OpenAI Revenue and Run Rate: 2026 Financial Analysis

OpenAI’s late-September 2026 annualized revenue was about $50 billion on the company’s own books—not the ~$70 billion figure that circulated a week earlier. The gap is an accounting construction, not a $20 billion sales miss. The $70 billion number investors and outlets used in late September was built to match Anthropic’s gross treatment of cloud-partner sales. Separately, OpenAI has told investors it expects its own (net) annualized revenue to reach or exceed $70 billion by year-end 2026. That target is a forecast of exit pace, not recognized revenue for the calendar year, which internal plans still put near $36 billion.

The $50 billion vs. $70 billion reconciliation

On October 8, 2026, the Financial Times reported that OpenAI had told investors its annualized revenue was “approaching $50 billion” at the end of September. Bloomberg, CNBC, and Reuters confirmed the same figure. Nine days earlier, Axios had reported a run rate nearing $70 billion (some outlets cited about $68 billion). The Financial Times and a person familiar with the matter, as relayed by CNBC and Reuters, said that higher number came from OpenAI’s own investors trying to produce a like-for-like comparison with Anthropic—not from a company disclosure of $70 billion already achieved. [1] [2] [3] [4]

The mechanism is partner-channel accounting. Anthropic books the full customer payment on sales through cloud partners (AWS, Google Cloud) as revenue and records the partner’s cut as an expense. OpenAI, for certain partner sales—especially models sold through Microsoft Azure—recognizes only its own share. Reporting on the partnership, including CNBC citing a person familiar with the April 2026 renegotiation, puts that share at 20 percent, continuing through 2030 but now capped at a cumulative $38 billion. On a simplified $1 Azure sale, OpenAI may book roughly 20 cents while an Anthropic-style gross method books the full dollar. Secondary accounts of the investor math describe exactly this adjustment: take OpenAI’s net base and gross up the partner channel so the headline can be compared with Anthropic’s. That is how a ~$50 billion net figure became a ~$68–70 billion “comparable” figure. Neither company has published audited quarterly statements, and annualized revenue has no GAAP definition, so both approaches can be defended—and neither is cash in the bank. [5] [6] [7] [8]

A further wrinkle: after the April 2026 commercial update, OpenAI stopped receiving the revenue share Microsoft had been paying OpenAI on Azure resales of OpenAI models. Microsoft still collects its share of OpenAI’s own revenue. Mid-August reports of a run rate above $40 billion still included some of that inbound Microsoft revenue, which then dropped out of OpenAI’s recognized top line. The late-September investor “gross-up” and OpenAI’s own net figure are therefore not the same series. [9]

On October 9, Bloomberg reported a different $70 billion: OpenAI expects to reach or exceed $70 billion in annualized revenue by the end of 2026, on the basis it uses with investors, driven mainly by enterprise. That is a three-month forecast, not a restatement of September. Moving from roughly $50 billion to $70 billion is about a 40 percent increase in the run rate in one quarter. The same investor materials showed third-quarter run-rate growth of 77 percent overall and 107 percent in enterprise. [2] [10] [11]

What this means for anyone underwriting the company or its suppliers: quote the metric and the basis in the same sentence. A $1.4 trillion pre-money valuation discussion (Bloomberg, September–October 2026 fundraising talks for $30 billion or more) is about 28 times the September net run rate and about 20 times the investor gross-up. The multiple is not the story; the definition is. Suppliers are paid from the smaller number—the cash OpenAI actually keeps—not from a grossed-up comparison built for Anthropic.

Run rate is not recognized revenue

Annualized revenue multiplies a recent month (or short period) by 12. It measures pace. Recognized revenue is what the books record over a full period. Because growth is so steep, exit run rate has run far ahead of the year’s booked total.

