Bernstein estimates that OpenAI’s annual spending on infrastructure will peak at $184 billion in 2031, according to a Bernstein chart posted on X by Small Cap Snipa on 3 November and by P/E Research on 4 November. The chart’s header says the estimate covers contracted spending and “includes both leases and assets”.
On the same chart, the share going to hyperscalers falls from about 82 per cent of the 2026 bar to about 64 per cent at the 2031 peak and about 15 per cent in 2035, by our approximate read of the chart. Hardware and co-location vendors take most of the rest.
Business Pill · Vertical Integration
An evergreen Business Engineer explainer on vertical integration: buying or building parts of the supply chain instead of renting them. It teaches the general idea only and does not discuss OpenAI, Bernstein or this estimate.
The key insight: Bernstein estimates OpenAI’s annual infrastructure spending peaks at $184 billion in 2031, as relayed on X. On the same chart the hyperscalers’ share of each year’s bill falls from about 82 per cent in 2026 to about 64 per cent in 2031 and about 15 per cent in 2035, by our approximate read of the chart.
Small Cap Snipa wrote at 22:06 UTC on 3 November that “Bernstein estimates OpenAI’s infrastructure spend will top out at $184B in 2031”, adding: “The model estimates roughly ~$1.25-1.30T in AI infrastructure spend from 2025-2035”.
P/E Research quoted that post at 05:05 UTC on 4 November: “Bernstein expects estimated annual spending by OpenAI on infrastructure to peak at $184 billion in CY31”.
Both posts carry the same image, labelled “EXHIBIT 1” with the footer “Source: Bernstein estimates”. The note itself goes to Bernstein clients; we have not read it, and the posts do not name its analysts or its date.
Bernstein’s estimate of OpenAI’s annual contracted infrastructure spending, leases and assets, 2025 to 2035, with the hyperscaler segment beside each total. The $184 billion 2031 peak is stated on the chart; every other value is our approximate read of the chart image relayed on X by Small Cap Snipa and P/E Research, not independently verified. The hyperscaler shares are our arithmetic.Animated: Bernstein’s estimate of OpenAI’s annual contracted infrastructure spending, 2025 to 2035, as relayed on X. The $184 billion 2031 peak is stated; other values are our approximate read of the relayed chart, and the hyperscaler shares are our arithmetic.
The Shape of the Curve
The chart shows eleven yearly bars from 2025 to 2035. Reading them against the chart’s gridlines, the bars rise from about $4 billion in 2025 to about $40 billion in 2026, about $88 billion in 2027 and about $139 billion in 2028.
They reach about $163 billion in 2029 and about $177 billion in 2030 before the stated $184 billion peak in 2031. After that they fall to about $173 billion in 2032, about $135 billion in 2033, about $99 billion in 2034 and about $91 billion in 2035.
Our read of the eleven bars adds up to about $1.29 trillion, inside the “~$1.25-1.30T” range that Small Cap Snipa relayed. Every bar except 2031 is our approximation, to the nearest billion.
Who Gets Paid
The chart splits each bar five ways: “Hyperscalers”, “neoclouds”, “co-location vendors”, “self-built datacenters” and “Hardware (semis etc.)”.
By our read, hyperscalers account for about $33 billion of the roughly $40 billion in 2026 and about $117 billion of the $184 billion in 2031. Their segment is largest in 2030, at about $127 billion.
By 2035 the hyperscaler segment is about $14 billion of about $91 billion. Hardware is about $38 billion and co-location vendors about $37 billion that year, by our read. Neoclouds show as a thin band from 2026 to 2031, largest in 2028 at about $16 billion.
As we read the chart, the business lever is supplier revenue: on Bernstein’s estimate as relayed, most of OpenAI’s yearly bill goes to cloud providers until about 2031, and a growing share goes to hardware and co-location vendors after that.
Against OpenAI’s Own Plan
In July The Wall Street Journal reported that OpenAI had raised its projected computing-power spending through 2030 to about $750 billion, “up from roughly $600 billion earlier this year”, as relayed by Yahoo Finance and TechCrunch.
Bernstein’s bars for 2025 to 2030 add up to about $610 billion, by our arithmetic, about $140 billion below that plan. The two figures are not like-for-like: one is projected compute spending, the other estimated contracted infrastructure spending including leases and assets.
On the funding side, the FT reported on 31 October that OpenAI was seeking about $20 billion at a $1.25 trillion valuation, which we covered in OpenAI Cuts Its Funding Target to About $20B at $1.25T. The $184 billion peak year is about nine times that round, by our arithmetic.
The Posts That Carried the Estimate
Small Cap Snipa, an account that describes itself as stock market research on AI, data centres and power, posted the Bernstein chart first, at 22:06 UTC on 3 November, with the $184 billion peak and the 2025-2035 range. It matters because it is the earliest relay we found of the chart.
P/E Research quoted it seven hours later with the same chart and a list of tickers. It matters because it frames the figure as annual spending, the unit the chart uses.
Our AI compute contracts tracker lists the disclosed rental contracts, with OpenAI’s $250 billion Azure commitment the largest single figure.
The Structural Read
As we read the chart, Bernstein models two phases: a build phase to 2031 in which most of the money goes to hyperscalers, and a later phase in which hardware and co-location vendors take most of a smaller bill.
As we read it, the estimate sits below OpenAI’s own reported plan through 2030 on a sum of the bars, but the two figures use different definitions, so the gap says more about scope than about a cut.
Bernstein chart header, as relayed on X
“Estimated annual contracted spending by OpenAI on infrastructure tops out at $184B in CY31 and includes both leases and assets”
Three Implications
FOR HYPERSCALERS By our read, their segment is about $117 billion of the $184 billion in 2031 and about $14 billion of about $91 billion in 2035, on Bernstein’s estimate as relayed.
FOR HARDWARE AND CO-LOCATION VENDORS By our read, hardware is about $38 billion and co-location vendors about $37 billion of the 2035 bar; together they are most of that year’s bill.
FOR OPENAI’S FUNDING The FT reported OpenAI was seeking about $20 billion; the $184 billion peak year is about nine times that, by our arithmetic.
The framework puts it this way: “The $1T capex cycle is being underwritten by a $1T concentrated demand commitment from two counterparties whose combined free cash flow is deeply negative.” It names OpenAI and Anthropic as those two counterparties.
As we read it, Bernstein’s chart, as relayed, puts a yearly schedule on one of those two commitments: most of OpenAI’s estimated bill through 2031 is hyperscaler spending. The framework predates the Bernstein estimate and does not discuss it.
What Is Not Established
We have not seen the Bernstein note, its assumptions, its analysts or the date it was issued. The figures here come from two X posts and the chart image they share, and we have not verified them independently.
Apart from the stated $184 billion for 2031, every yearly and segment value in this piece is our approximate read of the chart image. In searches up to about 06:10 UTC on 4 November we did not find an outlet write-up of the note. OpenAI has not commented on the estimate in any source we read.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Bernstein estimates OpenAI’s annual infrastructure spending peaks at $184 billion in 2031, as relayed on X. On the same chart the hyperscalers’ share of each year’s bill falls from about 82 per cent in 2026 to about 15 per cent in 2035, by our approximate read.
A note on sourcing. The Bernstein note goes to clients and we have not read it. The $184 billion peak and the 2025-2035 range come from posts on X by Small Cap Snipa and P/E Research and the Bernstein chart they share; we have not verified them independently. Other yearly and segment values are our approximate read of that chart; the shares, sums and comparisons are our arithmetic. The $750 billion plan is The Wall Street Journal’s report as relayed by Yahoo Finance and TechCrunch.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
