the board. [5] [9]

What the S-1 settles. As of June 30, 2026, Anthropic said it had raised an aggregate of $141.1 billion. Roughly $110 billion of that arrived in the first half of 2026, with the Series H the largest single piece (about $65.2 billion in the filing’s accounting). [10] [11] Aggregator totals that land anywhere from about $68 billion to more than $160 billion are not reliable; they mix equity, contingent vendor commitments, and debt, or stop before the S-1.

Cash, burn, and whether the business already funds itself

The cash number is not a rumor. The S-1 shows $112.4 billion of cash and cash equivalents and $19.3 billion of short-term investments — $131.7 billion — at June 30, 2026. CNBC’s comparison of non-bank U.S. companies put that pile behind only Berkshire Hathaway and Alphabet. [10] [12]

The burn story is the opposite of the headline loss:

• First-half 2026 revenue was $16.29 billion, versus $1.16 billion a year earlier. Full-year figures: $41 million (2023), $386 million (2024), $4.59 billion (2025). [13] [14]
• Operating loss narrowed to $693 million in the first half, from $2.14 billion a year earlier. The June quarter itself was a $974 million GAAP operating profit — the first operating profit the filing reports. [11] [15]
• The $80.7 billion first-half net loss is mostly non-cash: about $72.4 billion is a fair-value mark-up on convertible notes, derivative liabilities, and SAFEs as the private valuation rose. Adjusted operating income was $1.62 billion (about a 10% margin). Adjusted net income was about $2.7 billion. [11] [14] [16]
• Operating cash flow was +$3.93 billion in the first half. Capex was only about $482 million because Anthropic rents most compute. The Information reported $3.5 billion of free cash flow on that $16.3 billion of revenue, reversing a $1.6 billion cash burn a year earlier. [17] [18] [19]

SemiAnalysis and Yahoo Finance both read the filing as roughly $9–10 billion of cumulative cash consumed out of $141.1 billion raised — less than $10 billion of lifetime burn to a business that later told investors it did $16.8 billion in a single quarter. [11] [15]

Breakeven status, with the caveats. On an operating-cash and adjusted-profit basis, Anthropic has already crossed. Management told investors in mid-October that the third quarter was cash-flow positive, adjusted-profitable for a second straight quarter, with an 11% adjusted operating margin and $1.70 of revenue per $1 of compute. Third-quarter revenue was $16.8 billion, up about 45% from roughly $11.6 billion in the second quarter and about $4.7 billion in the first. [20] [21] [22] GAAP net income is still deeply negative because of those fair-value charges; the notes are expected to convert around the IPO, subject to an ownership cap on Amazon. The prospectus also says cash, sales, and the credit facility should cover at least the next 12 months — and then warns the company may need more financing sooner than it expects, precisely because the compute contracts are fixed and revenue is not. [12]

The balance-sheet “convertible notes payable” of $108.6 billion is a fair-value liability, not cash principal. Analysts reading the filing put actual outstanding Amazon note principal near $6.2 billion. [12] [19]

How the business actually makes money

Anthropic is an enterprise consumption business that happens to have a consumer chatbot, not the reverse.

In the first half of 2026, consumption-based (token) fees were $13.1 billion, about 80% of revenue; subscriptions were $3.2 billion, about 20%. [23] [24] More than 11,900 customers spending at least $100,000 a year generated about $12 billion, or 74% of first-half revenue. That cohort includes every Fortune 10 company and more than half the Fortune 500. More than 1,900 customers spend over $1 million a year and about 250 spend over $10 million. Net revenue retention, excluding Claude Pro and Max, was about 900% as of September 30, 2026. [11] [23]

Distribution is the structural quirk. Sales through Amazon, Google, and Microsoft cloud marketplaces were 48% of first-half revenue (about $7.8 billion), up from 11% in 2023, 32% in 2024, and 47% in 2025. Distribution fees to those resellers were about 17% of first-half revenue in one reading of the filing. [25] [26] [27] The same three companies are investors, compute suppliers, sales channels, and, in Google and Microsoft’s case, model competitors.

Gross margin needs a definition. Cost of revenue implies an 85% gross margin in the first half, up from 12% a year earlier. [23] [28] That is not the full compute bill. “Compute and infrastructure,” largely below gross profit in the operating statement, was $8.95 billion in the first half (55% of revenue) and fell from 251% of revenue in Q1 2025 to 48% in Q2 2026. Revenue per dollar of that compute line went from $0.40 to $2.08 over the same span. [29] [30] Training and new capacity still hit in discrete steps, so the ratio can reverse when a large block of contracted machines comes online.

