ANALYSIS: OpenAI, 750 billion, and the "electric debt" of intelligence

The Number That Eats the Horizon

A Projection, Not a Check

$750 billion. The number is so vast that it immediately risks becoming a myth. First, we must clarify its true nature. The Wall Street Journal reported on July 22 that OpenAI planned to spend roughly this amount on its computing power through 2030.

This is neither an expense that has already been incurred nor an official itemized bill. The information is based on a source familiar with the company’s projections. Earlier in 2026, the figure mentioned was around 600 billion. The projected increase thus amounts to about 150 billion.

The Useful Dizziness

Caution regarding the status of this figure does not diminish its significance. It makes it all the more concerning: a company is already organizing its future around a computational need on a scale comparable to an industrial policy.

One can debate the feasibility, financing, and timeline. We can no longer pretend that artificial intelligence floats in a weightless cloud. Behind the sleek interface lie leases, power plants, power lines, guarantees, chips, and regions compelled to believe in tomorrow.

The cloud now carries the weight of a financial continent.

Computing is no longer just a tool

From Resource to Destiny

For a long time, a tech company bought computing power to build a product. OpenAI seems to be reversing that relationship: it’s building a financial and industrial architecture to ensure computing power remains available, then betting that this capacity will generate the revenue needed to justify it.

This reversal is at the heart of this report. When infrastructure becomes this vast, it no longer simply follows demand. It precedes it, calls for it, and sometimes even compels it. Future models must become useful enough, ubiquitous enough, and profitable enough to fill campuses that do not yet exist.

The promise locked in concrete

Every gigawatt reserved transforms an assumption about AI into a physical obligation. The model may change in a matter of months; a data center lease may last twenty years.

Software loves reversibility. Infrastructure hates it. A chip becomes obsolete, but the debt remains. A training method improves, but the transmission line stays. The company is therefore betting simultaneously on its growth, on the persistent need for computing power, and on its ability to refinance the gap.

The digital future is built on commitments that, unlike the technology itself, cannot be undone.

Georgia Becomes a Unit of Measurement

Project Camellia

In Effingham County, Georgia, OpenAI is developing Project Camellia. The company says it has contracted with Georgia Power for 3.2 gigawatts to be delivered in phases between 2028 and 2032. It has announced an initial commitment of $20 billion for this project.

OpenAI itself acknowledges that significant work remains to be done: infrastructure, phasing, design, financing, and the operational model. This statement deserves as much attention as the financial figure. The campus is a work in progress, not yet a finished machine.

3.2 gigawatts

The key figure may not be 750 billion. It is 3.2 gigawatts for a single project: the electrical measure of an ambition that now demands its own landscape.

OpenAI promises that Georgia families will not subsidize the site, that it will pay the full costs of infrastructure and electricity service, and that the center will be able to reduce its consumption before residential customers are affected during peak times.

An energy promise becomes necessary only when demand is large enough to worry everyone around it.

Paying Its Own Way

The Community Doctrine

In January, OpenAI released its Stargate Community plan. The core commitment seems simple: each campus must cover its own energy costs so that its operations do not drive up local bills. Depending on the site, this involves dedicated generation, storage, new power plants, or power transmission.

The company also proposes that campuses become flexible loads, capable of reducing or interrupting their consumption when the grid experiences a peak. The idea is appealing: a power-hungry infrastructure that behaves as a partner rather than a predator.

The Promise and Its Arbiter

But “paying your own way” is not a self-executing formula. It requires rates, contracts, regulators, guarantees, audits, and the government’s ability to say no.

The mechanisms vary from state to state. Some rely on a dedicated rate, others on private financing for batteries or new capacity. Consumer protection therefore depends not only on OpenAI’s goodwill, but on verifiable clauses and institutions capable of enforcing them.

A public promise is worth exactly as much as the contract that ensures it is fulfilled.

Ohio Reveals the Real Story

The Filing That Dispels the Rumor

On August 17, a regulatory filing by Nvidia gave the matter a clearer shape. Nvidia has entered into several residual value guarantees tied to leases covering approximately 4.25 gigawatts of computing load at the PORTS-Pike campus in Pike County.

Nvidia’s total liability is capped at $105 billion for the initial commitment. It does not trigger an immediate payment. It depends, in part, on the availability of the facilities, which is expected in phases starting in 2028.

What Nvidia Is Actually Guaranteeing

The filing does not state that Nvidia will hand over $105 billion to OpenAI. It describes a conditional safety net: if OpenAI becomes insolvent or defaults on certain payments, Nvidia will cover a defined shortfall.