• End of 2025: CFO Sarah Friar said annualized revenue had surpassed $20 billion, up from $6 billion in 2024 and $2 billion in 2023, as compute capacity rose from 0.6 GW to 1.9 GW. Audited full-year 2025 revenue was $13.07 billion. By December 2025 the company was generating nearly $2 billion a month—an exit pace near $24 billion against a $13 billion year. [12] [13] [14]
• First half of 2026: investors were told revenue was $5.7 billion in Q1 and $6.7 billion in Q2 (up 18 percent quarter on quarter), or about $12.4 billion booked in six months. Anthropic’s Q2 was reported at $11.5–11.6 billion, the first quarter it led on that metric. Those quarterly figures are recognized revenue, not run rates, and they are not adjusted onto a common partner-accounting basis. [15] [3]
• Full-year 2026 plan: the July presentation reviewed by the Financial Times projects about $36 billion of revenue for 2026, rising to $350 billion in 2030, with cumulative revenue of about $840 billion through 2030. Secondary summaries have also cited a Bloomberg figure near $35 billion for recognized 2026 revenue. Either way, a $50 billion September run rate and a $70 billion December target do not imply $50–70 billion of GAAP revenue in calendar 2026. [16] [17] [18]

Consumer versus enterprise is a mix shift, not a separate top line that explains the $20 billion gap. Axios reported that third-quarter consumer revenue alone exceeded all consumer revenue in 2025, while enterprise more than doubled since July. The growth rates OpenAI showed investors—77 percent total, 107 percent enterprise—are run-rate growth, not a split of the $50 billion. One mid-2026 account put consumer near 70 percent of revenue and enterprise above 40 percent; those shares can reflect different definitions and should not be added. The year-end $70 billion target is described as enterprise-led. [19] [20]

Cash burn and projected losses

The operational picture in 2025 is an operating loss, not the headline net loss. Audited figures obtained by Ed Zitron and reviewed by the Financial Times show 2025 revenue of $13.07 billion (from $3.7 billion in 2024), R&D of $19.18 billion, total costs of about $34 billion, and an operating loss of $20.92 billion (from $8.78 billion in 2024). Net loss attributable to the company was about $38.5 billion, swollen by a roughly $41.55 billion non-cash charge tied to the nonprofit-to-for-profit conversion. A person familiar with the matter told the FT that stripping that charge and other non-cash items left losses near $8 billion. Payments to Microsoft in 2025 totaled about $17.2 billion across R&D, cost of revenue, and other items. Year-end assets were just over $50 billion, roughly half in cash. [13] [21] [22]

Internal plans have been revised upward on both revenue and burn, then partially walked back on cumulative free cash flow:

• February 2026 (The Information): cash burn of $25 billion in 2026, $57 billion in 2027, $85 billion in 2028, $51 billion in 2029, and positive $39 billion in 2030. That vintage added about $111 billion of cumulative burn versus the prior forecast and assumed about $665 billion of training and inference cost through 2030. 2025 cash burn was about $8 billion. [23] [24]
• May 2026: negative free cash flow of about $305 billion over 2026–2030. [16]
• July 2026 presentation (FT, September 18): negative free cash flow of $278 billion from 2026 through 2030—better than May, still enormous—against $856 billion of computing and infrastructure spend and the revenue path from $36 billion to $350 billion. The $122 billion raised in March 2026, at an $852 billion valuation, is projected to be exhausted in 2028. [16] [17]

Quarterly losses were still widening into mid-2026. Investor disclosures reported by outlets covering the Wall Street Journal put the Q2 operating loss, including stock-based compensation, at $12.3 billion, up from $9.3 billion in Q1, while revenue rose only $1 billion. Losses grew faster than sales in that quarter. [25] [26]

The implication is a funding treadmill, not a near-term profit inflection. Even the improved July plan does not turn free cash flow positive before 2030, and the cash raised in March is modeled to run out two years earlier. A new round of $30 billion or more, if it closes, covers a fraction of the $278 billion five-year gap. Microsoft’s fiscal 2026 disclosure—$24.1 billion of revenue from commercial arrangements with OpenAI, including revenue share and Azure, and $6.0 billion of receivables at June 30—shows the cash is already flowing to the compute supplier, not sitting as OpenAI margin. [27] [28]

Compute commitments versus the revenue that has to pay them

Commitments are multi-year contractual or reported obligations. They are not annual P&L expense, they overlap, and several of the largest are press figures rather than line items OpenAI has confirmed in its own filings.

