China faces $105B AI investment gap

AI infrastructure is the story here — not chatbots, not demos, but the boring expensive stuff underneath. Data centers, chips, networking gear, power contracts, and the cash needed to keep buying all of it. The new wrinkle is that the US and China are not just competing on model quality anymore. They are building on very different financial scales, and that gap is starting to look big enough to shape who can train, deploy, and monetize the next wave of AI. ### Where does the $105 billion number come from? The rough comparison making the rounds is simple: the biggest US tech companies are on pace to spend about $700 billion on AI-related capital expenditures in 2026, while China’s comparable total is closer to $105 billion. That framing showed up in fresh coverage around Chinese tech spending pressure, and it lines up with broader Wall Street work arguing that US hyperscaler capex has reached a historically extreme level. ### Why is capex the metric that matters? Because AI is constrained by physical bottlenecks. If you want bigger models, faster inference, or cheaper enterprise products, you need more compute and more power. Capex is basically the scoreboard for who can afford to keep expanding that base layer. Revenue matters later. Right now the race is still heavily about who can finance the buildout without blinking. ### So is China actually underinvesting? Not exactly. China is still spending heavily, and some analysts expect Chinese internet and AI firms to keep raising outlays as demand for AI applications grows. The issue is relative scale. If one side can put up six or seven dollars for every dollar the other side spends, small gaps in chips, power access, and model training capacity can compound fast, with tighter prioritization and more focus on efficient deployment. ### Why does the Manus deal matter so much? Because it suggests the constraint is not only money. Reports this week said China blocked or forced the unwinding of Meta’s reported $2 billion acquisition of Manus, a Chinese AI startup. If that account is right, Beijing is signaling that advanced domestic AI assets are strategic enough to keep under tighter national control. That is a very different message from “the best bidder wins.” ### Is that just one deal? Maybe — but one deal can still be a tell. The important part is the pattern it hints at. Washington has already tightened export controls around advanced chips and AI technology flows. If Beijing is also getting more restrictive about outbound ownership or foreign takeovers of sensitive AI companies, then both sides are hardening the edges of their ecosystems at the same time. ### What does that do to the market? It pushes AI toward fragmentation. The old assumption was that capital, talent, software, and acquisitions would keep crossing borders even if politics got noisy. The catch is that AI now sits too close to national security, industrial policy, and strategic infrastructure for that assumption to hold cleanly. Companies may need separate products, separate cloud footprints, and separate ownership structures for different jurisdictions. ### Does more spending automatically mean the US wins? No. Big spending can also mean waste. Investors are already asking whether the return on all this AI infrastructure will justify the burn. But scale still buys time, optionality, and room for mistakes. A company that overspends with a huge balance sheet can survive bad turns. A company that underspends in a compute race may never get the chance to catch up. ### Bottom line This is turning into a two-front AI story: America has the bigger wallet, and China looks increasingly willing to police where its most valuable AI assets can go. Put those together and you get a less global, more jurisdiction-aware AI industry — one where the winners are not just the best model builders, but the companies best positioned inside their own national stack.