Can OpenAI actually afford all this compute?
In our updated market reports, you will find everything you need
SUMMARY
Yes, OpenAI can afford its compute expansion for now, but it still cannot pay for the full plan from its own operating business.
The uncomfortable number is not simply the roughly $750 billion spending projection. It is the contrast between that long-dated infrastructure ambition and a current revenue pace of about $24 billion a year.
OpenAI is not writing one enormous check. Oracle, Amazon, Microsoft, CoreWeave, SoftBank, Nvidia, AMD, banks and infrastructure developers are funding buildings, chips and power systems upfront, while OpenAI accepts years of future payments.
That structure makes the buildout possible, but it does not make it cheap. Long cloud reservations, leases, chip orders and electricity contracts are much harder to unwind than research projects or hiring plans.
The early commercial evidence is unusually strong. OpenAI says available compute rose from 0.2 gigawatts in 2023 to about 1.9 gigawatts in 2025 while annualized revenue climbed from $2 billion to more than $20 billion, keeping revenue per gigawatt surprisingly stable.
Still, the first two gigawatts are not proof that the next ten or twenty will earn the same return. Training, inference, chip efficiency and utilization can make one gigawatt economically very different from another.
OpenAI’s recent gross margin near 39% suggests that paid usage can cover its direct serving costs. The company is nowhere near mature-software economics, though, once research, staff and infrastructure expansion are included.
Supplier financing is circular, but it is not empty. Amazon, Nvidia and AMD are helping finance a customer that buys their clouds and chips, yet OpenAI also has hundreds of millions of users, tens of millions of subscribers and a rapidly growing enterprise business.
Enterprise agents are the most credible way out. Businesses can pay for completed engineering, research or operational work, giving OpenAI far more pricing room than casual consumer chat, even as those agents consume far more tokens.
The stress test is harsh: even 50% annual revenue growth would produce only about $316.5 billion of cumulative revenue over five years. OpenAI needs something closer to repeated near-doubling, plus better margins, cheaper chips and high utilization.
An IPO could add tens of billions of dollars and widen access to debt, but it would mainly buy time. The durable answer has to come from the economics of the products, not from one more financing event.
OpenAI is making a survivable bet rather than a comfortably affordable one. It can keep building while investors and suppliers remain confident; whether the company can eventually carry the infrastructure itself is still unproven.
Why has OpenAI’s compute bill become so scary?
OpenAI’s compute bill has become frighteningly large because the company now expects to spend around $750 billion through 2030 while generating roughly $24 billion a year at its current revenue pace.
That gap has widened lately. OpenAI’s previous internal projection was reportedly closer to $600 billion. The company has also started moving beyond cloud contracts and taking direct responsibility for parts of the infrastructure.
Project Camellia shows how far the strategy has moved. OpenAI is designing and developing the Georgia data-center project itself, with 3.2 gigawatts of power scheduled to arrive in phases between 2028 and 2032. OpenAI says it will cover the infrastructure and electricity-service costs rather than passing them to local ratepayers.
The broader Stargate program has expanded just as quickly. OpenAI originally aimed to secure 10 gigawatts of US infrastructure by 2029. The company says it has already passed that target, including more than three gigawatts added within 90 days.
A gigawatt is roughly the power demand of a small city. OpenAI is now planning infrastructure across multiple sites whose combined electricity needs can be compared with entire countries.
The financial risk comes from the length of the commitments. OpenAI can cancel a research experiment or reduce hiring fairly quickly. Data-center leases, chip orders, power contracts and dedicated cloud capacity can run for five, eight or even ten years. Once construction starts, stepping away becomes expensive.
What does “afford” really mean for OpenAI?
OpenAI can fund the next few years of compute today, while sustainable affordability remains unproven.
The word “afford” hides three separate questions.
The first concerns cash. Can OpenAI make the payments that come due soon? Its recent fundraising and cash reserves make that likely.
The second concerns access to capital. Will investors, banks, cloud companies and chipmakers keep financing OpenAI as the infrastructure expands? They currently appear willing to do so.
The third question is much harder. Can ChatGPT, Codex, enterprise agents, the API, advertising and future products eventually produce enough gross profit to cover the full cost of running OpenAI?
The first two tests tell us whether OpenAI can keep building. The third tells us whether the strategy produces a durable company.