Run-rate figures that circulated before the S-1 — over $47 billion in May 2026, over $65 billion by late July — are company or source-cited annualizations, not GAAP. Growth then slowed: Q3’s 45% sequential gain was a sharp deceleration from Q2’s more-than-doubling, and Sacra reported annualized revenue only around $70 billion by October. [2] [31] [20]

The compute overhang that the cash pile does not retire

Positive cash flow does not retire the contracts. Analysts summing the seven dollar figures in the S-1 get about $637 billion:

• Broadcom equipment leases (TPU capacity): about $161.2 billion of rent, largely non-cancellable
• SpaceXAI (Nvidia GPUs): up to $120.5 billion, about $3.1 billion a month once fully delivered; cancellable on 90 days’ notice after early 2027
• Google Cloud: at least about $114.5 billion, April 2026–July 2033, payable regardless of usage
• Amazon: at least $110 billion, May 2026–April 2036, payable regardless of usage
• Microsoft: at least $44.9 billion
• Nscale and Lambda: up to roughly $44.6 billion and $41.3 billion

[24] [32] [30]

A September draft prospectus had put commitments at at least $518 billion, about 80% non-cancellable — the gap is mostly later amendments (Microsoft rose from about $31 billion) plus the SpaceXAI and Broadcom lines being fully specified. [33] [34] Additional TPU capacity under discussion was estimated at $125–150 billion, and a full AMD buy would exceed $20 billion; those sit outside the $637 billion sum. [24] The filing’s own risk factor states the mismatch: revenue is mostly consumption-based and not locked in, while compute commitments run for years.

For someone entering this market, that is the competitive fact. Anthropic’s near-term P&L can look like software. Its decade-long cost base looks like a utility with take-or-pay contracts written by its own shareholders.

Side-by-side with OpenAI

OpenAI’s last primary mark is its own: on March 31, 2026 it closed $122 billion of committed capital at an $852 billion post-money valuation. The round was anchored by Amazon (up to $50 billion, of which about $35 billion is contingent on an IPO or an AGI milestone), Nvidia ($30 billion), and SoftBank ($30 billion), with Microsoft continuing and a broader book including a16z, D. E. Shaw, MGX, TPG, and T. Rowe Price. About $3 billion came from individual investors through bank channels. The credit facility was expanded to about $4.7 billion. [35] [36] [37] Prior steps: a $6.6 billion round at $157 billion in October 2024, a $40 billion SoftBank-led round at $300 billion in March 2025, and a secondary around $500 billion in October 2025. [38] Lifetime totals in commercial databases cluster around $176–187 billion, but they disagree on whether contingent tranches count; Statista had $176.4 billion as of February 27, 2026, before the round fully closed. [39] [40] Treat “total raised” for OpenAI as less clean than Anthropic’s S-1 figure.

