Alibaba Just Traded 75% of Its Profit for the AI Future — and the Market Punished It Anyway

Alibaba Just Traded 75% of Its Profit for the AI Future — and the Market Punished It Anyway

Let me tell you something that's been sitting with me since Thursday's numbers crossed the wire. Alibaba reported its June quarter, and the headline was brutal: net income down 75%, to 10.54 billion yuan. That's about $1.57 billion, against a Wall Street consensus that wanted 21.8 billion. The stock dropped nearly 5% premarket. Every financial headline in the world read it the same way — Alibaba's AI bet is eating the company alive.

And here's the thing that makes me want to scream at the screen: the market is punishing Alibaba for doing exactly what the American hyperscalers are doing, with exactly the same playbook, on exactly the same timeline. I've been running hosting infrastructure for over a decade, and I've watched the AI buildout from the cheap seats this whole time. So let me tell you what this earnings print actually says — because it's not the story the headlines are selling you.

The Numbers — What Alibaba Just Printed

First, the raw data, because the numbers matter more than the spin. Revenue came in at 268.95 billion yuan, up 9% year over year — actually ahead of estimates. Cloud revenue grew 45%, which the company says is a 22-quarter high. AI-related product revenue has now grown triple digits for twelve consecutive quarters, running at an annual rate of 49.5 billion yuan, roughly $7.3 billion.

Then the cost side. Capital expenditure hit 67.68 billion yuan — about $9.98 billion, nearly $10 billion in a single quarter — up 75% from the same period last year. And here's the number almost nobody is talking about: free cash flow was an outflow of 44.67 billion yuan. Not a small positive. An outflow. The company is burning through billions of yuan of cash every quarter, and it's all going into AI infrastructure.

So yes, profit down 75%. Yes, capex up 75%. Yes, the market took the stock out back. All of that is true. It's just not the whole truth.

The Two Readings — Broken Company or Buying the Future

There are two ways to read this print, and they're both sitting right there in the same earnings release. The first reading is the obvious one: Alibaba is spending money faster than it can make it, its margins are getting crushed, and the market is right to be nervous. Net income down 75% on revenue up 9% is not a healthy-looking P&L to anyone who doesn't live inside the AI bubble.

The second reading is the one the founders understand. Alibaba is doing what Microsoft, Amazon, Meta, and Google are all doing — trading today's profit for tomorrow's position. Its cloud business grew 45% in the quarter on surging AI demand. Its AI product revenue has grown triple digits for three straight years of quarters. Its adjusted cloud EBITDA margin expanded to 11.6%, up sequentially, which means the AI infrastructure it's building is starting to monetize even as it scales.

This is the exact same trade the hyperscalers are making with $700 billion of combined annual capex. The only difference is that Alibaba is doing it with Chinese economics, Chinese hardware constraints, and Chinese competition. And if you think the American version of this trade is risky, wait until you see the Chinese version.

The Part Nobody's Talking About — Open-Weight Deflation

Here's where this story separates from the Meta-and-Microsoft version. Alibaba is not just building infrastructure and renting it out at a premium. It is simultaneously giving away the crown jewels of its AI stack as open-weight models. Qwen has now open-sourced more than 460 models. The ecosystem has spawned over 300,000 derivatives. Alibaba's models have passed 3 billion downloads — the company says that puts it ahead of Meta and Google on open-model adoption. Chinese open-source models went from 1.2% of global open-model downloads in late 2024 to roughly 30% by early 2026, and DeepSeek and Qwen drove most of that shift.

Think about what that means. Every dollar Alibaba spends on AI capex has to compete against a free, downloadable version of its own best work. The same models it's renting out on its cloud are available to anyone with a GPU server and a weekend. That is simultaneously the smartest distribution strategy in the history of enterprise software and the most aggressive margin-killer the industry has ever invented. It's the reason Alibaba's AI revenue can grow triple digits while its profit collapses — volume is exploding, but every unit is priced against a free alternative.