OpenAI currently passes the first two tests and fails the third. Its products generate substantial revenue, while the business continues to consume far more cash than it produces.
That position can continue for years when investors remain confident. It becomes dangerous when capital providers start asking for proof instead of promises.
If you want more recent data on this point, please see our latest frontier AI labs market report.
How much compute has OpenAI already promised to buy?
OpenAI has already signed at least $460.4 billion of publicly identified cloud commitments, and several large hardware and Microsoft agreements sit outside that total.
The largest reported contract is OpenAI’s agreement to buy around $300 billion of Oracle computing capacity over roughly five years, beginning in 2027.
OpenAI has also expanded its AWS commitment to about $138 billion. The relationship began with a $38 billion agreement covering Nvidia-based infrastructure and later added $100 billion of AWS Trainium capacity over eight years.
CoreWeave says its agreements with OpenAI have reached as much as $22.4 billion. Those contracts cover reserved cloud capacity for training advanced models and running demanding workloads.
Adding those three figures gives us $460.4 billion. That total remains incomplete. OpenAI continues to use Microsoft Azure, has added Google Cloud and is developing infrastructure with SoftBank, SB Energy and other Stargate partners.
Hardware announcements create another layer. OpenAI plans large deployments involving Nvidia, AMD, Amazon Trainium and custom Broadcom accelerators. Some of that hardware will operate inside the cloud contracts already counted, so adding every announced gigawatt would count the same infrastructure several times.
OpenAI has already committed to several hundred billion dollars of compute, even before every part of the plan has received a public price.
| Provider | Publicly identified commitment | Main purpose |
|---|---|---|
| Oracle | Approximately $300B | Large Stargate cloud deployments |
| AWS | Approximately $138B | Nvidia and Trainium capacity |
| CoreWeave | Up to approximately $22.4B | Training and advanced AI workloads |
| Combined | Approximately $460.4B | Excludes several unpriced or overlapping programs |
Who is paying upfront for OpenAI’s data centers?
OpenAI’s partners are carrying much of the upfront construction bill, while OpenAI commits to years of future payments.
Cloud providers and infrastructure developers generally raise the money for buildings, chips, cooling equipment, power connections and networking. OpenAI then pays through cloud contracts, capacity reservations, leases and usage commitments.
The original Stargate structure made that division unusually clear. SoftBank received financial responsibility, while OpenAI received operational responsibility. Oracle, MGX and other partners supplied additional capital, infrastructure and technology.
The structure lets OpenAI secure assets far beyond what its own balance sheet could purchase today. It also transfers part of the early construction risk to companies with easier access to debt markets.
The newest proposed arrangement pushes that model further. Nvidia is reportedly discussing a roughly $250 billion financial guarantee that would help OpenAI lease a planned 10-gigawatt data-center complex in Ohio. The entire project could cost more than $500 billion once the chips are included.
An Nvidia guarantee would lower the financing cost because lenders could rely partly on Nvidia’s credit strength. OpenAI would gain access to the infrastructure without first raising hundreds of billions of dollars itself.
Oracle shows the risk on the other side of these deals. Oracle recently reported $638 billion of remaining performance obligations, meaning contracted revenue it has yet to recognize. S&P Global estimates that roughly half of that backlog is tied to OpenAI.
S&P recently cut Oracle’s credit rating to BBB-, the lowest investment-grade level, citing heavier infrastructure spending, uncertain profitability and high customer concentration. Wisconsin regulators have also sought billions of dollars of collateral for power infrastructure serving an Oracle data-center project.
OpenAI has persuaded suppliers to finance a remarkable share of its expansion. Those suppliers expect long contracts and future cloud revenue in return. Their willingness to carry that risk is currently one of OpenAI’s most valuable financial assets.
If you want more recent data on this point, please see our latest frontier AI labs market report.
How much revenue is OpenAI making right now?
OpenAI currently generates about $2 billion of revenue a month, giving the company an annualized pace of roughly $24 billion.
Annualized revenue describes the current monthly rate multiplied across a year. It should not be confused with the revenue OpenAI has already recognized over a completed financial year.
OpenAI reportedly recorded around $13.1 billion of revenue in 2025. Its first-quarter revenue then reached approximately $5.7 billion, about three times the level from a year earlier. The recent monthly figure suggests that growth has continued since then.
