Three items logged on the Big Tech beat this week look unrelated: an $8 million seed round for satellite AI software, a White House plan that leans on voluntary safety tests from AI firms, and Google's struggling effort to pay websites for AI answers. Read together, they show one pattern. The companies closest to the model keep the revenue and the regulatory standing, while partners, contractors, and publishers absorb the cost and the risk of the AI buildout. In the United States, that asymmetry is now the defining feature of how Big Tech does business.
The Satellite Deal Shows Who Carries the Capital Risk
TechCrunch reported that Satlyt Inc., a startup founded in 2024 that develops artificial intelligence software for satellites, raised $8 million in a seed round led by early-stage fund Non Sibi Ventures. Its chief executive, Rama Afullo, was previously a product manager at SpaceX, according to the same report. On its own, that is a small early-stage financing. In the context of the Big Tech beat, it is a reminder of where AI capability is actually being assembled. Satlyt is not a platform giant; it is a supplier building intelligence for orbital infrastructure. The model may run on servers owned by a handful of American cloud companies, but the software that makes it useful in space is being built by a young firm with seed capital and a founder with a resume from the launch industry. If the model works, the upside flows toward the platforms that host and distribute it. If it does not, the loss lands first on Satlyt, its founders, and its early backers. That is the standard arrangement in the current AI cycle: the largest American technology companies get optionality, and the smaller firms take the balance-sheet risk.
Voluntary Safety Tests Are a Transfer of Responsibility
Ars Technica reported that the Trump plan to combat AI risks hinges on Big Tech firms policing themselves, with dozens of AI companies agreeing to voluntary safety tests. The word doing the work in that sentence is "voluntary." A voluntary regime does not impose penalties for failing a test, does not create an independent enforcement body, and does not produce a public record that consumers or investors can rely on. It shifts the burden of proof onto the firms themselves and, by extension, onto the public that has to trust the results. For US technology companies, this is a favorable arrangement in the short run. It avoids compliance costs and preserves the speed that has driven the American AI market. It also concentrates risk. If a voluntary test misses a serious failure, the consequences fall on users, on downstream businesses that built on the model, and on the credibility of the entire US AI sector. The largest firms have the legal and communications resources to manage that outcome. Smaller suppliers and the customers who depend on them do not.
Google's AI Payments Show the Same Drift
Ars Technica also reported that Google's early attempt to pay websites for AI answers is struggling, with many sites seeing just one-tenth of one percent of their advertising revenue from AI payments. That is not a rounding error; it is a signal about who captures value when an AI system answers a question that used to send a user to a publisher. The publisher supplies the underlying content, the AI system packages the answer, and the platform keeps the relationship with the user. Payments of one-tenth of one percent of prior advertising revenue, as reported, suggest that the compensation being offered is not close to replacing what the old arrangement provided. For US consumers, the near-term effect is convenient answers. The longer-term effect is a publishing base with less money to do original work, which eventually reduces the quality of the information the AI systems themselves depend on. For US technology companies, this is a familiar pattern: internalize the value, externalize the cost.
