• China’s AI leaders compete on cost and distribution, not on matching United States spending line for line.

• Alibaba is spending harder on infrastructure, while Tencent is trying to invest without breaking investor trust.

• Better chip access helps, but monetising agentic AI remains the real test.

Cheap is a lovely word in markets until the bill arrives. This week, Tencent and Alibaba reminded investors that Chinese artificial intelligence may look like a cheaper alternative to United States names, but it is not cheap to build. Tencent unsettled investors after saying it would more than double investment in new AI products in 2026. A day later, Alibaba reported revenue growth of just 2% and a 66% drop in net income, even as cloud revenue rose 36% and Qwen passed 300 million monthly active users.

That contrast is the real hook. China’s AI race is not trying to beat Amazon, Microsoft, Alphabet and Meta yuan for dollar. It is trying to win by doing more with less: lower-cost open-source models, giant consumer platforms, faster product rollouts and, increasingly, home-grown chips.

Cheaper does not mean smaller

The big difference with China is where AI shows up first. In the United States, the clearest monetisation paths so far sit in cloud contracts, software tools and advertising upgrades. In China, the near-term opportunity looks more consumer-led and platform-led. Agentic AI, meaning software that completes tasks across apps instead of only answering questions, fits neatly into ecosystems that already handle chat, payments, shopping, travel and delivery. Tencent wants that front door to be WeChat. Alibaba wants Qwen and its new enterprise tools to become the operating layer for shopping, work and cloud usage.

That helps explain why Chinese AI can look “cheaper” than its United States peers while still being commercially interesting. These companies already own the traffic. They do not always need to persuade users to adopt a brand new habit. They can bolt AI onto habits that already exist. The catch is that low prices and wide adoption do not guarantee good profits. China’s enterprise market has been slower to spend heavily on information technology services, which makes consumer usage and token billing look more important, but also more uncertain. Tokens, put simply, are the units of AI usage that companies charge for when a model reads, writes or acts.

Alibaba brings the shovel, Tencent brings the traffic

Alibaba looks like the more obvious infrastructure bet. In February 2025, it pledged at least 380 billion yuan over three years for AI and cloud infrastructure. In its latest results, it said its T-Head chip arm now has a proprietary graphics processing unit, or GPU, in production at scale, supporting training, fine-tuning and inference while contributing meaningfully to cloud supply. That is the closest thing in China to the classic AI shovel story. Alibaba is trying to own more of the stack, from model to chip to cloud bill.

Tencent’s route is cleverer, but narrower. Its new AI product costs were 7 billion yuan in the December quarter and 18 billion yuan in 2025, and management now expects that figure to more than double in 2026. Investors did not love the sound of that, especially when it came with a slower buyback programme and limited detail on near-term returns. AI bills have a habit of arriving before AI profits, and markets are rarely famous for patience.

Still, Tencent has one advantage Alibaba cannot copy: distribution. WeChat remains one of the strongest digital gateways anywhere, and Tencent is already using AI to improve ad targeting, gaming economics and cloud services. So the spending debate is not really about whether Tencent can build useful AI. It is about whether those improvements can fund the next wave of investment quickly enough to keep investors calm. Alibaba has chosen heavier spending and clearer infrastructure ownership. Tencent is trying to spend just enough while letting its existing engine carry the weight.

The semiconductor story adds another twist. Nvidia has won approval to resume sales of H200 chips to China, and Tencent has said foreign accelerators are becoming available again. That matters because export controls had constrained 2025 spending plans. But the news did not transform the mood because the harder question is no longer simple access. It is return on investment. More chips help only if they lead to more useful services, more paying users and better margins. Otherwise they are just a more expensive electricity bill.

Chinese groups also are not waiting politely for Washington to solve their supply chain. Alibaba says its in-house GPU is now contributing to cloud infrastructure supply. Tencent says its GPU capacity should step up during 2026 and 2027. China’s broader AI ecosystem has already learned to work under tighter hardware constraints, which has pushed firms towards algorithm and hardware efficiency. That is one reason the “cheaper China AI” idea keeps resurfacing. Scarcity can be an unpleasant teacher, but it does teach.