The customer base is enormous. OpenAI says ChatGPT has more than 900 million weekly users and over 50 million subscribers. That means only a minority of users need to pay for OpenAI to create a very large consumer subscription business.
Business customers now contribute more than 40% of revenue. OpenAI expects enterprise revenue to reach parity with consumer revenue, giving the company a second large engine beyond ChatGPT Plus and other individual subscriptions.
Advertising is also arriving faster than expected. OpenAI says its initial advertising pilot crossed $100 million in annualized revenue within six weeks. That remains small beside subscriptions, although the early pace suggests that ChatGPT’s free audience can become commercially valuable.
The company already has the revenue scale of a major software business. Its compute commitments belong to a much larger financial category, closer to a global cloud provider or energy company.
Has more compute really produced more OpenAI revenue?
OpenAI’s own three-year figures show a striking pattern: available compute rose 9.5 times and annualized revenue rose about ten times.
Available capacity increased from 0.2 gigawatts in 2023 to 0.6 gigawatts in 2024 and approximately 1.9 gigawatts in 2025. Annualized revenue climbed from $2 billion to $6 billion and then more than $20 billion.
That works out to around $10 billion of annualized revenue per gigawatt in each period.
The stability is unusual. OpenAI added capacity, released stronger products and found enough customer demand to monetize the extra supply at almost the same rate for three consecutive years.
The pattern supports OpenAI’s claim that limited compute has constrained growth. ChatGPT has faced usage caps, API customers have encountered capacity limits and new products have frequently launched gradually because OpenAI could not serve everyone immediately.
Still, three observations cannot establish a permanent rule. One gigawatt used for mature inference may produce a very different return from one used for training experiments. Newer chips also deliver more useful computation from the same amount of electricity.
OpenAI has proved that its first two gigawatts helped create a rapidly growing business. The company now needs to repeat that performance across a fleet several times larger.
| Year | Available compute | Annualized revenue | Annualized revenue per gigawatt |
|---|---|---|---|
| 2023 | 0.2 GW | $2B | Approximately $10B |
| 2024 | 0.6 GW | $6B | Approximately $10B |
| 2025 | Approximately 1.9 GW | More than $20B | More than $10.5B |
Can OpenAI survive its current losses?
OpenAI can survive its current losses for several years, although the business remains far from self-funding.
OpenAI reportedly produced an operating loss of around $20.9 billion on $13.1 billion of revenue in 2025. The company spent roughly $1.60 more than it earned for every dollar of revenue.
The previous year looked even worse in percentage terms. OpenAI generated approximately $3.7 billion of revenue and lost around $8.8 billion from operations. The operating loss equaled about 237% of revenue in 2024, falling to roughly 160% in 2025.
That improvement deserves attention. Revenue grew faster than the operating loss, showing some early operating leverage. The absolute loss still increased by more than $12 billion.
More recent shareholder figures reportedly showed $5.7 billion of quarterly revenue, a $3.7 billion quarterly cash burn and a gross margin near 39%. A 39% gross margin means OpenAI keeps around 39 cents after the direct cost of delivering each dollar of product revenue, before paying for research, staff, sales and other operating expenses.
That margin is much lower than the 70% to 90% often associated with mature software companies. It is high enough to show that paying customers can generate positive product-level economics.
OpenAI’s financial cushion is currently huge. The company closed a funding round with $122 billion of committed capital at an $852 billion valuation. It also expanded its undrawn revolving credit facility to approximately $4.7 billion.
Reported cash and marketable securities exceeded $73 billion after the first quarter. At the recent quarterly burn rate, that cash alone would theoretically cover several years.
The burn will probably rise as new capacity arrives. Committed capital may also be delivered in stages or depend on conditions. OpenAI has plenty of runway today, yet the company will need further funding if infrastructure spending follows the current plan.
If you want more recent data on this point, please see our latest frontier AI labs market report.
Are Nvidia, Amazon and AMD financing their own customer?
OpenAI’s supplier financing is circular, yet real customer demand currently gives the structure economic substance.
Amazon has committed $50 billion of investment to OpenAI while securing a cloud relationship worth around $138 billion. Amazon gains a major AWS customer, demand for Trainium chips, OpenAI products on Bedrock and custom models for Amazon services.
Nvidia intends to invest as much as $100 billion as OpenAI deploys up to ten gigawatts of Nvidia systems. Nvidia benefits twice when the plan moves forward: its investment may appreciate and the infrastructure ecosystem buys millions of its chips.