Metric	Anthropic	OpenAI
Latest primary valuation	$965 billion (May 28, 2026 Series H)	$852 billion post-money (March 31, 2026)
Capital raised	$141.1 billion aggregate through June 30, 2026 (S-1)	$122 billion committed in the March 2026 round alone; lifetime estimates ~$176–187 billion, partly contingent
Cash and equivalents	$131.7 billion cash + short-term investments at June 30, 2026	About $40 billion at end-2025; more than $73 billion cash and securities at end of Q1 2026, after the round began funding
Booked revenue	2025: $4.59 billion. H1 2026: $16.3 billion. Q3 2026: $16.8 billion (investor meeting, not yet in the S-1)	2025: $13.07 billion. Q1 2026: $5.7 billion. Q2 2026: $6.7 billion. H1 2026 about $12.4 billion
Run rate	Over $47 billion (May); over $65 billion (late July). October annualized figures slowed toward ~$70 billion in some estimates	About $50 billion at end of September 2026, confirmed by the company. A ~$70 billion figure counted gross partner revenue and was walked back
Gross margin	~85% in H1 2026 on cost of revenue; compute-and-infrastructure still 55% of H1 revenue and 48% in Q2	About 33% in 2025; one secondary compilation put Q1 2026 near 39%. Not from an S-1
Operating result	H1 2026 operating loss $693 million; Q2 operating profit $974 million; adjusted operating income $1.62 billion	2025 operating loss $20.92 billion. Q1 2026 cash burn $3.7 billion. Q2 operating loss including stock compensation reported at $12.3 billion
Cash flow	H1 operating cash flow +$3.9 billion; H1 free cash flow about +$3.5 billion. Q3 described as cash-flow positive	Internal July 2026 plan: $278 billion of negative free cash flow from 2026 through 2030. March raise expected to be exhausted by 2028, with about $116 billion still to raise
Compute commitments	~$637 billion itemized in the S-1 across seven counterparties; cloud minimums with Amazon, Google, and Microsoft are largely payable regardless of use	Supplier filings: incremental $250 billion to Microsoft Azure, $138 billion to Amazon (a $38 billion deal expanded by $100 billion), up to about $18–22 billion to CoreWeave. Internal plan: about $856 billion of compute and infrastructure through 2030. Altman has also cited on the order of $1.4 trillion of data-center commitments
Business mix	~80% token consumption, ~20% subscriptions; 48% of revenue via cloud marketplaces; enterprise-heavy, ~74% of H1 from $100k+ accounts	Entered 2026 about 60/40 consumer/enterprise. By August 2026, CFO Sarah Friar told investors enterprise had overtaken consumer, two quarters ahead of the year-end parity target. ChatGPT scale (900 million-plus weekly actives; later ~1 billion monthly actives) is the consumer moat Anthropic does not have

Anthropic column: S-1 and the October investor meeting. [10] [11] [14] [23] [20] OpenAI column: company announcement, FT-verified 2025 financials, The Information, FT, and CNBC. [41] [42] [43] [44] [45] [35] [46] [47]

Do not compare the revenue lines as if they use the same ruler. Anthropic books the gross value of sales through cloud partners and expenses the partner’s cut. OpenAI books only its own share of certain partner arrangements. OpenAI itself said reports of a $70 billion run rate were wrong and confirmed about $50 billion. On a gross basis the gap to Anthropic would be smaller; on OpenAI’s net basis, Anthropic’s quarterly booked revenue ($16.8 billion in Q3) is already in the same neighborhood as a $50 billion annualized rate. [42] [48] Third-quarter run-rate growth at OpenAI was still fast — about 77% for the company and 107% for enterprise, per what it told investors. [47]

Capital efficiency is the real split. Anthropic reached a first-half free-cash-flow margin that one cloud-comps benchmark put near 21% (11% on a trailing-twelve-month basis) after consuming on the order of $10 billion of cash in its entire history. [11] OpenAI’s audited 2025 result was an operating loss larger than revenue ($20.9 billion on $13.1 billion), R&D alone of $19.2 billion, and a plan that still has expenses outrunning cash until the end of the decade even after cutting the five-year burn forecast from about $305 billion to $278 billion. [41] [43] Its Q1 2026 cash burn of $3.7 billion was more than half of that quarter’s $5.7 billion of revenue. [45]

The mechanism is mix plus cost timing. Anthropic’s revenue is concentrated in high-intensity enterprise and coding workloads (Claude Code was cited at an $8 billion annualized rate as of May, and agents multiplied billable tokens). That produces high net revenue retention and a compute bill that, for a few quarters, grew much more slowly than sales. OpenAI still carries the cost of a free and lightly monetized consumer base at ChatGPT scale, plus heavier training spend, even after enterprise crossed 50% of revenue. Gross margin in the low-to-mid 30s cannot fund an $856 billion compute plan. Anthropic’s 85% gross margin can fund operations today and still cannot, by itself, fund $110 billion minimums to Amazon and Google that are owed whether customers show up.

Both labs are financed in a circle. Amazon, Google, Microsoft, and Nvidia are simultaneously investors, suppliers, and in several cases distributors. Of the capital OpenAI has raised since the start of 2025, a large share came from companies it buys compute from — Amazon and Nvidia alone account for $80 billion of commitments in that window, and Amazon had paid $28.7 billion of its $50 billion by the end of June. [33] Anthropic’s Series H explicitly folded in previously committed hyperscaler money. Equity in these rounds is partly a prepayment for future cloud invoices.