The Secondary Bottleneck Nobody's Talking About — Every Capex Dollar Competes Against Free

This is the structural problem that the mainstream coverage is missing, and it's the one I keep coming back to. The American hyperscalers have a monetization moat: proprietary frontier models, locked-in enterprise contracts, switching costs measured in millions of dollars. When Microsoft builds a data center, it can charge cloud prices backed by Azure's integration moat. When Amazon builds one, it has AWS's install base. Even with open-weight pressure from DeepSeek and Qwen, the American giants can defend premium pricing on the enterprise layer.

Alibaba doesn't have that luxury. Its open-weight strategy means its own capex is subsidizing the deflation of the exact product it's trying to sell. The AI inference market in China is a knife fight — ByteDance's Doubao has 155 million weekly users, DeepSeek is undercutting everyone, and the entire Chinese ecosystem is competing on price per token in a way that makes the American cloud wars look polite. Alibaba is spending $10 billion a quarter to build infrastructure that will be rented out in a market where the going rate for its own models is zero dollars on the download page.

That's the real bottleneck, and it's not a hardware problem or a power problem. It's a pricing problem. The biggest Chinese hyperscaler is betting that it can out-spend the deflation its own open-source strategy creates. That bet is not going to resolve in a quarter or two. It's going to take years, and the free cash flow is going to stay ugly the whole way.

What This Means for Independent Hosting Providers

First — stop pricing your infrastructure against the hyperscalers' list price. The list price is a fiction. Qwen is free to download, and 30% of the world's open-model downloads are Chinese. If you're building a hosting or inference business, your real competitor is not the cloud's sticker price — it's the open-weight model running on hardware you already own.

Second — watch Chinese cloud pricing as a leading indicator. The American hyperscalers can hold premium pricing for a while, but open-weight deflation is a tide. Every price cut in the Chinese inference market eventually washes up on the global market. If Alibaba is renting tokens at Chinese price points, that pressure comes for your pricing too, eventually.

Third — the open-weight wave is your friend, not your enemy. The 300,000-plus Qwen derivatives and 3 billion downloads mean there's a massive ecosystem of people who need somewhere to run these models. Independent hosting providers are the natural home for that workload — the people who want the free model without the hyperscaler lock-in. That's a real market, and it's growing faster than the proprietary cloud business.

Fourth — do not confuse revenue growth with profitability. Alibaba's AI revenue is growing triple digits and its cloud is up 45%, and it's still losing billions in free cash flow. When the biggest player in China can grow like that and still burn cash, it tells you the AI infrastructure trade has a longer road to profit than anyone wants to admit. Build your capacity plans accordingly.

The Structural Reality — This Trade Runs on Both Sides of the Pacific Now

Here's the uncomfortable truth. For the last eighteen months, the "AI infrastructure is a bubble" argument has been mostly a one-sided story — American hyperscalers spending money they don't have on infrastructure the market keeps saying it wants. Thursday's Alibaba print kills that framing. The Chinese hyperscalers are now making the identical trade, with the identical structure, and the identical negative free cash flow. The world's two biggest AI markets are both betting that the future pays for the present.

That doesn't make the trade right. It makes it global. When the American buildout hits a speed bump, there's no longer a Chinese counterweight to point at — the Chinese players are levered to the same thesis, with thinner margins, harder hardware constraints, and an open-weight strategy that compounds the pressure. Whatever the AI capex cycle looks like from here, it now looks the same in both hemispheres.

The Bottom Line

Alibaba didn't break itself on Thursday. It showed the world what the AI trade actually costs when you're not Microsoft. Profit down 75%, capex up 75%, free cash flow negative — and cloud revenue up 45% with AI growing triple digits for the twelfth straight quarter. The market punished the P&L and ignored the trajectory, which is what markets do when they're scared.

But here's what I'd tell you if we were sitting at the bar: don't be scared of the headline, and don't be naive about the strategy. Alibaba is building the future and giving away the present, all at once. The question every hosting provider should be asking isn't whether Alibaba is broken. It's whether you can make money in a market where the biggest competitor's best model costs nothing to download. That's the trade. And it's coming for all of us.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: Bloomberg, Reuters, CNBC, AP, TechNode, Wall Street Journal, Fortune, Alibaba Group earnings release, Quartz, Blockonomi.