Disclosed or company-stated commitments that suppliers have put in their own documents or announcements:

• Microsoft: an incremental $250 billion of Azure services under the restructured agreement (Microsoft subsequent-event disclosure). Revenue-share payments to Microsoft continue through 2030, capped at $38 billion in total. [11] [7]
• Amazon: an existing $38 billion multi-year commitment expanded in early 2026 by $100 billion over eight years, including obligations tied to AWS chips—about $138 billion combined, per Amazon’s June 30, 2026 10-Q language as reported. [11] [29]
• CoreWeave: 2025 contracts of up to $11.9 billion, up to $4 billion, and up to $6.5 billion, totaling about $22.4 billion. [30]

Press-reported and not independently confirmed in the counterparty’s filings under OpenAI’s name in the sources reviewed here: Oracle at more than $300 billion over roughly five years for about 4.5 GW (Wall Street Journal, September 2025). Stargate’s $500 billion over four years (January 2025) is an investment intention, not a single take-or-pay invoice. Chip-side figures (Nvidia up to $100 billion of investment tied to at least 10 GW of systems; AMD 6 GW) mix equity, warrants, and purchase intent. Adding every headline produces a trillion-dollar stack that double-counts capacity OpenAI will not pay for twice. [31] [32] [33]

The number that belongs next to the revenue forecast is the internal spend plan, not the sum of press releases. July’s presentation puts computing power and infrastructure at about $856 billion through 2030—roughly $171 billion a year against a September net run rate of $50 billion, or about 3.4 times current annualized revenue. A July Wall Street Journal report had already described a compute spending plan raised to about $750 billion through 2030. Against the three supplier figures that are in filings ($250 billion Microsoft + $138 billion Amazon + up to $22.4 billion CoreWeave ≈ $410 billion), even the grossed-up $70 billion run rate is under six years of revenue, and the $50 billion net figure is over eight—before Oracle, training clusters, and power. Those contracts are largely take-or-pay or long-dated; they do not shrink if a headline run rate is restated. [16] [29] [11]

Figure table

Definitions matter more than the headline. “Annualized” means a short-period pace times 12. “Recognized” means period revenue. “Net” means OpenAI’s own share. “Gross-comparable” means an investor construction that grosses up partner sales. “Plan” means an internal projection reported by the press, not a result.

