Paradox S-1: an $80.7B net loss on revenue of $16.3B in H1 2026, up from a $13B net loss on revenue of $1.2B in H1 2025, and a net revenue retention of 900% (Bailey Lipschultz/Bloomberg)

Paradox’s first-half 2026 figures show explosive growth alongside an enormous loss. The company reported $16.3 billion in revenue but a net loss of $80.7 billion. Revenue measures sales earned during the period, while net loss includes costs and other expenses deducted from those sales.

The comparison with H1 2025 is dramatic. Revenue rose from $1.2 billion to $16.3 billion. The net loss also widened sharply, from $13 billion to $80.7 billion. Thus, sales grew by far more than ten times, while the reported loss increased by more than six times.

These figures describe a company scaling rapidly but spending even faster. The supplied article does not provide a detailed expense breakdown, so it does not identify which costs produced the loss. It does show the central tension facing Paradox as it moves toward an IPO: exceptional commercial momentum paired with very large negative earnings.

An S-1 filing is a registration statement submitted to the U.S. Securities and Exchange Commission before a company offers shares to the public. It typically describes the business, financial results, risks, use of proceeds, and ownership structure. The document helps potential investors evaluate what they might buy.

Submitting an S-1 does not complete an IPO. It begins a review and disclosure process. Regulators can ask questions, and the company may amend the filing. Underwriters and executives also prepare for marketing the offering, including discussions about valuation and investor demand. The company ultimately decides whether and when to proceed.

For Paradox, the filing matters because it moves the company closer to becoming publicly traded. Its reported figures, 900% net revenue retention, and warning about existential risks would become part of the information investors examine. The supplied article does not state a listing date, share price, exchange, or final IPO decision.

Revenue and net loss measure different sides of a company’s finances. Revenue records sales during a period. Net loss subtracts operating expenses, investment in development, financing costs, taxes, and other accounting charges. If those costs exceed sales, the company reports a loss, even when revenue is growing quickly.

For example, a company could collect $16.3 billion from customers but spend $97 billion on computing capacity, employees, research, facilities, and other expenses. It would then report a $80.7 billion net loss, before considering the precise accounting treatment of each item. That example illustrates the mechanism, not Paradox’s actual expense breakdown.

The article provides the two totals but does not explain the specific costs behind them. The key reality is that Paradox’s sales expanded from $1.2 billion in H1 2025 to $16.3 billion in H1 2026, while spending or other losses grew faster. Its IPO filing puts that growth-versus-profitability gap before prospective investors.

Net revenue retention, or NRR, tracks how much revenue a company keeps and expands from an existing customer group over time. A 900% rate means that group produces 900% of its starting revenue. In plain terms, $1 of earlier revenue becomes $9, before counting new customers.

The key mechanism is expansion within the original customer base. Customers may buy more usage, adopt additional products, move to larger plans, or increase their spending as their needs grow. For example, if a group paid $10 million during one period, 900% retention would mean $90 million from that same group in the comparison period, after accounting for cancellations and downgrades.

The article reports Paradox’s 900% figure but does not explain its measurement period or customer mix. Such a rate signals unusually strong expansion among existing customers, yet it does not guarantee profitability. Paradox still reported an $80.7 billion net loss in H1 2026, despite revenue reaching $16.3 billion.

An AI company’s sales can grow rapidly while its cost base grows faster. Training advanced models requires large computing systems, specialized chips, data, engineers, and testing. Running those models for customers also consumes computing capacity, electricity, storage, and network resources. These costs can make each new customer expensive to serve.

Other possible expenses include research and development, employee compensation, offices and data centers, equipment depreciation, sales, legal work, security, and safety programs. A company might also record financing or investment-related charges. For example, doubling customer usage can raise revenue while requiring more servers and power, limiting or reversing the profit from that usage.

The supplied article gives no itemized explanation for Paradox’s $80.7 billion loss. These are established examples of costs that can affect an AI company, not confirmed Paradox line items. The reported pattern remains clear: revenue rose to $16.3 billion in H1 2026, but the company’s total costs and other charges were much larger.

An existential-risk warning tells prospective investors that the technology could, in the company’s assessment, create consequences threatening humanity’s long-term future. Including that warning in an IPO filing makes the risk part of the formal information investors receive when evaluating Paradox and its shares.

The practical effect is greater scrutiny. Investors may examine Paradox’s safety controls, research decisions, governance, and plans for handling dangerous capabilities. Lawyers, regulators, business partners, and the public may also focus more closely on the company’s disclosures and responsibility for its systems. The warning can affect how people assess uncertainty and potential liability.

The article does not describe a specific incident, safeguard, regulatory response, or investor reaction. Therefore, the warning alone does not establish that existential harm has happened. It shows that Paradox moved closer to an IPO while acknowledging that its technology carries risks potentially far beyond ordinary business losses.

Generative AI models are systems that learn patterns from large collections of examples and then produce new content. Depending on the model, that content can include text, images, audio, video, or code. They do not simply retrieve one stored answer; they calculate likely outputs from learned relationships in the data.

Training adjusts a model’s many internal parameters through repeated calculations across enormous datasets. That process requires powerful chips, substantial memory, and long periods of computation. Once trained, the model still needs computing power each time it responds to a user. More users, longer requests, and more complex outputs increase the workload.

Electricity is needed to run the computing equipment and cooling systems, while data supports the learning process and ongoing testing. These demands help explain why AI companies may face large infrastructure and operating costs. The article does not detail Paradox’s systems, but its reported revenue, loss, and warning concern a company developing this kind of technology.