What this means if you are competing, investing, or selling into either company

Anthropic is the one with public books, a November listing targeted under the ticker ANT (reports have pointed to the week of November 16, a raise that could exceed $100 billion, and a valuation discussed in a $1.5–2 trillion range — down from $2 trillion talk after Q3 growth slowed). Goldman Sachs and Morgan Stanley lead the syndicate. The last primary price is $965 billion; a secondary indication the day before the filing’s market reaction was already lower. [1] [49] [16] OpenAI has talked about a listing, then pushed it; by late summer Friar was describing an IPO as a 2027 milestone, and October reports had it discussing another private round near $30 billion at a valuation above $1 trillion. Those later OpenAI round figures are press reports, not a closed deal. [50] [40]

Three implications follow from the numbers rather than from the narrative:

1. Near-term cash is not the constraint at either lab. Contracted compute is. Anthropic can fund operations from sales and still has to meet usage-independent minimums that dwarf 2025 revenue. OpenAI’s own plan says the March round runs out in 2028. A competitor that cannot sign take-or-pay deals with Amazon, Google, or Microsoft is not in the same supply market, regardless of model quality.

2. Enterprise API economics are currently more efficient than consumer scale. Anthropic got to operating profit without ChatGPT’s distribution. OpenAI’s enterprise crossover is the attempt to import that efficiency onto a consumer cost base. If coding and agent workloads keep concentrating on Claude, Anthropic’s 900% net retention is a harder moat than weekly active users. If those workloads commoditize — the S-1 risk factors and Sacra’s October note both flag customers cutting Claude Code usage and cheaper open models — the fixed compute bill lands on a shrinking numerator.

3. Reported revenue and reported losses are both easy to misuse. Anthropic’s $80.7 billion loss is mostly a mark-to-market on its own rising valuation. OpenAI’s $50 billion versus $70 billion run-rate dispute is an accounting-policy dispute. Any comparison that ignores gross-versus-net marketplace revenue, or that treats fair-value charges as cash burn, will get the capital-efficiency ranking backward.

---

Recent Findings Supplement (October 2026)

Anthropic closed its largest-ever private round in May 2026, raising $65 billion at a $965 billion post-money valuation—surpassing OpenAI’s prior $852 billion mark—and filed its S-1 on October 26, 2026, for a Nasdaq IPO (“ANT”) potentially valuing it at $1.8–2 trillion.[1][2]

This round incorporated $15 billion of prior hyperscaler commitments (including $5 billion from Amazon) alongside new capital from Altimeter Capital, Dragoneer, Greenoaks, Sequoia, Coatue, GIC, Capital Group, and strategic chip partners (Samsung, SK Hynix, Micron). Run-rate revenue had reached $47 billion earlier in May and climbed to approximately $65 billion by July (gross basis, including partner sales).[3][4]

• Total capital raised reached $141.1 billion by June 30, 2026 (roughly $110–111 billion in 2026 alone), with earlier rounds including $30 billion Series G in February at $380 billion post-money.[5]
• Google committed up to $40 billion in April 2026 ($10 billion immediate at a $350 billion valuation, plus $30 billion milestones). Amazon’s cumulative commitments now stand at up to $33 billion potential.[6]
• The IPO filing positions Anthropic for a potential $100 billion+ raise (existing investors also selling shares), with expectations of a blockbuster debut testing broader investor appetite beyond Silicon Valley.[5]

This capital influx and IPO path give Anthropic a war chest and liquidity event ahead of OpenAI, pressuring rivals to match scale while highlighting how hyperscaler partnerships convert into both funding and locked-in compute demand.

Anthropic’s S-1 reveals explosive H1 2026 revenue growth to $16.3 billion (14x YoY from $1.16 billion), narrowing operating losses, positive operating cash flow, and a massive $131.7 billion cash and short-term investments balance as of June 30.[2][7]

H1 compute/infrastructure spend was $8.95 billion against that revenue, supporting ~85% gross margins on existing models and an 85%+ net revenue retention rate. Q2 operating income turned positive at $974 million (8.4% GAAP margin, 19% adjusted), with full H1 operating loss at just $693 million. Operating cash flow reached +$3.9 billion for the half (free cash flow +$3.5 billion reported), reversing prior burns.[8][9]

• 2025 full-year revenue was $4.59 billion; H1 2026 alone exceeded all of 2025. Q2 revenue hit ~$11.6 billion (up sharply QoQ).[7]
• Cash pile ($112.4 billion equivalents + $19.3 billion short-term investments) exceeds most major U.S. non-bank companies and will grow further with the IPO.[10]
• Net loss of $80.7 billion was driven almost entirely by $72.4 billion non-cash fair-value adjustments on convertible notes; underlying operations show rapid improvement.[2]
• Over $518 billion in compute commitments (largely non-cancellable) over ~10 years, plus hundreds of billions in future lease/equipment obligations, underscore the scale of infrastructure bets.[11]

Anthropic has transitioned from heavy burn to cash-flow positive operations faster than expected, using its data and enterprise focus to achieve high margins on deployed capacity—creating a capital-efficiency moat that forces competitors to either match pricing power or absorb deeper losses.