Date of figure	Metric	Amount	Definition	Source (report date)
2023	Annualized revenue	$2B	Company-stated ARR	Friar blog, via Reuters (Jan 19, 2026) [12]
2024	Recognized revenue	$3.7B	Audited full-year revenue	Zitron docs, FT-verified (Jun 16, 2026) [13]
2024	Annualized revenue	$6B	Company-stated ARR	Friar blog, via Reuters (Jan 19, 2026) [12]
2024	Operating loss	$8.78B	Audited loss from operations	Same audited set (Jun 2026) [22]
2025	Recognized revenue	$13.07B	Audited full-year; ahead of internal $10B target	FT / Zitron (Jun 16, 2026) [13]
End-2025	Annualized revenue	>$20B	Company-stated ARR; monthly revenue nearly $2B by December	Friar (Jan 2026); FT (Jun 2026) [12] [13]
2025	Operating loss	$20.92B	Audited; R&D $19.18B; total costs ~$34B	FT / Zitron (Jun 2026) [21]
2025	Net loss attributable to company	~$38.5B	Includes ~$41.55B non-cash conversion charge; adjusted loss ~$8B per FT source	FT (Jun 16, 2026) [13]
2025	Payments to Microsoft	$17.2B	R&D, cost of revenue, and other	Audited docs via Quartz/FT (Jun 2026) [21]
2025	Cash burn	~$8B	Internal actual vs plan	The Information (Feb 20, 2026) [23]
End-2025	Cash	Roughly half of >$50B assets	Audited balance sheet	FT / Zitron (Jun 2026) [21]
Q1 2026	Recognized revenue	$5.7B	Told to investors	WSJ (Aug 18, 2026) [15]
Q1 2026	Operating loss (incl. SBC)	$9.3B	Investor disclosure	Secondary reports of WSJ figures (Aug–Sep 2026) [26]
Q2 2026	Recognized revenue	$6.7B	+18% QoQ; Anthropic Q2 $11.5–11.6B	WSJ (Aug 18, 2026); Reuters (Oct 8, 2026) [15] [3]
Q2 2026	Operating loss (incl. SBC)	$12.3B	Widened vs Q1	Secondary reports of investor figures [25]
Mid-Aug 2026	Annualized revenue	>$40B	Net basis; still included some inbound Microsoft model-resale revenue that later stopped	Bloomberg, via later reporting (Aug 2026) [34] [9]
End-Sep 2026	Annualized revenue	~$50B (“approaching”)	OpenAI’s figure to investors; own share only (net of partner cut)	FT (Oct 8); Bloomberg, CNBC, Reuters same week [1] [2]
Late Sep 2026	“Comparable” annualized revenue	~$68–70B	Investor gross-up to match Anthropic’s partner-sales method; not OpenAI’s reported revenue	Axios (Sep 29); FT/CNBC explanation (Oct 8) [19] [4]
Q3 2026	Run-rate growth	77% total; 107% enterprise	Investor presentation; growth rates, not dollar levels	CNBC, person familiar (Oct 2026) [11]
End-2026	Annualized revenue target	≥$70B	Company forecast on its own basis, enterprise-led; not recognized FY revenue	Bloomberg (Oct 9, 2026) [2]
FY 2026	Revenue plan	$36B	Internal projection of recognized/full-year revenue	FT, July presentation (Sep 18, 2026) [16]
2026	Cash-burn plan (Feb vintage)	$25B	Projected cash outflow for the year	The Information (Feb 20, 2026) [23]
2026–2030	Negative free cash flow	$278B	July plan; May plan was ~$305B	FT (Sep 18, 2026) [16] [17]
Through 2030	Compute and infrastructure	~$856B	July plan, largest expense category	FT (Sep 18, 2026) [16]
2030	Revenue plan	$350B	Internal; cumulative 2026–30 revenue ~$840B	FT (Sep 18, 2026) [16]
2027 / 2028 / 2029 / 2030	Cash flow (Feb vintage)	−$57B / −$85B / −$51B / +$39B	Year-by-year; July plan’s annual split not published	The Information via The Decoder (Feb 21, 2026) [24]
Multi-year	Azure commitment	Incremental $250B	Contracted purchase of Azure services	Microsoft filing, reported Oct 2025 onward [11]
Multi-year	AWS commitment	$38B + $100B over 8 years	Amazon disclosure of expanded arrangement	Amazon 10-Q language, as reported [11]
Through 2029	CoreWeave contracts	Up to ~$22.4B	Three 2025 agreements	CoreWeave disclosures / Reuters (Oct 15, 2026) [30]
Through 2030	Microsoft revenue-share cap	$38B cumulative	20% rate continues, subject to cap	The Information via Reuters (May 12, 2026); CNBC (Apr 27, 2026) [7] [6]
Mar 2026	Equity raised	$122B at $852B valuation	Closed round; modeled exhausted by 2028 under July plan	FT / Reuters (Sep 18, 2026) [17]
Oct 2026	Fundraising talks	≥$30B at $1.4T pre-money	In discussions; not closed in the reports cited	Bloomberg (Oct 9, 2026) [2]

How to use the numbers

Treat three series as non-substitutable. The series OpenAI manages to is net annualized revenue: above $20 billion exiting 2025, above $40 billion in August 2026, approaching $50 billion at the end of September, with a company target of $70 billion or more by year-end. The series investors built in late September (~$68–70 billion) is a gross-up for comparison with Anthropic and should not be cited as OpenAI revenue. The series that pays the bills is recognized revenue and cash flow: $13 billion booked in 2025, about $12.4 billion in the first half of 2026, a $36 billion full-year 2026 plan, and a July projection of $278 billion of negative free cash flow through 2030 against $856 billion of compute and infrastructure.

For a competitor or a supplier, the binding constraint is not whether the exit run rate prints $50 billion or $70 billion in December. It is whether enterprise growth can lift recognized revenue and gross margin fast enough to service take-or-pay capacity that was signed against a much larger ambition—and whether the next private round arrives before the March 2026 capital is modeled to run out in 2028. OpenAI has not commented publicly on the October figures. All of the 2026 run-rate, loss, and plan numbers above are from investor communications and leaked or reviewed internal presentations, not from audited public financials.

