Eddie Wu: Alibaba trains next-gen AI on its own chips after earnings beat
Alibaba's CEO used a better-than-expected quarter to call homegrown silicon the foundation of its AI stack, a pivot that reshapes its cost base and its leverage over suppliers.

Alibaba Group Holding chief executive Eddie Wu Yongming said the company is training its next-generation AI models on in-house, domestically developed chips, a disclosure he made alongside better-than-expected revenue and profit. For decision-makers, the move signals that AI infrastructure is shifting from a procurement line item to a strategic asset, with cost, supply security, and competitive position all riding on who owns the silicon.
Alibaba Group Holding is training its next-generation AI models on chips it designed in-house, chief executive Eddie Wu Yongming said Thursday, delivering the disclosure alongside a quarter that beat expectations on both revenue and profit. "Infrastructure powered by our own domestically developed chips is becoming the foundation of Alibaba's full-stack AI capabilities," Wu told analysts on the company's earnings call. Read that sentence twice. The news is not that Alibaba has a chip program. The news is that the program has graduated from a research line item to the stated foundation of the company's AI strategy, announced on the same call where management had good numbers to show.
That word "foundation" is doing real work. A full-stack AI pitch means owning every layer: the silicon, the cloud infrastructure that runs it, the models trained on top, and the applications sold to customers. Historically, most cloud and AI companies rented the bottom layer from merchant chip suppliers and competed on the layers above. Owning the bottom layer changes three things at once: cost per training run, exposure to supply shocks, and how much of the value chain the company keeps for itself.
The timing is not accidental. Wu's comments landed on the back of better-than-expected revenue and profit, and that matters because chip development is one of the few corporate programs that behaves like a mortgage: expensive up front, paid off over years, and brutally hard to walk away from once you have started. A beat gives management the political cover to keep funding a program whose returns show up in gross margin long after the quarter that funded it is forgotten. Weak numbers invite questions about discipline. Strong numbers buy patience.
For context, the strategic logic here is older than the current AI boom. Washington's export restrictions in recent years have limited Chinese companies' access to the most advanced AI accelerators, which turned domestic chip development from a cost-optimization exercise into a supply-chain insurance policy. Any Chinese company whose AI roadmap depends entirely on imported silicon is, in effect, renting its future from a counterparty that can change the terms. Wu's framing, domestically developed chips as the foundation, is the language of a company trying to stop renting.
What Wu did not say matters just as much. The disclosure as reported does not come with performance benchmarks, manufacturing details, or volume commitments, and those are the numbers that separate a genuine substitution from a hedge. Four gates decide that: performance per watt against the leading merchant accelerators, memory bandwidth, the maturity of the software toolchain that developers actually have to use, and whether the chips can be manufactured at the volumes Alibaba's training runs demand. Clear all four and this is a structural shift. Clear two and it is a useful fallback for less demanding workloads.
The second-order effects run in both directions. If in-house silicon carries a meaningful share of Alibaba's training load, the company's capex profile changes shape, its dependence on external suppliers falls, and it gains room to price cloud services more aggressively, because the largest input cost is now partly internal. That is a competitive problem for everyone selling into the same market. Cloud and AI peers across China have historically pursued their own in-house silicon efforts for exactly these reasons, and a credible deployment at Alibaba's scale would convert that from a differentiator into table stakes.
There is a real risk on the other side of the ledger. Vertical integration is only a moat if your internal layer keeps pace with the frontier. Lock into a chip stack that trails the best available silicon by a meaningful margin and you have not saved money, you have capped your own ceiling, with switching costs high enough to make the mistake durable. That is the tension every board weighing a build-versus-buy decision on AI infrastructure now has to sit with: control versus capability, and the fact that you usually cannot maximize both.
For executives outside Alibaba, the practical read is that the AI cost curve is becoming a strategic variable rather than a procurement detail. The companies that win the next few years are likely to be the ones that know, workload by workload, what they are paying for intelligence and which parts of that stack they could own. Wu's line on Thursday was a signal about where the value is moving: away from renting intelligence and toward owning the machinery that produces it.
This story's Key Insights and Take-aways are locked.
Create a free account to unlock Executive Actions for one credit.
Register to UnlockAlways free for Executives Club members. Join the Club
More in Business
S&P 500 Enters Correction, Down 10% From Its High on AI Worries
The index's slide past the 10 percent line is a symbolic verdict on investor mood, and the AI trade that powered the rally is now the thing testing it.
Dow and S&P 500 enter correction as Nasdaq logs worst week since 2020
All 11 S&P 500 sectors fell more than 1% in the same week for the first time since April 2025, a breadth signal that matters more than any single index level.
Chip stocks lose $2.3 trillion in a week, worst run in 25 years
A $2.3 trillion weekly wipeout in semiconductors is forcing every board with an AI story to ask whether it owns a thesis or just a crowded trade.



