W.061Bernstein puts OpenAI’s compute contracts at $614B of spending through 2030, peaking at $184B in 2031. OpenAI’s own plans have ranged from $600B to $856B
Bernstein analyst Mark Moerdler estimates that OpenAI’s disclosed compute contracts imply about $614B of infrastructure spending through 2030, below the widely cited $750B, and only if every contract is delivered on time. On his estimate, spending peaks at about $184B in 2031. The four cloud contracts he counts add up to about $710B, more than the 2030 total, partly because they run years beyond it. OpenAI’s own plans for the same period have ranged from $600B to $856B this year.
What Bernstein said
Bernstein’s figure is lower than the contract headlines because the bills arrive in stages. As PANews relayed the note, Bernstein argues that compute is delivered in batches and paid as facilities come online. According to Zhitong Finance’s write-up, it treats chips from Nvidia, Broadcom and AMD as likely financed by the suppliers and paid over about six years, says warrants attached to the AMD and Cerebras deals could make those chips almost free to OpenAI, and leaves Google out for lack of verifiable contracts. It counts a roughly $20B data center shell in Georgia upfront. The $614B assumes every contract is delivered on time, which Zhitong reports Bernstein calls extremely unlikely.
The path rises steeply before it peaks. The relays state about $139B for 2028, up about 58%, or $129B without the Georgia shell, and a peak of about $184B in 2031. Read from Bernstein’s chart, spending is about $40B this year and about $88B next year. The note’s date has not been reported. A chart from it was posted on X late on November 3, FourWeekMBA relayed it on November 4, Zhitong Finance wrote it up on November 5, and Investing.com and PANews reported it on November 8.
About $40B this year, about $139B in 2028, about $184B at the 2031 peak
OpenAI’s annual contracted compute spending in Bernstein’s estimate, $ billions, assuming every contract is delivered on time; the series includes leases and assets. Pink bars are figures stated in reports of the note (2028 and the 2031 peak); the others are FourWeekMBA’s readings of Bernstein’s chart. 2025–2030 sum to about $611B on those readings, against the $614B stated.
The contracts
Most of the value sits with Oracle and Microsoft. Microsoft said in October 2025 that OpenAI had contracted to buy an incremental $250B of Azure services. The Wall Street Journal reported a roughly $300B Oracle contract over about five years starting in 2027; Bernstein describes several five-to-six-year contracts. Oracle has not confirmed the figure. Amazon’s two agreements, $38B over seven years and $100B over eight, come to $138B. CoreWeave’s three come to up to $22.4B, which Bernstein treats as ceilings rather than guaranteed revenue. Bernstein rates Oracle and Microsoft outperform and CoreWeave underperform, and places the risk with speculative builds and smaller cloud providers rather than with the large labs’ contracts.
Bernstein’s case is that revenue can outrun the bill. OpenAI’s annualized revenue was about $50B in September, the FT reported. Bernstein expects 2027 revenue above $80B, more than 2027 spending excluding the part of the Georgia shell falling that year, about $73B by our arithmetic ($88B read from the chart less $15B), and says revenue stays ahead as long as it grows at least 60% a year. The 2031 peak of $184B is 3.7 times September’s annualized revenue.
The four cloud contracts come to about $710B
OpenAI contracts with cloud providers that Bernstein counts, $ billions, as announced or reported. Segments are separate agreements: Amazon’s $38B (2025) and $100B (2026); CoreWeave’s up to $11.9B, $4B and $6.5B (all 2025). Oracle’s figure was reported by the Wall Street Journal and has not been confirmed by Oracle. The terms run past 2030, so the total exceeds Bernstein’s spending through 2030; it also excludes chip deals and the Georgia shell, which Bernstein’s path includes.
Against OpenAI’s own numbers
OpenAI’s plans for compute through 2030 have moved this year. In February it set a target of about $600B, CNBC reported; in July the Wall Street Journal reported $750B; and in September the FT reported a company presentation showing about $856B, with about $278B of negative cash flow from 2026 to 2030. Those are spending plans, not contracts, and the reports do not all use the same scope. In November 2025 Sam Altman put commitments at about $1.4T over eight years. On November 6 OpenAI closed an $11B first tranche of a round at a $1.25T valuation, a source told Reuters, 47% above its March price.
Bernstein’s $614B sits near OpenAI’s February target, below its later plans
Estimates of OpenAI’s compute spending through 2030, $ billions, with the date and outlet that reported each. Bernstein’s is its estimate of what disclosed contracts imply if delivered on time; the others are OpenAI’s own plans or projections as reported, and their scope may differ.