---

Recent Findings Supplement (October 2026)

OpenAI's September 2026 annualized revenue run rate (ARR) was reported to investors as approaching $50 billion (net of partner shares), roughly $20 billion below the ~$70 billion figure that circulated in late September 2026.[1][2]

The gap stems from accounting differences rather than a revision of underlying performance. OpenAI reports only its own share of revenue (net), excluding or limiting the full value of sales routed through cloud partners such as Microsoft. Rival Anthropic counts gross sales through partners. Investors had adjusted ("grossed up") OpenAI's figures for direct comparison, producing the higher ~$68–70 billion estimate.[3][4]

OpenAI separately told investors it expects to reach or exceed $70 billion ARR by year-end 2026, driven primarily by enterprise growth (77% total run-rate growth and 107% enterprise growth in Q3).[5][6]

This reconciles the October reports (~$50 billion) against earlier indications (~$70 billion) as a definitional distinction (net vs. gross/partner-adjusted) rather than conflicting actuals or forecasts. Run rate itself is a forward-looking snapshot (recent monthly sales × 12) and differs from recognized/full-year revenue.[4]

Recent compute and cash-flow projections (from a July 2026 investor presentation) show rising infrastructure commitments alongside modestly improved free-cash-flow expectations.[7]

• Compute and infrastructure spend planned at ~$856 billion through 2030 (up from ~$600 billion cited in February 2026 materials).
• Projected negative free cash flow of ~$278 billion over 2026–2030 (improved from ~$305 billion in a May projection).
• Revenue trajectory in the same plan: ~$36 billion in 2026 rising to $350 billion in 2030.

These figures underscore the scale of capex relative to current revenue and the need for sustained growth to support the commitments.[8]

2025 actuals and early 2026 context (referenced in October 2026 reporting) provide baseline for the run-rate trajectory.[6]

• 2025 revenue: ~$13.07 billion with a reported net loss of ~$38.5 billion.
• Started 2026 with ~$20 billion ARR (up from ~$6 billion in 2024).

For competition or entry in this space, the net-vs-gross distinction highlights the importance of standardized metrics when benchmarking AI labs; partner-channel revenue recognition can materially affect perceived scale. Massive multi-year compute commitments create high fixed-cost barriers that favor players with strong enterprise traction and access to capital, while pressuring margins and requiring rapid ARR conversion into sustainable cash flow. The $50 billion net figure (with a $70 billion+ year-end target) suggests continued momentum but also that headline run rates require careful definition to avoid over- or under-estimating true economics.[9]

Summary table of key recent figures (post-April 2026 reporting focus; all are investor-shared or media-reported estimates, not audited GAAP results):

• Metric: Sept 2026 ARR (net, to investors) — Figure: Approaching $50B — Date/Source: Oct 8–9 2026 (FT/Reuters/Bloomberg/CNBC) — Definition: OpenAI's share only.
• Metric: Late Sept 2026 ARR indication (investor-adjusted) — Figure: ~$68–70B — Date/Source: Sept 29–Oct 2026 (Axios/Reuters et al.) — Definition: Grossed-up for Anthropic-style partner sales comparison.
• Metric: Year-end 2026 ARR target — Figure: $70B+ — Date/Source: Oct 9 2026 (Bloomberg) — Definition: Expected net run rate, enterprise-driven.
• Metric: Q3 2026 run-rate growth — Figure: 77% total / 107% enterprise — Date/Source: Oct 2026 investor updates (via CNBC et al.).
• Metric: Compute/infrastructure commitments 2026–2030 — Figure: ~$856B — Date/Source: July 2026 presentation (FT/TNW Sep 19 2026) — Definition: Planned spend.
• Metric: Projected negative FCF 2026–2030 — Figure: ~$278B (vs. prior ~$305B May estimate) — Date/Source: July 2026 presentation.
• Metric: 2025 revenue / net loss — Figure: ~$13.07B revenue / ~$38.5B loss — Date/Source: Referenced in Oct 2026 reports (e.g., Benzinga via Finpresso).
• Metric: 2026 revenue projection (plan) — Figure: ~$36B — Date/Source: July 2026 presentation.
• Metric: 2030 revenue projection (plan) — Figure: $350B — Date/Source: July 2026 presentation.[8]

These updates are the primary new data points from October 2026 coverage; earlier 2026 figures (e.g., Q1/Q2 revenue or February compute plans) are referenced only as context for changes. All monetary values are in USD as reported.