Nvidia may then take over the lease, seek a new tenant, initiate a sale, or pursue other remedies. OpenAI, for its part, has agreed to reimburse and indemnify Nvidia for the amounts actually paid. The risk is shifted, shared, and managed; it is not eliminated.

AI financing resembles less a fundraising effort than an architecture of mutual survival.

From 250 to 105 billion

The Scale of the Rumor

Before the filing, discussions had circulated about support potentially reaching 250 billion. Reuters later reported that Nvidia was considering an initial commitment of less than 120 billion. The regulatory filing ultimately set a cap of 105 billion for the first tranche.

These figures are not interchangeable. They correspond to different moments and structures: negotiation, reported estimate, documented commitment. Stacking them as if they were successive checks would create a story more spectacular than reality.

A Step Back That Doesn’t Diminish the Scale

Moving from the discussed 250 billion to a capped 105 billion may seem like a step back. Yet, 105 billion in conditional commitments for 4.25 gigawatts remains a colossal exposure.

The adjustment matters above all because it reveals the discipline of capital. Even in the midst of euphoria, investors assess their exposure, limit guarantees, and break the process down into phases. The headlong rush is therefore not blind. It moves forward with lawyers, clauses, and exit strategies.

Financial vertigo wears a seatbelt; it still moves very fast.

Twenty Years Versus Four Years

The Long-Term Lease

SB Energy is to build, own, and operate the Ohio campus, while OpenAI will be the tenant under leases that could last up to twenty years. Nvidia would provide the computing infrastructure and initial credit support. Each player bears a share of the risk.

At the other end of the spectrum, the $750 billion projection ends in 2030. Herein lies the tension: a spending horizon spanning just a few years is underpinned by material commitments that extend much longer. Revenues must follow quickly, but the real estate and power obligations may persist through several generations of models.

The Pace of Generations

Four years in AI can encompass several technical breakthroughs. Twenty years in a lease consist only of payments, renewals, and the patience of creditors.

Nvidia estimates that the site will be able to accommodate several upgrade cycles. This is precisely what makes the arrangement logical: the building and the power infrastructure remain, while the systems change. But this logic requires sustained demand, a renewed competitive advantage, and customers willing to pay.

The bet isn’t that a single model will win; it’s that the need for models never stops growing.

Electricity Is Becoming the Real Currency

Gigawatts before revenue

Project Camellia is targeting 3.2 gigawatts. PORTS-Pike anticipates approximately 8 gigawatts of computing power in the long term, of which 4.25 are covered by the initial commitment and approximately 3.8 are potential additions. The first 800 megawatts are expected in 2028, primarily thanks to existing infrastructure.

After this first phase, OpenAI indicates that new power plants connected to the grid—particularly natural gas plants—as well as new transmission lines will be needed. Digital growth thus encounters a very old limitation: generating and delivering energy to the right place.

The Local Cost

Communities don’t buy tokens. They host power poles, power plants, construction sites, roads, promised jobs, and a collective risk should safeguards fail.

OpenAI and its partners promise to pay for the necessary improvements. This promise must be fulfilled without automatic cynicism or naivety. A campus can broaden the tax base and strengthen the grid. It can also lock in energy choices and shift future costs that are difficult to predict.

Every instant response begins somewhere with a slow decision about copper, gas, and concrete.

Advertised Jobs vs. Actual Jobs

35,000, then 2,500

For Ohio, OpenAI projects 35,000 construction jobs over the six-year development period through 2032 and 2,500 long-term operational jobs. These are projections, not current figures. They depend on funding, permits, infrastructure, and execution.

The contrast between the construction phase and operations is worth noting. The construction phase can require a massive workforce; the completed facility operates with far fewer people. The local economic impact therefore depends on training, contracts awarded to regional businesses, and the quality of the jobs that remain.

The Language of Profit

A projected job is not a job created. An announced fund is not a dollar disbursed. Expected tax revenue is not yet a funded school.

OpenAI has announced $40 million for a community fund in Ohio, in addition to the $40 million pledged by SB Energy. In Georgia, it is committing $80 million over the life of the project. These commitments are significant, but their impact will be measured by actual disbursements, beneficiaries, and public reports.

The significance of an announcement is judged by what remains once the cranes are gone.

Circularity is no longer a vague accusation

The seller becomes the guarantor

Nvidia sells the systems that will power the campus. Nvidia is also investing 1.5 billion in SB Energy and guaranteeing a portion of the infrastructure that will exclusively host its hardware, subject to limited exceptions. OpenAI leases the capacity and promises to reimburse Nvidia if the guarantee is called upon.

