The New AI Conglomerates: Investors, Suppliers and Competitors

The relationships between the largest cloud providers and the leading AI labs have become the defining commercial structure of the industry, and they are difficult to describe in conventional terms.

Investor, host, reseller and rival

A cloud provider may simultaneously be a significant equity investor in a lab, the exclusive host of that lab’s models, a reseller of those models to its own customers, and a competitor building its own models. Each role is understandable individually. Together they produce a set of incentives that regulators are still trying to characterise.

• Investment gives influence over strategy without ownership or control.
• Hosting creates dependency in both directions.
• Reselling means the provider’s own product quality depends on a partner.
• Competing internally creates an obvious conflict over priorities.

A merger review that never happened

Antitrust authorities in several jurisdictions have begun examining whether these arrangements amount to de facto acquisitions that avoided merger review. The question is not whether the investment was legal but whether the market effects are the same as a merger, in which case the usual review should have applied.

Companies argue that the deals are ordinary commercial partnerships and that breaking them would slow deployment of useful technology. Regulators argue that the absence of formal control does not mean the absence of practical control.

Structural remedies, and contracts drafted for them

Whether any jurisdiction imposes structural remedies, and whether the next generation of deals is written differently as a result. Contracts are being drafted now with regulatory risk explicitly in mind, which is itself a measure of how much has changed.

Consolidation turns on control, not size

Whether an investment is consolidated depends on control rather than on size. A large minority stake with board representation may be accounted for as an equity method investment, which keeps the investee’s losses off the investor’s operating results.

The practical effect is that a strategically integrated relationship can look, in the financial statements, like a passive holding. That is not improper, and it makes the true economic exposure difficult to assess from the outside.

Unwinding would hurt both sides first

Should a partnership unwind, the consequences would be asymmetric. The lab would lose its primary infrastructure and distribution channel; the cloud provider would lose a marquee customer and face a competitor that now knows its systems intimately.

That mutual dependency is precisely what makes the arrangements unlikely to be dissolved voluntarily, and precisely why regulators are interested in them.