TikTok's U.S. operation is looking to secure multiple gigawatts of computing capacity over the coming years, with one person familiar with the conversations telling CNBC that the target is as much as 4 gigawatts by the early 2030s. The figure, first reported by CNBC on October 23, has not been confirmed by the company, and the conversations are described as an early market sounding rather than a signed contract or a construction plan. Two people familiar with the discussions told CNBC the entity was exploring options to obtain the capacity over the next several years, and one source framed the need as four gigawatts by early next decade.
The entity doing the shopping is TikTok USDS Joint Venture, the Delaware-incorporated business that operates TikTok in the United States. Its ownership structure is unusual for a major consumer platform: Silver Lake, Oracle and MGX each hold 15 percent, while ByteDance retains a 19.9 percent stake in the new entity, a structure reported by CNBC and echoed across international coverage of the announcement, including Greek financial media Reporter.gr. Since the joint venture was formed, TikTok U.S. has relied on Oracle Cloud Infrastructure as its anchor provider rather than owning data centers at scale, and a company spokesperson confirmed that Oracle will remain its principal cloud partner in the U.S. The algorithm, the company has previously said, is protected within Oracle's U.S. compute environment as part of the platform's American isolation commitments.
The scale of the ask
Four gigawatts is a large number even by the standards of the current AI buildout. Independent analysis of Oracle's disclosed capacity, carried by CTOL Digital Solutions, estimates that 4 GW is roughly equivalent to nearly five quarters of Oracle's total capacity deliveries. For a company that historically leased rather than built, the ambition signals a shift in how TikTok U.S. sees its compute future: AI workloads are expected to dominate, and the platform is preparing products and operations tied to artificial intelligence, according to the CNBC report. Intelligence service Intellectia.AI summarized the aim as addressing "the increasing demands of artificial intelligence workloads," while Briefs Finance noted TikTok's flagship B2B AI products sit at "the frontier of large-model inference and creative automation."
The context is a global scramble for digital infrastructure. The same week, Nvidia executives were describing AI capacity demand as outstripping supply while operators wrestled with capital constraints. TikTok's own parent is building abroad: the AI Data Center Index lists ByteDance's 500 MW Johor, Malaysia, facility as under construction, alongside a 400 MW Thailand hub, part of a broader Asian buildout that also includes projects in Indonesia and North Dakota. In the U.S., TikTok's search places it among the hyperscalers and AI companies competing for power, land and grid connections that are increasingly the binding constraint on new capacity.
What Oracle's position means
Oracle's role matters for how the search plays out. Because Oracle's cloud business historically operates largely on leased and third-party infrastructure, securing additional capacity for TikTok means finding sites, power and construction capacity — the same pipeline every hyperscaler is chasing. CNBC said it has reached out to Oracle for comment on the reported plans. The reporting makes clear that no provider has been selected for the incremental capacity and that the company has not confirmed a figure publicly. What is firm is the direction: the operator of one of the largest U.S. consumer platforms is planning for AI-heavy, multi-gigawatt infrastructure in the United States this decade, and the market is already reacting to the procurement pressure that implies.
The wider market context helps size the ask. The AI Data Center Index tracks 346 AI data center facilities worldwide with 117.2 GW of total capacity, of which 43.6 GW sits in the United States and 163 facilities are already operational. Within that footprint, hyperscaler-scale projects in Oklahoma, Ohio and Wisconsin are measured in hundreds of megawatts each, which is what makes a single 4 GW customer program stand out: it would represent roughly 3 percent of the tracked global capacity for one platform's workloads. Power availability, grid interconnection queues and transformer lead times are now the pacing items for this buildout, and every gigawatt-scale buyer is effectively competing for the same constrained pipeline.
For B2B buyers of compute and infrastructure services, the episode is a reminder that demand forecasting at platform scale has moved from server counts to power envelopes. TikTok U.S. itself has commercialized part of its stack — its flagship B2B AI products for large-model inference and creative automation, noted by Briefs Finance, depend on the same capacity the joint venture is now trying to secure. A 4 GW commitment, if it materializes, would be among the largest single-customer capacity programs in the U.S. cloud market, and its procurement — whether through Oracle, new build, colocation or a mix — will ripple through power utilities, equipment vendors and the financing markets that have become the real bottleneck in AI infrastructure. The conversations are early, the sources anonymous, and the number unconfirmed; that is precisely why it is worth tracking.