This cycle does not prove fraud or a collapse. Rather, it shows how an emerging market creates its own demand by financing the ecosystem that will buy its products. The supplier supports the customer, who supports the manufacturer, who buys the supplier’s infrastructure.

Risk may shift from one pocket to another

Capital circulates, but risk never disappears. It shifts to the balance sheet, the lease, the insurer, the taxpayer protected by regulation, or the investor willing to wait.

The provisions of the SEC filing are specifically designed to define this shift. They limit the amount, set the triggers, and provide avenues for recourse. This precision is reassuring because it reduces legal uncertainty. It is unsettling because it reveals the magnitude of the risk that must now be managed.

When an industry must guarantee its own customers, its growth also becomes its creditor.

Revenue Must Grow Faster

The Implicit Financial Model

OpenAI states that it will fund its commitments through revenue, cash flow from growth, and capital raised from investors. This is a statement of strategy, not a demonstration of sustainability. The detailed financial statements needed to make this assessment are not public.

We must therefore resist two easy certainties. There is no basis for asserting that the company will pay without difficulty. Nor is there any basis for concluding that failure is inevitable. The 750 billion projection measures an appetite; it does not provide the result.

The Subscription Model

To survive, OpenAI must convert massive computing capacity into everyday utility, and then that utility into sustained revenue, before the cost of computing exceeds the value created.

This requires more corporate clients, more paid use cases, falling unit costs, and business models that remain more attractive than those of competitors who are also capable of purchasing gigawatts. The financial gamble is therefore also a gamble on the loyalty of users who are still free to leave.

Computing is paid for in dollars; debt is repaid in human habits.

Efficiency Can Feed Hunger

More Energy-Efficient Models

A reasonable objection argues that technical advances will reduce the cost of each query, each training session, and each unit of intelligence produced. That’s possible. A better chip, optimized software, or a more efficient architecture can deliver more with less.

But efficiency does not guarantee a decrease in total consumption. If the unit cost falls while usage skyrockets, overall demand may rise. Industrial history is familiar with this paradox: making a resource less expensive can expand its market faster than it reduces its footprint.

The Price of Success

The best business scenario for OpenAI may be the worst energy scenario: models efficient enough to become ubiquitous, and useful enough that no one wants to shut them down.

That is why the discussion cannot be limited to technical efficiency. It must include volumes, peak hours, the source of electricity, load-reduction commitments, and the cost of infrastructure. A more energy-efficient query repeated a thousand times over is not an automatic victory.

Efficiency reduces the ration; success multiplies the number of guests.

Natural Gas Enters the Cloud

The Tangibility of the Compromise

OpenAI states that the full expansion of the Ohio site will require new power plants, particularly those fueled by natural gas. This detail stands in stark contrast to the intangible nature of AI. The next generation of models could depend directly on fossil-fuel infrastructure built to meet a new load.

This does not mean that all electricity will be generated by gas, nor that the final outcome is known. The project has not been fully designed. But the official possibility is enough to raise the conflict: the pace of expansion, grid reliability, and climate goals do not align automatically.

The Climate Debt

A company can pay for its power line and honor its rate commitment while leaving another bill unpaid: additional emissions and energy choices locked in for decades.

Community plans mention batteries, solar power, new generation capacity, and flexibility. They will need to publish the actual breakdown, not just the general principles. The promised annual audit of these projects must track energy consumption, peak-load reductions, and the infrastructure that is actually funded.

Paying one’s share of the grid doesn’t yet settle one’s share of the sky.

The region has the right to negotiate

Neither automatic rejection nor a red-carpet welcome

The targeted counties have reasons to welcome these projects: construction jobs, tax revenue, training, investment in the grid, and the revitalization of industrial sites. The Ohio campus would partially occupy a former uranium enrichment complex—an area already shaped by national ambitions.

They also have reasons to demand strong safeguards. The power capacity requested is immense. The projected benefits span several years. There are risks of delays, scale adjustments, or technological changes. A community is not hostile to the future simply because it demands guarantees.

Legitimate questions

Who pays if the project is halted? Who owns the new transmission lines? What happens to the site if demand drops? Which jobs will be local? How much water will actually be used?

OpenAI promises annual reports, community agreements, and independent audits. These tools can foster genuine accountability if they publish objectives, results, and variances. They will become mere marketing if they only highlight the benefits without making the costs and delays transparent.

A community is not merely a backdrop for intelligence; it is a party to the contract.

The government cannot stand idly by

Regulate Before Connecting

Public utility commissions, grid operators, environmental authorities, and local governments will determine whether the promises are kept. Companies may offer to pay. Only an enforceable rule can prevent the bill from being passed on elsewhere after a market shift.

