Alibaba's February 2025 commitment of RMB380 billion over three years [6] averages RMB126.7 billion a year, and the company spent RMB67.7 billion in the June 2026 quarter alone [8], which annualises to RMB270.8 billion, about 2.1 times the pace it announced eighteen months earlier [1]. Multi-year programme numbers out of Chinese tech have become floors, with the original budgets already exceeded.
Aggregate the three listed names for that quarter and you get RMB131.9 billion [8][9][10], which at the 6.67 rate implied by Moody's own conversion of Alibaba's programme into $57 billion [6] is $19.8 billion in three months, or $79 billion annualised; add the up-to-$70 billion ByteDance is reported to be planning for 2026 [7] and the arithmetic lands at $149 billion against Moody's $140 billion forecast for the year [2]. The forecast is a run rate rather than a stretch, and the ByteDance line is the least verifiable figure inside it.
Moody's expects the spending difference to produce a widening gap in installed computing capacity [11], and on the IEA figures it cites the absolute gap does widen, from 24 gigawatts at the end of 2025 to 33 GW in 2030 [4]. Set the money against the physical build, though, and the proportions sit awkwardly: the US is spending 5.6 times China's 2026 budget [6] while adding 48 GW by 2030 against China's 39 GW, a difference of 1.23 times [7]. That comparison is crude, one year of capex against five years of capacity, and it leaves out the thing that decides the outcome, which is what each gigawatt is computing with.
This is probably wrong, but the cash-flow half of the report is the softer half. Moody's expects AI investment to weaken free cash flow and push leverage higher on both sides of the Pacific, with several companies turning free-cash-flow negative in 2026 and 2027 [15], which makes thin cash generation a shared condition of differing depth rather than a China-specific brake. The brake the agency ranks first is hardware, since restrictions on leading-edge Nvidia chips have pushed Chinese hyperscalers and state-linked operators towards domestic parts that still trail Nvidia's technology and ecosystem [13][14]; the structural point underneath, that Chinese hyperscalers are building from a materially smaller revenue base while US peers fund the build out of profitable non-cloud businesses [19], is the harder one to argue with. The counter-thesis is the same fact read the other way, or rather the more interesting version of it: if domestic accelerators deliver materially less usable compute per watt, a Chinese gigawatt is worth less than a US gigawatt, and the dollar gap is the better proxy after all.
This question turns on two figures. ByteDance's $70 billion is a Bloomberg report cited by Moody's rather than a filing [7], and if the outlay comes in at $40 billion the 2026 Chinese total misses by 21% [9]. And if the US hyperscalers reach the near-$1 trillion 2027 figure [3] while China holds at $165 billion [1], the US increment alone, roughly $215 billion, exceeds China's entire 2027 budget by about $50 billion [8], and the cash-flow reading wins without needing the chip argument at all.
