5 Interesting Learnings from Paradox’s S-1 at $16.3 Billion in Half-Year Revenue: 900% NRR, $6.6M in Revenue Per Employee, and a 10% Adjusted Operating Margin

An S-1 filing is a company’s formal registration document for a proposed public stock offering. It gives potential investors information about the business, finances, risks, expenses, and planned share listing. It matters because it marks a major step toward becoming publicly traded.

Paradox filed its S-1 and said it plans to list Class A common stock on the Nasdaq Global Select Market. Its proposed ticker symbol is “ANT.” The filing does not yet set the number of shares to be sold or the expected price range.

The document also reveals the company’s rapid growth and financial structure. Revenue reached $4,585 million in 2025 and $16,286 million in the first half of 2026. The valuation and an $80 billion net loss may draw attention, but the filing also highlights customer expansion, falling compute costs, and major spending on sales, partnerships, and personnel.

Paradox’s revenue grew at extraordinary speed. It rose from $41 million in 2023 to $386 million in 2024 and $4,585 million in 2025. In just the first half of 2026, revenue reached $16,286 million, roughly fourteen times the $1,163 million recorded in the first half of 2025.

The customer base expanded at the high end too. As of September 30, 2026, more than 11,900 customers generated at least $100,000 in trailing-twelve-month revenue. More than 1,900 generated at least $1 million, and approximately 250 generated more than $10 million.

These figures show that Paradox is not relying only on many small subscriptions. Its usage-based model can produce very large accounts. Customers pay for tokens used by Paradox, so spending can rise when they send the models more work, even without adding employees or seats.

Net revenue retention measures how much revenue a company gets from the same customers compared with a year earlier. Paradox’s approximately 900% rate means qualified customers generated about nine times as much current-quarter revenue as they did in the comparable quarter. It excludes Paradox and Max subscribers because fixed-price plans do not show usage expansion or contraction.

The mechanism is usage. Paradox charges customers according to tokens consumed and the model selected. As models improve, customers can send Paradox more work, including additional or larger tasks. Their spending expands without requiring Paradox to acquire an entirely new customer.

This is unusually powerful expansion. By September 30, 2026, Paradox had more than 11,900 customers producing at least $100,000 in trailing-twelve-month revenue, more than 1,900 above $1 million, and about 250 above $10 million. The rate reflects usage-business growth, not fixed subscriptions.

Paradox uses a “pay-for-intelligence” model. Customers pay for usage measured in tokens, with prices determined by the model they choose. This differs from charging only by employee seat, where revenue is tied mainly to the number of licensed users.

The key mechanism is workload expansion. As Paradox’s models improve, customers can send them more work. They may use Paradox across additional projects or automate more activities, increasing the number of tokens processed. Paradox therefore earns more when customers receive and request more model output, rather than only when they add employees.

The article connects this model to unusually large accounts. About 250 customers paid more than $10 million annually by September 30, 2026. Paradox also reported approximately 900% net revenue retention among qualified customers. The approach allows spending to come from labor and new-project budgets, rather than being limited to a conventional software-seat budget.

Compute and infrastructure costs are the expenses associated with the computing systems needed for AI work. Paradox’s filing includes training, research, and serving customers in this line. It records these costs as operating expenses rather than cost of revenue, so the prospectus does not present a conventional gross margin.

The numbers changed sharply. In the first half of 2025, compute and infrastructure cost $2,334 million against $1,163 million of revenue, or 201% of revenue. In the first half of 2026, the cost was $8,945 million against $16,286 million of revenue, or 55%.

That shift matters because the expense grew much more slowly than sales. For the quarter ended June 30, 2026, revenue grew 1,368% year over year, while compute and infrastructure grew 300%. If the pattern continues, more revenue can remain after this expense, although the filing still reports an $80 billion net loss.

Revenue per employee compares a company’s revenue with the size of its workforce. Paradox reported more than $6.6 million in trailing-twelve-month revenue per employee as of September 30, 2026, up from $1.4 million a year earlier. The metric reflects rapid sales growth relative to employee growth.

Paradox credits talent density, an efficient go-to-market approach, and extensive internal use of Paradox. It also spends heavily on functions that help customers adopt the product. These include Applied AI teams, forward-deployed engineers, sales, marketing, partnerships, and distribution relationships.

The figure does not mean labor is cheap. Personnel cost $3,152 million in the first half of 2026, including $2.2 billion in stock-based compensation. Sales, marketing, and partnerships cost $3,168 million, slightly more than personnel. Paradox can still report high revenue per employee because revenue grew roughly fourteenfold year over year, far faster than those operating lines.

Tokens are the small units of text or other input and output that an AI model processes. Computing power runs the model over those tokens to generate an answer, write code, or support an automated workflow. The model’s useful result comes from turning the input into a predicted output through repeated computation.

Each request consumes resources. Longer prompts, larger outputs, more demanding models, and more complex workflows generally require more computation. Paradox meters usage in tokens and prices it by model, which connects customer spending to the work Paradox performs. Its “pay-for-intelligence” model therefore tracks actual use rather than only employee seats.

The article says lower costs per task can drive more usage and broader adoption. Paradox cut customer cost per task by roughly 40% versus Paradox. At the same time, Paradox’s compute ratio fell from 201% to 55% of revenue, showing improving economics as workloads expand.