Regulation must also withstand competition between jurisdictions. When a project promises tens of thousands of jobs, a county may be tempted to offer more than it will receive. Transparency regarding tax exemptions, network costs, and decommissioning obligations becomes essential.

A Quasi-Public Infrastructure

On this scale, the data center is no longer just a private asset. It shapes the grid, power generation, regional employment, and sometimes national energy policy.

This does not justify the government dictating business models. It does justify the government treating their infrastructure as a matter of public interest. Gigawatts reserved for one company alter the options available to other users. The choice must therefore be evaluated beyond the immediate commercial contract.

The calculation is up to the company; its consequences extend beyond its own boundaries.

750 billion conditions

What Must Be True

For the projection to become sustainable, several factors must align: revenue growth, continued access to capital, demand for AI, chip availability, permits, energy, on-time construction, stable partnerships, and genuine protection for taxpayers.

A single failure does not doom the whole endeavor. Projects are phased precisely to adjust commitments. But this list shows why the figure of 750 billion should not be celebrated as a sign of strength. It is, above all, an exceptional concentration of dependencies.

What We Already Know

We know that OpenAI wants more computing power. We know that contracts are getting longer and guarantees more sophisticated. We do not yet know whether the resulting economy will be large enough.

That is where the line of reason lies. It is possible to criticize this headlong rush without predicting a crash. It is possible to recognize the usefulness of AI without endorsing every single project. It is possible to admire the ambition while refusing to let society foot the bill.

Doubt is not a weakness when the future is already costing hundreds of billions.

Intelligence will have to learn to count

After the spectacle of numbers

750 billion is impressive because the number defies our intuition. The real test will be more modest and more severe: every phase delivered, every megawatt billed, every job created, every community promise kept, every guarantee called upon or avoided.

The future of AI will not be decided by a single press release. It will be decided by thousands of contracts and public decisions, balancing the actual utility of the tools against the material costs required to keep them within reach.

The Electricity Debt

OpenAI isn’t just building data centers. It is transforming its faith in artificial intelligence into an electrical, territorial, and financial debt that future success will have to repay.

Perhaps the revenue will come. Perhaps the models will make businesses more productive, research faster, and tools more accessible. But if the promised computing power becomes indispensable before it becomes profitable, then the industry will have created its own constraint before proving its value.

3.2 gigawatts in Georgia.

4.25 gigawatts guaranteed in Ohio.

750 billion projected by 2030.

At what point does the calculation cease to serve the future and begin to jeopardize it?

By Maxime Marquette, columnist

Columnist’s Transparency Box

Editorial Stance

I am not a journalist, but a columnist and analyst. My expertise lies in observing and analyzing the geopolitical, economic, and strategic dynamics that shape our world. My work consists of dissecting political strategies, understanding global economic trends, contextualizing the decisions of international actors, and offering analytical perspectives on the transformations that are redefining our societies.

I do not claim to possess the dispassionate objectivity of traditional journalism, which is limited to factual reporting. I strive for analytical clarity, rigorous interpretation, and a deep understanding of the complex issues that affect us all. My role is to make sense of the facts, place them within their historical and strategic context, and offer a critical analysis of events.

Methodology and Sources

This text respects the fundamental distinction between verified facts and interpretive analyses. The methodological rule is consistent: factual information is published only if it is supported by a verifiable source, and the sources actually used in this article are listed under “Sources,” never here.

Categories of primary sources used by the publication, when applicable: official press releases from governments and international institutions, public statements by political leaders, reports from intergovernmental organizations, and dispatches from recognized international news agencies.

Types of secondary sources: specialized publications, internationally recognized news media, analyses from established research institutions, and reports from sector-specific organizations.

When an article cites statistical, economic, or geopolitical data, it comes from data-producing institutions (intergovernmental organizations, central banks, national statistical institutes), and the specific institution is listed under “Sources.”

Nature of the Analysis

The analyses, interpretations, and perspectives presented in the analytical sections of this article constitute a critical and contextual synthesis based on available information, observed trends, and expert commentary cited in the sources consulted.

My role is to interpret these facts, contextualize them within the framework of contemporary geopolitical and economic dynamics, and give them coherent meaning within the broader narrative of the transformations shaping our era. These analyses reflect expertise developed through continuous observation of international affairs and an understanding of the strategic mechanisms that drive global actors.

This article describes a documented state of affairs as of its publication date, not a prediction: subsequent developments may alter these perspectives. No updates are promised in advance; when an article is corrected or supplemented, the change is dated within the text.

ANALYSIS: OpenAI, 750 billion, and the "electric debt" of intelligence