OpenAI’s annualized revenue run rate reached approximately $50 billion (net basis) by end-September 2026—up from ~$40 billion in August and ~$20 billion at end-2025—while the company explores a $30+ billion pre-IPO round at a $1.4 trillion pre-money valuation and delays its own IPO to 2027+.[12][13]

Enterprise revenue has overtaken consumer, with business customers exceeding two million; the company projects $70+ billion run rate by year-end 2026. Its March 2026 $122 billion round (at $852 billion post-money) brought cash in hand to roughly $162 billion, but internal plans project a $278 billion cumulative cash burn through 2030, with current resources potentially exhausted by 2028 and ~$116 billion more needed.[14][15]

• Revenue accounting differs from Anthropic’s gross method (OpenAI reports only its share of partner sales, e.g., via Microsoft).[16]
• Compute and infrastructure commitments remain enormous (~$856 billion planned through 2030 in one projection).[15]
• Recent growth accelerated post-GPT-5.6 (July), with enterprise doubling in Q3, though some investor estimates of higher run rates were revised downward.[17]

OpenAI maintains stronger consumer scale and partnership leverage but faces steeper projected cash needs and slower reported profitability trajectory, making its next funding round a critical test of whether enterprise momentum can close the efficiency gap with Anthropic.

Side-by-side (most recent public or reported figures, post-April 2026):

• Revenue/Run-rate: Anthropic ~$65 billion (July gross); OpenAI ~$50 billion (Sep net, targeting $70+ billion YE). Anthropic showed faster recent scaling in H1 S-1 data.[3][12]
• Margins/Cash Flow: Anthropic H1 2026 operating margin improving to positive in Q2, +$3.9 billion operating cash flow, high gross margins on deployed models; OpenAI still projecting heavy multi-year burns ($278 billion through 2030).[8][15]
• Cash on Hand: Anthropic $131.7 billion (June 30); OpenAI ~$162 billion (post-March round).[2][15]
• Capital Raised/Valuation: Anthropic $141.1 billion total (~$65 billion Series H at $965 billion May); OpenAI $122 billion March round at $852 billion, seeking $30+ billion at $1.4 trillion pre-money.[1][13]
• Commitments: Both face hundreds of billions in compute obligations; Anthropic’s are more explicitly quantified in S-1 with high non-cancellable portions.[11]

Anthropic leads on near-term capital efficiency and cash generation while OpenAI retains advantages in absolute consumer reach and certain partnership structures—differences that will be tested in their respective IPO timelines and ability to fund non-cancellable infrastructure without eroding margins.

Business mix and capital efficiency diverge sharply: Anthropic’s enterprise/API focus (over 1,000 customers >$1M/year, strong in regulated sectors via safety/interpretability positioning) pairs with higher reported gross margins and quicker path to positive cash flow on existing capacity.[3] OpenAI balances a massive consumer base (ChatGPT) with fast-growing enterprise but reports net revenue and faces deeper projected losses tied to aggressive scaling.[14]

Anthropic’s model treats safety research and enterprise governance as core differentiators that unlock sticky, high-value contracts, enabling it to underwrite compute spend more efficiently. OpenAI’s consumer flywheel drives volume but requires heavier discounting or subsidies in some enterprise deals, pressuring margins.[18]

New entrants or incumbents must choose: replicate Anthropic’s enterprise/safety moat for better unit economics or OpenAI’s consumer scale for volume—while both paths demand unprecedented capital discipline amid non-cancellable compute bets that dwarf current revenues.

For competitors, Anthropic’s S-1 and positive cash-flow inflection set a new benchmark for capital efficiency in frontier AI, while OpenAI’s larger consumer base and upcoming raise highlight the ongoing premium on distribution and partnerships; the race now hinges on who sustains revenue growth without exhausting cash reserves or triggering valuation resets ahead of public markets.