Risks in plain sight

The risks are not hard to find. First, monetisation is still the soft spot. Tencent admits returns from new AI products will take time, while Alibaba is still relying on e-commerce cash flows to fund a large AI push. Second, the core businesses are under pressure. Alibaba’s quick commerce battle is hurting profits, and Tencent still depends heavily on games and advertising to finance future bets. Third, regulation and security could become a real brake on agentic AI just as usage explodes.

Investor playbook

• Watch revenue growth and margin together. In AI, usage without economics is only half a story.

• Compare ecosystems, not just models. Distribution often matters more than benchmark glory.

• Treat chip access as an enabler, not the end result. Monetisation remains the real finish line.

Cheap AI, costly proof

China’s AI appeal is easy to understand. The companies are cheaper than the United States mega-cap favourites, the products are moving fast, and the user bases are huge. But this week’s earnings show the real test is not who launches the cleverest agent or shouts loudest about the next model. It is who can fund the compute, protect the core business and turn AI from a subsidy into a service people will keep paying for.

Cheap AI is not the same as free AI. In China, the opportunity may lie in a market trying to build practical AI under tighter budgets and tighter constraints. That can create winners. It also sends the bill early.

Listen to the full episode now or follow the Saxo Market Call on your favorite podcast app.

Today’s Links

USD in terminal decline or is the US set to use stablecoins to dollarize the global economy? Again, have a listen to a great conversation among some analysts who I admire (but don’t necessarily 100% endorse, though I also appreciate their humility), including Michael Every, Brent Johnson (of the US Dollar Milkshake Theory) Izabella Kaminska and Whitney Baker - the last is someone I was not familiar with before listening to this conversation and she certainly holds her own.

Does the US have China over a barrel with this Hormuz Strait naval escort idea? The very busy Shanaka Anslem Perara certainly thinks so: Very interesting analysis that suggests the US is aggressively playing the energy card to shore up its leverage against China as we wind toward a possible Trump-Xi summit, with the USD as reserve currency angle in play as well - depending on whether China joins to help re-establish traffic through the Hormuz Strait with naval escort.

About those thousands of Chinese fishing boats in geometric formations… Are they there to provide cover for possible shenanigans? Craig Tindale weighs in.

Markets too complacent as Iran War continues? You know my view - the FT’s Katie Martin thinks they are too complacent, too.

Total economic war, China style A Chinese company acquired a major antimony mine in Canada back in 2009, the country’s only of its kind and a significant source of global antimony supply, only to shutter it in 2023, a year before embargoing its own exports of antimony to the US which spiked the price some nine-fold. Antimony is useful for many industrial applications but it is a mission-critical element for military ones, like hardening surfaces and for ammunition. The Bureau substack with an extensive coverage of this interesting story (you can do a Free Trial for access).

Data centers, state of play in nuclear power, Oklo and more. This interview highly recommended by the legendary “chigrl”, Tracy Shuchart, who wrote a piece on her substack throwing shade on the impact of the big strategic reserve release.

Chart of the Day - JP Morgan / S&P 500 Triptych

On today’s podcast, I mentioned a financial services ETF and its relative weakness as a possible negative harbinger of the outlook for equities as financials are often seen as a leading sector due to the credit cycle. Below I substitute that ETF with the US’ largest commercial bank JP Morgan, showing at some major negative turning points in which the broad S&P 500 was booming to new highs when the financials and JP Morgan had already turned lower and were not confirming that high. No indicator is a sure thing, but this is a loud one that demands explanation before we are to believe that, even if we do get a normalization of flows of oil and gas through the Hormuz Strait and very soon, the market is just set to turn back higher again. The huge internal rotations across equities over the last several months are another sign that something deeper and profound is going on in this market.

Questions and comments, please!