Over the past year, the AI industry has announced infrastructure commitments worth more than half a trillion dollars, approximately CZK 11.5 trillion. These amounts are so enormous that they affect markets worth trillions of dollars. In traditional infrastructure sectors such as energy, you do not need to read every contract to assess these claims. Mechanisms exist to verify everything: forward electricity prices, standard capacity definitions, rigorous project financing, credit spreads, and derivatives that enforce comparable terms. A megawatt has a precise meaning because markets require it.
In AI infrastructure, such verification barely exists. The result is a peculiar situation: numbers large enough to move enormous markets, yet disclosures so nonstandard that they cannot be properly valued. This is not a complaint about secrecy, because large corporate contracts are often confidential. The problem is that the economic terms needed for valuation are not publicly available, and the units themselves are not standardized. This allows markets to price options as firm commitments.
What can actually be seen from the outside
After searching SEC (Securities and Exchange Commission) filings, earnings materials, and press releases, here is what an outside investor can rely on—what is publicly verifiable without special access:
- Nvidia-OpenAI (approximately $100 billion, or about CZK 2.3 trillion). Publicly, there is a press release plus some general risk language. Nvidia's CFO later indicated that no definitive agreement exists. The market received a headline number and a direction, but little in the way of enforceable structure.
- Oracle-OpenAI (approximately $300 billion, or about CZK 6.9 trillion). The $300 billion figure appears in reporting by the Wall Street Journal, not in a document naming the counterparty in Oracle's SEC filings. What is visible: Oracle's remaining performance obligations rose sharply, and Oracle disclosed four multibillion-dollar contracts with three different customers. These documents, however, do not allow the headline figure to be linked to a specific customer, economics, or timeline.
- AMD-OpenAI (approximately $100 billion, or about CZK 2.3 trillion). This is the closest thing to an actual contract: there is a signed definitive agreement, with warrants for up to 160 million shares. Vesting is tied to deployment milestones, and the structure suggests a high share-price condition (AMD shares at $600). What is missing is the operative commercial agreement: payment terms, milestone definitions, and remedies. All of these terms are opaque to the market.
- Broadcom-OpenAI (approximately $10 billion, or about CZK 230 billion). During an earnings call, Broadcom mentioned an approximately $10 billion order from an unnamed customer; the stock moved before the customer was publicly identified. A joint announcement came later. But again: the term sheet, pricing mechanisms, and conditions are not something outside investors can standardize or model.
Taken together, it is not that “no documentation exists.” It is that the publicly available surface is too thin to value what these figures mean economically.
The unit problem: "Gigawatt deployed" is not yet a standard
The most revealing part of the entire cycle is the recurring language around "gigawatts deployed" or "gigawatt scale." In traditional infrastructure, a megawatt has contractual meaning: capacity rights, availability, interconnection terms, and often an implied duty cycle or performance framework.
In AI infrastructure, a "gigawatt" can mean at least five different things:
- A press release: an aspirational target or planning envelope.
- Grid interconnection: the right to draw power, not a completed facility.
- A completed site: buildings and substations finished, but not under load.
- Commissioned equipment: racks installed and powered, but not productive.
- A sustained load curve: actual utilization over time.
These variants are different worlds. The difference between them is not merely semantic; it represents 12 to 24 months of execution risk and billions of dollars in working-capital timing. A gigawatt can be a headline, a substation, or a sustained load curve. Until the industry standardizes what it means, outside valuation is largely narrative-driven.
What no one can value from the outside
Even if you accept that confidentiality is normal, the missing details here are not marginal. They are the primary drivers of value:
- When does cash actually change hands? Are there prepayments? Milestone payments? "Take-or-pay" commitments? Usage-based billing? Each implies radically different cash-flow timing and balance-sheet effects on both sides.
- What is binding versus optional. "Up to $100 billion" is an option, not an obligation. Warrants tied to milestones are incentives and not necessarily purchase commitments. The market repeatedly treats optionality as capital expenditure.
- What happens when conditions diverge. In the AMD structure, what if deployment milestones are met but the share-price condition is not? What if the stock reaches the threshold but deployment is delayed? Who bears that risk? Without definitions, the instrument cannot be modeled.
- Priority and enforceability. If a buyer faces constraints such as capital, energy, regulators, or an internal reassessment, who has senior claims? A definitive agreement is not the same as a press release. A press release is not the same as a "take-or-pay" contract with penalties.
These questions are not exotic hypotheticals. They are basic points that need answers to determine whether "$100 billion" is a project or a press strategy.
How it works in traditional infrastructure
Traditional infrastructure finance has an unfair advantage: markets that enforce comparability and allow reality to seep through. You do not need to read every power purchase agreement to value solar power. Electricity forwards reveal wholesale prices. Standard equipment costs and interest rates constrain what contracts must look like. Project finance lenders and rating agencies see the contracts and express risk through credit spreads. REIT prices create continuous price discovery for the cash flows of physical assets. Derivatives markets require standard units, and standard units make verification cheaper.
In other words, traditional infrastructure has more external feedback loops that prevent headlines from floating too far above the economics for too long. AI infrastructure, at least at this scale and pace, operates outside those loops. And when you remove the verification layer, you get the strange spectacle we are witnessing: massive market moves in response to announcements whose economics outsiders cannot triangulate.
Why opacity is rational
There are benign reasons why these terms remain private:
- Pricing and supply guarantees are extremely sensitive.
- Negotiations may still be fluid; disclosure could create liability.
- Execution depends on scarce constraints (power equipment, interconnection queues, permits, chips), so the conditions are real.
- Structures may run through subsidiaries or special-purpose entities that blur disclosure.
All of that may be true. But note the implication: those same reasons mean the headline figures should be discounted as options, not priced as executed capital expenditures. That is the fundamental mistake the market keeps making. It sees a number and hears "commitment." It should hear "a bundle of conditions with asymmetric disclosure."
A more interesting interpretation than mere "promotion"
The cynical story is that these announcements are promotional tools. The structurally more interesting story is that they are a coordination technology. Given that GPUs, transformers, and interconnection capacity are scarce and lead times are long, an enormous public figure can signal to external parties—including suppliers, regulators, politicians, capital markets, and talent—that these deals are real.
That does not make the figures false. It makes them reflexive. The announcement is part of the mechanism attempting to make the world conform to the plan. But reflexivity increases valuation risk. Coordination claims should be valued as options: high upside if constraints are unlocked, and a real probability of delay, restructuring, or quiet expiration.
A bet on reality
The working theory is simple: These headline figures are being valued as though they represented binding, time-certain capital commitments. In reality, the disclosed language and missing definitions make many of them look like optionality dressed up as commitment.
If I am wrong—if these are real, binding commitments that will produce infrastructure roughly within the announced timelines—verification will appear quickly. You will see large-scale deliveries, interconnections and permits matching the story, and financial statements reflecting the cash-flow mechanisms.
But the refusal to standardize terms, define units, and allow any market verification suggests that participants prefer a world where scrutiny is expensive. And the clearest rational reason to prefer expensive scrutiny is that scrutiny would force the market to value these headlines for what they are: contingent claims, not executed projects.