AMD issued OpenAI warrants covering as many as 160 million AMD shares. The warrants vest as OpenAI deploys up to six gigawatts of AMD hardware and as AMD reaches commercial and share-price milestones. AMD expects the partnership to generate tens of billions of dollars of revenue.
These arrangements give OpenAI capital or potential equity value that can help it purchase infrastructure from the same companies. The suppliers gain sales, long-term customers and a stake in OpenAI’s success.
Supplier finance has existed for decades in aircraft, telecommunications, energy and heavy industry. The danger appears when the financing creates purchases that end customers never justify.
OpenAI currently has hundreds of millions of users, tens of millions of subscribers, rapidly growing business revenue and substantial API demand. There is real activity beneath the financial engineering.
A downturn would expose the circularity quickly. Falling OpenAI demand would reduce the value of the suppliers’ investments, weaken their largest future customer and leave newly built infrastructure underused at the same time.
Can multiple chip suppliers lower OpenAI’s compute costs?
OpenAI’s multi-chip strategy should lower long-term compute costs, provided the company can run workloads efficiently across very different systems.
OpenAI once depended heavily on Microsoft Azure and Nvidia GPUs. It now uses Microsoft, Oracle, AWS, CoreWeave and Google Cloud, alongside several chip architectures.
Its announced hardware programs include up to ten gigawatts of Nvidia systems, six gigawatts of AMD GPUs, around two gigawatts of Amazon Trainium capacity and ten gigawatts of OpenAI-designed accelerators developed with Broadcom.
Those figures overlap. A Broadcom or Nvidia system can sit inside a Microsoft, Oracle or OpenAI data center. The announcements describe supplier relationships and hardware pipelines rather than a clean total of separate physical capacity.
The variety still gives OpenAI stronger bargaining power. Frontier training can use the most capable hardware. Predictable, high-volume inference can move to cheaper systems. Less urgent jobs can run when capacity and electricity are less expensive.
OpenAI’s Jalapeño chip offers the clearest example. OpenAI designed the accelerator with Broadcom specifically for large-language-model inference. Early laboratory testing reportedly shows substantially better performance per watt than existing systems, although final performance data has yet to be published.
Custom chips could remove expensive features that OpenAI rarely uses and optimize memory, networking and software around its own models. Google followed a similar path with TPUs, while Amazon developed Trainium and Inferentia.
The challenge is software. OpenAI must keep models, kernels, networking and scheduling efficient across Nvidia, AMD, Trainium and its own processors. Poor utilization can erase the savings from cheaper hardware.
The strategy should reduce OpenAI’s cost per useful task over time. It also gives the company an escape route if one supplier’s prices, delivery schedule or technical roadmap becomes unattractive.
Can cheaper OpenAI tokens rescue the business?
Cheaper OpenAI tokens help enormously, although agents are increasing total token consumption faster than simple chat ever did.
The original GPT-4 API cost $30 per million input tokens and $60 per million output tokens. OpenAI’s current cost-focused GPT-5.6 Luna model charges $1 for input and $6 for output.
Headline input prices have therefore fallen by 30 times, while output prices have fallen by ten times. The models differ in capability, but the comparison still shows how aggressively OpenAI has pushed down the cost of serving useful AI.
Hardware improvements explain part of the reduction. OpenAI also saves money through caching, batching, model routing, quantization, better kernels and higher hardware utilization.
Lower prices expand the market. A developer who could afford one million tokens can now buy far more work for the same budget. Tasks that once looked too expensive become normal product features.
That creates a new problem. OpenAI’s APIs already process more than 15 billion tokens per minute. Agents consume many more tokens than a single chatbot response because they read files, search, plan, call tools, check results, correct mistakes and continue working.
Reasoning models may also spend large numbers of hidden tokens before returning a short answer. Persistent agents can keep context and remain active for hours or days.
A falling token price can therefore produce a rising total compute bill. Electricity became cheaper and society used more of it. Computing followed the same pattern for decades.
OpenAI needs the cost of completing a useful task to fall faster than the price customers are willing to pay for that task. Token efficiency alone cannot settle the equation.
If you want more recent data on this point, please see our latest frontier AI labs market report.
Will enterprise agents make OpenAI’s compute affordable?
