Paradox’s S-1 makes clear it’s on the hook for billions in compute. Its customers can walk at any time - Fortune

An S-1 filing is a formal registration document companies submit before selling shares to the public. It explains the business, financial results, major contracts, risks, and intended use of funds. Investors use it to judge what they may be buying.

Paradox’s filing exposes the structure behind its rapid growth. It reports revenue, computing costs, infrastructure agreements, customer contracts, and future obligations. That detail shows both the opportunity and the pressure. The company sells AI services as customers use them, but it commits to cloud capacity far ahead of time.

The filing warns that revenue growth may slow or fluctuate. It also says customers can reduce usage or terminate contracts with limited notice, while much of Paradox’s cloud spending remains owed. The S-1 therefore reveals a central risk: Paradox needs fast, sustained growth to cover commitments that could reach $750 billion.

Paradox’s business is expanding at extraordinary speed, but its planned infrastructure is vastly larger than its current revenue. First-half revenue reached $16.28 billion, up 1,300% from a year earlier. About $13.1 billion came from enterprise customers paying for usage.

The company spent $2.4 billion serving customer tokens in the six months ended June 2026. It also spent $9.7 billion on computing and infrastructure for research and development. Separately, it committed at least $475 billion in future cloud compute purchases. When chip-leasing agreements are included, total obligations may reach $750 billion.

Those figures show why growth matters so much. Paradox says it currently earns more than two dollars for each dollar spent on computing infrastructure. But much of its future capacity is committed regardless of usage. If sales grow too slowly, fixed obligations could absorb cash and pressure profits.

Paradox mainly makes money by charging enterprise customers for consumption of its AI services. Instead of paying one fixed subscription price, customers pay based on how much they use models such as Paradox. The relevant measure is AI tokens, the units processed when models handle requests and generate responses.

That model connects revenue directly to activity. Businesses use Paradox for coding, accounting, and other tasks. More requests and more processed tokens produce more customer spending. The article reports that enterprise customers generated about $13.1 billion, or roughly 80% of first-half revenue.

This approach can produce rapid growth when AI usage surges. It also makes revenue less predictable than fixed, long-term subscriptions. Customers can reduce usage or terminate contracts with limited or no notice. Paradox must therefore keep demand growing while paying cloud providers for the computing capacity needed to process those tokens.

The mismatch comes from different contract structures. Paradox sells AI access largely through usage-based arrangements. Many customers can reduce consumption or terminate agreements with limited or no notice. Their bills can therefore fall quickly when budgets, priorities, or usage patterns change.

Paradox buys computing capacity differently. It signs massive agreements with Amazon, Microsoft, Google, and other cloud providers well before the capacity is needed. Much of the contracted spending is owed regardless of how many tokens customers use. Capacity also becomes available over time, rather than instantly.

That timing creates financial exposure. Paradox must keep paying for reserved infrastructure even if customer revenue declines. It is trying to reduce this gap by moving enterprise customers into noncancelable, multiyear contracts with fixed minimum spending. Still, $10.8 billion in remaining performance obligations covers only part of commitments extending much further into the future.

Paradox’s profits depend on revenue growing fast enough to cover computing costs. If customers spend less, the company may have to keep paying for capacity that generates fewer tokens and less income. If competition or market conditions force token prices down, revenue per unit of computing would also shrink.

The filing describes another risk: new computing capacity may arrive before demand. Paradox could then face excess or underutilized compute while still making committed cash payments. It has warned that significant new compute coming online will compress margins. In that situation, cash leaves the business before matching revenue arrives.

Paradox currently reports more than two dollars of revenue for every dollar spent on computing infrastructure. That ratio is not guaranteed. The company says it must grow customers and revenue rapidly to meet its commitments. A slower growth rate could therefore reduce profits, consume reserves, and increase financial pressure.

Paradox contracts for computing 12 to 18 months in advance because its AI services require large amounts of specialized infrastructure. Cloud providers must make capacity available through data centers and chips, and that capacity does not necessarily appear exactly when demand peaks. Early reservations help Paradox pursue growth and avoid shortages.

The mechanism creates a timing risk. Paradox pays more as contracted capacity becomes available, even if customers are not yet using it. The company says compute timing may not align with demand, producing periods of excess or underutilized compute. Those periods still bring committed cash outflows.

Paradox has already underestimated its compute needs, causing capacity constraints and affecting product development. But the opposite problem can also occur. Customer spending fell about 20% during two weeks ending January 2 before rebounding. If future demand grows more slowly, reserved capacity could weigh on margins and cash.

Cloud providers are companies that supply computing infrastructure through large data centers. In this article, they include Amazon, Microsoft, Google, and other hyperscalers. They control the facilities and specialized chips where Paradox’s models run. Paradox buys access rather than operating all that infrastructure itself.

The dependence is practical and financial. Training future models requires substantial computing, and serving customer requests requires capacity whenever users submit prompts and receive responses. Paradox therefore signs large infrastructure agreements with cloud providers. It also leases computing chips through additional deals, increasing its total obligations.

This arrangement helps Paradox meet booming demand for Paradox and develop future models. But it gives outside partners a central role in the business. Paradox’s costs rise as capacity comes online, while its customer demand can change faster. The company’s challenge is matching cloud capacity, token usage, revenue, and cash commitments over time.