The figures behind the charts
| $B | 2025 | 2026 | 2027 | 2028 | 2029 | 2030 | 2031 |
|---|---|---|---|---|---|---|---|
| Contracted spending | ~4 | ~40 | ~88 | 139 | ~163 | ~177 | 184 |
| Provider | $B | Agreements |
|---|---|---|
| Oracle | 300.0 | about $300B, reported (WSJ, Sep 2025); Oracle has not confirmed the figure |
| Microsoft | 250.0 | $250B of incremental Azure services (Oct 2025) |
| Amazon | 138.0 | $38B over 7 years (Nov 2025); $100B over 8 years, about 2 GW of Trainium (Feb 2026) |
| CoreWeave | 22.4 | up to $11.9B (Mar 2025); up to $4B (May 2025); up to $6.5B (Sep 2025) |
What this changes on inthesingularity.ai
- Labs funding pack: Bernstein’s year path (about $40B in 2026, $139B in 2028, $184B in 2031) is a contract-based check on the pack’s OpenAI compute line, which can be set against OpenAI’s own $856B plan.
- Revenue tracker: Bernstein’s 2027 floor of $80B and its 60% growth threshold are a benchmark worth recording beside the tracker’s OpenAI forecast.
- Worth tracking: whether Oracle’s next filing changes the share of its backlog due in years three to five, where most of a contract that ramps in 2028 would sit.
Sources and method
| Item | Figure | Source |
|---|---|---|
| Bernstein | $614B of infrastructure spending through 2030 if contracts are delivered on time, against a widely cited $750B; Oracle about $300B, Microsoft $250B, Amazon $138B, CoreWeave about $22.4B described as caps rather than guaranteed revenue | Investing.com, Nov 8 |
| Bernstein | 2028 about $139B, up about 58%, or $129B without the Georgia shell; peak about $184B in 2031; 2027 revenue above $80B; revenue stays ahead while growth is 60% or more; Google excluded; chips likely supplier-financed | Zhitong Finance, via Sina, Nov 5 |
| Bernstein | Annual values read from Bernstein’s chart: 2025 ~$4B, 2026 ~$40B, 2027 ~$88B, 2029 ~$163B, 2030 ~$177B | FourWeekMBA, Nov 4 |
| Microsoft | OpenAI contracted to buy an incremental $250B of Azure services | Microsoft, Oct 28, 2025 |
| Amazon | $38B over seven years | OpenAI, Nov 3, 2025 |
| Amazon | A further $100B over eight years, about 2 GW of Trainium | Amazon 8-K exhibit, Feb 27, 2026 |
| CoreWeave | Up to $6.5B more, bringing the total with OpenAI to about $22.4B | CoreWeave, Sep 25, 2025 |
| Oracle | About $300B over about five years, starting in 2027, per the Wall Street Journal; Oracle has not confirmed the figure | The Register, Sep 11, 2025 |
| OpenAI | Compute target of about $600B by 2030; 2030 revenue above $280B | Fortune, Feb 20, 2026 |
| OpenAI | Compute spending through 2030 raised to $750B (WSJ report) | Startup Fortune, Jul 2026 |
| OpenAI | About $856B of compute through 2030; 2026 revenue about $36B; about $278B of negative cash flow 2026–2030 (FT) | Reuters, via The Star, Sep 19 |
| OpenAI | Annualized revenue about $50B in September (FT) | Yahoo Finance, Oct 8 |
| OpenAI | $11B closed as the first part of a round at a $1.25T valuation, per a source | Reuters, via Central Oregon Daily, Nov 6 |
| OpenAI | $122B round at a $852B valuation | SiliconANGLE, Mar 31 |
| OpenAI | Sam Altman: commitments of about $1.4T over eight years | TechCrunch, Nov 6, 2025 |
We have not seen Bernstein’s note; its figures come from Zhitong Finance’s write-up of November 5 (the fullest), Investing.com and PANews. A chart from the note was posted on X late on November 3. Annual values other than 2028 and 2031 are FourWeekMBA’s readings of that chart. Contracts: 300 + 250 + 38 + 100 + 11.9 + 4 + 6.5 = 710.4; Amazon 38 + 100 = 138; CoreWeave 11.9 + 4 + 6.5 = 22.4. Below the widely cited figure: 750 − 614 = 136. 2025–2030 on the chart readings: 4 + 40 + 88 + 139 + 163 + 177 = 611. 2028 growth: 139 ÷ 88 − 1 = 58.0% (stated: about 58%). 2027 without the shell: 88 − 15 = 73 (Zhitong puts $15B of the roughly $20B shell in 2027). Peak against revenue: 184 ÷ 50 = 3.68. Valuation: 1250 ÷ 852 − 1 = 46.7%.
Wire reports are written within hours of the news, on a six-hour schedule. Each lists its sources and shows the arithmetic behind its figures. Not investment advice.