Enterprise agents are OpenAI’s best route to making compute affordable because companies can pay for completed work rather than cheap conversation.
A consumer may compare a ChatGPT subscription with entertainment services or other personal apps. A business compares an AI agent with the cost of an employee, contractor, outsourced process or delayed project.
That creates much more room for OpenAI to charge according to value. An agent that saves an engineering team hundreds of hours can justify thousands of dollars of compute. A chatbot that answers casual questions cannot.
OpenAI’s revenue mix is already moving in that direction. Enterprise products and APIs account for more than 40% of revenue. OpenAI expects business revenue to match consumer revenue, which would make companies responsible for roughly half of the business.
Codex provides an early test. The coding agent now serves more than five million weekly users. Coding tasks often require long sessions, repeated tool calls and substantial context, making Codex expensive to operate. Successful tasks can also create far more economic value than ordinary chat.
OpenAI Frontier pushes the same model into broader company workflows. Businesses can deploy agents with shared context, permissions, identity and access to internal systems. AWS will distribute Frontier to enterprise customers and provide persistent runtime environments for those agents.
The commercial opportunity goes beyond selling more tokens. OpenAI can charge for managed environments, security, governance, integrations, reliability and completed outcomes.
Competition will keep prices under pressure. Companies can send simple tasks to cheaper OpenAI models, Anthropic, Google or open-weight systems. Large customers will also negotiate volume discounts.
Enterprise growth improves the economics only when contract value rises faster than compute consumption. OpenAI’s future depends heavily on achieving that.
What happens if OpenAI’s growth slows?
OpenAI’s compute plan becomes financially uncomfortable when annual revenue growth falls much below a near-doubling pace.
We can see the problem with a simple stress test. Using the $24 billion annualized revenue pace discussed above, we calculated how much cumulative revenue OpenAI would generate over five years under three growth scenarios.
At 50% annual growth, cumulative revenue reaches approximately $316.5 billion. That would be extraordinary performance for almost any company, yet it covers less than half of the reported compute-spending projection.
At 75% annual growth, cumulative revenue reaches around $493.2 billion. The gap remains large before paying staff, research costs, sales expenses and other operating bills.
Doubling revenue every year produces approximately $744 billion of cumulative revenue. That finally approaches the infrastructure projection, although revenue and spending cannot be compared as if every revenue dollar were available to pay cloud providers.
OpenAI may negotiate, delay or cancel parts of the plan. Some spending is conditional on deployment, and partners will finance portions of the physical assets. New products could also increase revenue faster than these scenarios assume.
The calculation still shows the scale of the bet. OpenAI needs several years of exceptional growth, combined with improving margins, to make the commitments feel comfortable.
| Annual revenue growth | Approximate five-year cumulative revenue | What it would mean |
|---|---|---|
| 50% | $316.5B | Massive business, still far below planned compute spending |
| 75% | $493.2B | Stronger coverage, with a large remaining gap |
| 100% | $744B | Near the compute projection before any other expenses |
Can an OpenAI IPO solve the funding gap?
An OpenAI IPO could finance another stage of the buildout, while stronger operating economics would still be required afterward.
OpenAI has confidentially submitted a draft registration statement to the US Securities and Exchange Commission. The company says it has yet to decide when to proceed.
Going public would give OpenAI access to a much larger pool of investors. It could raise new equity, issue public debt more easily and use listed shares to compensate employees or finance acquisitions.
The scale of the compute plan limits what an IPO can achieve. At OpenAI’s recent $852 billion private valuation, issuing shares equal to 10% of the company would theoretically raise about $85 billion before fees, discounts and market effects.
That would be one of the largest equity raises ever completed. It would still cover only a fraction of the infrastructure under discussion.
Repeated equity raises could provide more money, although existing shareholders would face substantial dilution. Public investors would also demand clearer answers about gross margins, cash burn, long-term contracts and the return produced by each new gigawatt.
An IPO buys time and broadens the funding base. It cannot turn low-return infrastructure into high-return infrastructure.
Can OpenAI actually afford all this compute?
Yes, OpenAI can afford its compute expansion for now, although the company still depends on investors and partners to carry a large part of the risk.
OpenAI has assembled enough cash, committed capital, credit and supplier support to keep building through the next phase. An immediate funding crisis looks unlikely.
