TikTok's US Joint Venture Is Hunting Four Gigawatts of Computing Power by the Early 2030s

TikTok's US operation is quietly entering the global race for artificial intelligence infrastructure. The TikTok USDS Joint Venture, the entity that operates the short-video platform in the United States, has been sounding out opportunities to secure multiple gigawatts of computing capacity over the coming years, two people familiar with the discussions told CNBC on October 23. One of the sources said the company is looking for about four gigawatts of compute by the early 2030s — roughly the output of four large nuclear reactors.

The company did not confirm a target, but a spokesperson acknowledged the direction of travel: "TikTok runs primarily on Oracle Cloud Infrastructure in the US and Oracle will continue to be our primary cloud infrastructure provider." Oracle's cloud typically runs in facilities the database giant leases, and CNBC reported it had approached Oracle for comment on the capacity talks. No final agreements have been announced.

Why a video app needs gigawatts of compute comes down to two workstreams. First, TikTok US has been localizing its core machine learning: in January the company said it would retrain, test and update its recommendation algorithm on US user data, with the system secured inside Oracle's US cloud environment. Second, the company is hiring aggressively for AI — its careers site lists numerous machine learning roles, including one posting that describes work on "a flagship B2B AI product at the frontier of large-model inference and creative automation."

The structure behind the effort is new. TikTok USDS Joint Venture was established in 2026 under a US executive order requiring the divestiture of TikTok's American business from its Chinese parent company, ByteDance. The venture has three managing investors — Silver Lake, Oracle and MGX — each holding 15 percent, with Michael Dell's Dell Family Office also on the register. ByteDance retained 19.9 percent of the new entity, which now operates the platform for roughly 170 million US users under its own governance.

The capacity hunt puts TikTok US in the same line as the hyperscalers it depends on. Amazon's Andy Jassy said this week that AWS capacity for 2027 is "largely sold out" with a good chunk of 2028 committed; Meta, Microsoft and OpenAI have all signed multibillion-dollar compute deals; and even sovereign-backed ventures are queuing for power and silicon. For a company whose product is powered by one of the most compute-hungry recommendation systems in consumer software, securing long-dated capacity has become table stakes rather than a differentiator.

It is also a bet on the joint venture's independence. Buying four gigawatts is a decade-scale commitment — the kind of capital planning that only makes sense if TikTok's US entity expects to operate, and to keep training its own models on US soil, well into the 2030s. The details remain anonymous-sourced and unconfirmed, but the direction is clear: the most-downloaded app in America is about to become a serious customer in the market for gigawatt-scale AI compute.