Alibaba is trying to change the language investors use to understand its business.
For years, GMV—gross merchandise volume—was one of the defining numbers associated with Alibaba's e-commerce empire. It measured the value of transactions flowing through its marketplaces and helped investors understand the scale of China's digital consumer economy.
Now CEO Eddie Wu is pushing investors toward a very different concept: “Machine Thinking.”
At Alibaba's 2026 Apsara Conference in Hangzhou, Wu argued that the world is entering an era in which machine-generated intelligence becomes a scalable commodity, analogous to how industrialization turned physical power into something that could be produced and consumed at enormous scale. Alibaba's own framing says machines could eventually perform more than 1,000 times the total amount of “Thinking” performed by humanity.
It is an ambitious shift.
But there is an important distinction between measuring the amount of AI activity taking place and proving that the activity is economically valuable.
And that is where Alibaba's new metric becomes particularly interesting.
From GMV to machine intelligence
Alibaba built its global identity around commerce.
GMV answered a relatively intuitive question:
How much merchandise is moving through the platform?
The new AI-era question is radically different:
How much machine computation and cognitive work is being generated?
Wu describes this future using the concept of “Machine Thinking.”
Alibaba's official presentation argues that human cognitive capacity has historically been constrained by two factors: the limited number of humans capable of performing complex intellectual work and the economic cost of allocating that human talent.
AI changes both constraints.
Machines can theoretically run continuously, replicate expertise and execute enormous numbers of tasks simultaneously.
Alibaba therefore sees AI not simply as software that assists humans, but as a new industrial resource.
Just as electricity became a generalized source of physical power, Alibaba wants machine-generated intelligence to become a generalized source of cognitive power.
That is the conceptual foundation behind the company's new language.
The “machine thinking” economy
Wu made a striking comparison.
During the Industrial Revolution, machines initially imitated human and animal labor.
Steam engines pumped water.
Machines replaced physical labor.
The term horsepower itself emerged because early engines were understood in relation to the work horses could perform.
Eventually, however, machines stopped merely copying horses.
They enabled things horses could never do:
- airplanes;
- industrial factories;
- modern transportation;
- massive infrastructure;
- space exploration.
Alibaba believes AI is approaching a similar transition.
Today's AI applications largely imitate human cognitive tasks:
- writing;
- coding;
- summarizing;
- translating;
- searching;
- customer service;
- analysis.
But Wu argues that this represents only the beginning.
The next stage involves millions of AI agents working simultaneously on complex problems.
His example is deliberately extreme: give an AI system the objective of building a spacecraft capable of reaching Mars, and potentially millions of agents could divide the problem into millions of subtasks and work continuously toward the objective.
That is the economic idea behind machine-thinking volume.
The amount of machine cognition performed could become vastly larger than the amount of human cognition.
Alibaba says machine thinking is still tiny
Wu's most provocative number was that today's total volume of machine thinking is less than 3% of human thinking, according to Alibaba's own conceptual estimate.
His argument is that even if machine thinking eventually reaches 1,000 times human capacity, the world would still be at the beginning of the expansion curve.
This is not comparable to a conventional financial metric such as revenue.
It is closer to an industrial-capacity thesis.
Think of it this way:
GMV measures economic activity.
Machine-thinking volume attempts to measure computational/intellectual activity.
The latter could grow enormously without necessarily producing proportional economic returns.
And that distinction is critical for investors.
The problem: thinking volume isn't revenue
This is where the skepticism around Alibaba's new framing becomes justified.
Suppose an AI agent performs 10 million reasoning steps.
That tells us something about compute utilization.
It does not automatically tell us:
- how much customers paid;
- whether customers found the output useful;
- whether the AI generated profit;
- whether the task replaced a human worker;
- whether the customer will continue using the service;
- whether Alibaba's infrastructure costs exceeded revenue.
A company could theoretically generate enormous machine-thinking volume while losing money on every transaction.