The business underneath the buildout is also real. OpenAI generates around $2 billion of monthly revenue, serves hundreds of millions of people and has built a fast-growing enterprise operation. Its first 1.9 gigawatts of available compute accompanied a tenfold increase in annualized revenue.
The unresolved problem is scale. OpenAI remains deeply unprofitable, and its publicly identified cloud commitments already reach several hundred billion dollars. Its newest projects are pushing the company closer to direct data-center development, which increases its responsibility when costs overrun or demand arrives late.
Partners have absorbed much of the upfront burden. Oracle, Amazon, Microsoft, CoreWeave, SoftBank, Nvidia, AMD, Broadcom, banks and infrastructure investors are financing buildings, chips and energy systems that OpenAI could never purchase alone today.
That model works while OpenAI keeps growing and capital providers remain confident. It starts to crack when revenue slows, margins stay low or lenders demand stronger guarantees.
As seen above, the roughly $750 billion compute projection would require several more years of exceptional revenue growth. OpenAI also needs enterprise agents to create enough customer value, custom hardware to lower unit costs and supplier competition to keep contract prices under control.
Our judgment is simple: OpenAI can finance all this compute today, while its ability to pay for it from its own business remains unproven.
The company has built a strong enough commercial engine to justify continued expansion. It has yet to build one strong enough to carry the entire infrastructure plan without outside help.
OpenAI is making a survivable bet rather than an affordable one. If revenue keeps compounding near its recent rate, the compute will look visionary. A sustained slowdown would leave OpenAI and its partners with one of the largest infrastructure mismatches ever created by a private company.
If you want more recent data on this point, please see our latest frontier AI labs market report.
OUR METHODOLOGY
This analysis tests whether OpenAI can carry its rapidly expanding compute plan. We separate near-term financing capacity from long-term affordability, because cash, committed capital and supplier support can keep construction moving even when the operating business is not yet self-funding.
We evaluate six connected dimensions: the size and duration of OpenAI’s infrastructure commitments, who provides the upfront capital, OpenAI’s access to future financing, current revenue and losses, the commercial return associated with additional compute, and the consequences of slower growth.
We give the most weight to disclosed contracts, funding commitments, company announcements, regulatory filings, reported financial figures and infrastructure already under development. Long-range capacity goals are useful context, but they are weaker evidence than signed agreements or projects with named locations, partners and deployment schedules.
Cloud commitments, hardware programs and data-center developments are kept separate when they appear to describe overlapping layers of the same infrastructure. A Nvidia, AMD, Trainium or Broadcom deployment may sit inside an Oracle, AWS, Microsoft or OpenAI-operated facility, so adding every announced dollar or gigawatt would create double counting.
The approximately $460.4 billion commitment figure includes the publicly identified Oracle, AWS and CoreWeave agreements. It excludes several unpriced or potentially overlapping relationships, including Microsoft Azure, Google Cloud, SoftBank-led projects and parts of the broader hardware roadmap.
Current liquidity and partner financing are used to answer whether OpenAI can keep building now. Revenue growth, gross margin, cash consumption and the economics of enterprise products are used to judge whether the company could eventually support the infrastructure from its own business.
The revenue-per-gigawatt comparison is treated as an observed historical pattern, not a permanent law. Training and inference workloads have different economics, newer chips produce more useful computation per unit of power, and utilization can vary sharply across facilities.
The five-year growth scenarios are stress tests rather than forecasts. They show how much cumulative revenue OpenAI would generate from a $24 billion annualized starting point under 50%, 75% and 100% annual growth, without assuming that every revenue dollar could be used to pay infrastructure providers.
Key sources include OpenAI’s funding, revenue and compute update, OpenAI’s infrastructure strategy, the original Stargate announcement, the Oracle and SoftBank Stargate expansion, the 4.5-gigawatt Oracle partnership, and the SB Energy partnership.
We also used AWS on the expanded OpenAI agreement, OpenAI on the Nvidia partnership, AMD’s annual report, S&P Global’s analysis of AI-infrastructure credit risk, OpenAI’s user and subscriber update, OpenAI’s enterprise revenue update, current GPT-5.6 Luna pricing, and OpenAI’s confirmation of its confidential S-1 submission.
Looking across those sources supports the distinction at the center of the article: OpenAI has enough capital, partners and commercial momentum to finance the expansion for now, but it has not yet demonstrated that its operating business can sustainably carry the full cost.
