ByteDance’s Possible $70 Billion AI Capex Plan Signals China’s New Compute Arms Race

ByteDance is already one of China’s most aggressive AI spenders. Earlier this month, SCMP reported that the company had lifted its planned 2026 capital expenditure to more than 200 billion yuan, or about $30 billion, up at least 25% from a preliminary plan of 160 billion yuan. The report said the increase was driven by ByteDance’s deeper commitment to AI and rising memory-chip costs, while a larger share of the budget was being directed toward domestic AI chips to reduce geopolitical risk.

A potential move toward as much as $70 billion would put ByteDance closer to the spending scale of the world’s biggest AI hyperscalers. U.S. tech giants are still far ahead: SCMP reported that the largest U.S. tech companies are on track for more than $700 billion in AI capital expenditure this year, while Chinese cloud service providers are expected to spend around $105 billion. That gap explains why Chinese firms such as ByteDance, Alibaba, Tencent, and Baidu are under pressure to increase infrastructure spending even as they face export controls, chip scarcity, and pressure to use more domestic semiconductors.

ByteDance’s urgency is not abstract. AI is core to its business model, from TikTok and Douyin recommendation engines to advertising efficiency, content moderation, generative AI products, and cloud services through Volcano Engine. The company also appears to be looking beyond standard GPU purchasing. Reuters, citing Bloomberg, reported that Qualcomm reached a deal to supply ByteDance with millions of AI data-center chips known as application-specific integrated circuits, or ASICs, to support ByteDance’s AI agent software. That matters because custom chips could help ByteDance reduce dependence on Nvidia while building hardware more closely matched to its own AI workloads.

The chip strategy is also shaped by geopolitical friction. Reuters previously reported that ByteDance planned to spend about 100 billion yuan, or roughly $14.29 billion, on Nvidia AI chips in 2026 if Nvidia were allowed to sell H200 GPUs in China. But U.S. licensing rules, Chinese policy preferences, and Beijing’s push for domestic chip adoption make that path uncertain. This is why ByteDance’s AI spending should be read as both a technology investment and a supply-chain defense strategy: the company needs more compute, but it also needs more control over where that compute comes from.

For markets, the takeaway is that AI capex is becoming a balance-sheet arms race. Investors usually focus on revenue growth from AI products, but the real question is whether companies can turn huge infrastructure spending into durable margins. Reuters noted that Big Tech investors are already questioning whether hundreds of billions in AI spending can generate enough cloud and advertising growth to justify the cost. ByteDance faces the same problem, with an added layer of geopolitical risk. If the company successfully converts heavy AI capex into better recommendation systems, stronger ad monetization, cheaper inference, and competitive AI agents, the spending could strengthen its long-term moat. If not, it could become another example of the AI boom’s harshest truth: compute is expensive, and not every company spending big will earn the right returns.