That is why the metric should not be confused with a revenue model.
Alibaba itself has provided evidence that AI commercialization is already occurring. In its August 2026 earnings commentary, the company said annualized revenue from AI-related products had exceeded RMB49.5 billion ($7.3 billion), with AI-related products representing about 35% of Alibaba Cloud's external revenue at that point. Alibaba also said its model and application services had surpassed RMB16 billion in annual recurring revenue.
Those numbers are much closer to what investors traditionally want to know:
How much money is the AI business actually generating?
But Alibaba isn't abandoning monetization
That is an important counterpoint.
It would be wrong to interpret the machine-thinking narrative as Alibaba saying revenue no longer matters.
The company is simultaneously building an enormous commercial AI infrastructure business.
Alibaba says AI demand is increasing consumption of:
- GPU compute;
- tokens;
- storage;
- databases;
- networking;
- model-as-a-service products;
- AI applications.
The company calls this transformation an “Agentic Cloud.”
That creates an interesting business model.
Alibaba does not necessarily need every AI application to become a blockbuster consumer product.
If companies use more AI agents, they potentially consume more cloud infrastructure.
More agents → more inference.
More inference → more compute.
More compute → more cloud demand.
More cloud demand → potentially more Alibaba revenue.
The company's full-stack strategy is designed to capture several layers of that spending.
Alibaba wants to own the entire AI stack
This is arguably more important than the terminology.
Alibaba isn't simply betting on a chatbot.
It is attempting to control multiple layers of the AI economy:
Chips → Data centers → Cloud → Foundation models → Agents → Applications
At Apsara, Alibaba unveiled the Zhenwu V900, which it says is three times the performance of its predecessor and is scheduled for mass production in early 2027.
The company also announced plans for Alibaba Cloud's global data-center capacity to exceed 20 gigawatts by 2032.
At the model layer, Alibaba is developing future Qwen models with 5 trillion to 10 trillion parameters, substantially larger than its current flagship model.
That is a classic vertically integrated infrastructure strategy.
If AI becomes a new general-purpose computing layer, Alibaba wants to sell the electricity-equivalent, the hardware-equivalent, the operating infrastructure and the applications.
Why the GMV comparison matters
There is an important historical analogy here.
Alibaba's old economic engine was essentially:
People → platforms → transactions → commissions/services → revenue
The proposed AI-era engine becomes:
Machines → computation → agents → tasks → cloud consumption → revenue
The difference is profound.
E-commerce scales with the number of consumers and the amount each consumer purchases.
Machine intelligence potentially scales with the number of tasks that can be assigned to machines.
And unlike a human employee, an AI agent does not need:
- sleep;
- weekends;
- holidays;
- annual leave;
- additional office space for every additional worker.
If the cost of inference continues falling while AI capabilities improve, companies could economically assign machines an enormous number of tasks that would never have been economical to assign to humans.
That is the economic bet Alibaba is making.
The real metric investors should watch
This is where the “machine thinking” narrative needs to meet financial reality.
Investors should probably watch several numbers simultaneously.
Machine-thinking volume
This indicates how much AI activity is occurring.
Useful—but insufficient.
Token consumption
This tells us whether AI usage is actually increasing.
Alibaba has specifically linked rising AI-agent activity to increasing token and compute demand.
Revenue per unit of compute
This is much more interesting.
If compute doubles but revenue only increases 20%, the economics may be deteriorating.
Gross margin
AI infrastructure can be enormously capital intensive.
Revenue growth without attractive margins can create a very different investment story.
Capital expenditure
Alibaba needs huge infrastructure investment to capture the AI opportunity.
Its 20GW data-center ambition demonstrates the scale of the capital commitment being contemplated.
Recurring AI revenue
This may ultimately be more useful than an abstract measure of machine cognition.
Recurring revenue demonstrates that customers are willing to keep paying.
Free cash flow
Ultimately, the question is brutally simple:
Does the AI business generate more cash than it consumes?
The biggest uncertainty: AI infrastructure could become a commodity
There is another risk hidden inside Alibaba's thesis.
Alibaba is betting that the growth of machine intelligence will create enormous demand for infrastructure.
That is plausible.
But enormous demand does not guarantee enormous profits.
Cloud computing provides a useful historical analogy.
Infrastructure can become highly valuable while simultaneously becoming intensely competitive.
If:
- Nvidia sells the chips,
- TSMC manufactures them,
- Alibaba operates cloud infrastructure,
- Huawei develops competing accelerators,
- Tencent builds competing cloud capacity,
- ByteDance develops its own infrastructure,
- Baidu operates another AI ecosystem,
then customers have increasing choices.
That can push prices down.
The AI industry could therefore experience extraordinary growth in machine-thinking volume while margins remain under pressure.
China's chip restrictions make the strategy even more important
Alibaba's full-stack strategy also has a geopolitical dimension.
China faces restrictions on access to some of the world's most advanced AI accelerators.
That makes domestic semiconductor development strategically important.
Alibaba's Zhenwu V900 announcement therefore isn't simply another product launch.
It is part of a broader attempt to reduce dependence on foreign AI hardware while building an integrated Chinese AI ecosystem. Reuters reported that Alibaba's latest AI push comes amid intensifying competition between Chinese technology companies and U.S. firms and continuing U.S. restrictions on advanced AI chips.
If Alibaba can develop competitive chips, models and cloud infrastructure together, it potentially gains an advantage that a company dependent on external hardware may not have.
But that remains an execution challenge.
The biggest bet isn't the metric—it is the worldview
The fascinating thing about Eddie Wu's argument is that “machine thinking” is more than a KPI.
It is an attempt to redefine what an AI company actually is.
Under the traditional technology-company model:
Software helps humans perform tasks.
Under Alibaba's proposed model:
Machines become an industrial source of intelligence.
That is a much bigger proposition.
If Wu is right, then the AI economy isn't ultimately about chatbots.
It is about creating an industrial infrastructure capable of supplying cognition at scale.
That could eventually encompass:
- scientific research;
- engineering;
- software development;
- logistics;
- medicine;
- finance;
- manufacturing;
- education;
- entertainment;
- government services.
The economic unit wouldn't necessarily be a human user.
It could increasingly be a machine performing a task.
But investors should not confuse a powerful narrative with proof
This is the central issue.
Alibaba has a compelling story.
It has enormous cloud infrastructure.
It has the Qwen model family.
It is developing its own chips.
It is building an agentic cloud.
It is expanding data-center capacity.
And its reported AI-related revenue is already substantial and growing rapidly.
But the new machine-thinking framework remains fundamentally a forward-looking conceptual measure.
The ultimate test is financial.
Can Alibaba turn the explosive growth of machine computation into:
Revenue → recurring revenue → margins → free cash flow → returns on invested capital?
That is the chain investors ultimately need to see.
Alibaba is changing what it wants the market to count
The significance of the announcement is therefore not that GMV has suddenly become irrelevant.
It is that Alibaba wants the market to stop thinking of it primarily as an e-commerce company.
The company increasingly wants to be understood as an AI infrastructure and machine-intelligence company.
GMV described the world Alibaba helped build:
humans buying things through digital platforms.
Machine-thinking volume describes the world Alibaba believes is coming:
machines performing enormous amounts of cognitive work through digital infrastructure.
That is a much more ambitious vision.
But visions do not pay shareholders.
Revenue does.
And revenue becomes meaningful when it produces sustainable margins and cash flow.
So the most important question surrounding Alibaba's new “machine thinking” metric isn't how many trillions of AI tasks machines can perform.
It is much simpler:
How much money can Alibaba make from every unit of machine thinking it enables?
Until the company can answer that with increasingly strong financial results, machine-thinking volume remains a measure of technological horsepower—not yet proof of an enduring AI business model.
Comments
Post a Comment