Author｜Mei Wen

This may be the most noteworthy financial report in China's tech industry over the past decade. Not because the numbers themselves are particularly earth-shattering, but because they declare the end of an old paradigm in a near-blunt manner.

The core narrative of the financial report revolves around a sharp inflection point: the Cloud Intelligence Group, which carries Alibaba's future imagination, achieved a 38% year-over-year revenue growth rate, with AI-related product revenue achieving triple-digit year-over-year growth for the eleventh consecutive quarter, heading toward an annual revenue scale of 40 billion yuan. The growth rate of customer management revenue for the Tao Tian Group remains in the single digits, a tension that transcends the rise and fall of business segments.

This financial report strips away all sentimental embellishments and delivers a stark conclusion: the golden age of 'software-defined' systems, with GMV, traffic, and user engagement as its underlying code, is coming to an end. A new continent of 'model-driven' systems, composed of computing power, models, and intelligent agents, has emerged.

The Quiet Handover of Value Creation

Over the past two decades, nearly all internet giants have essentially been software companies. Their core assets were applications built with code—whether e-commerce, social media, or search. These applications defined user experiences, allocated traffic, and completed transaction loops. Software was the ultimate lever to move the world; mastering software definition meant controlling the commercial lifeblood of an era.

In that golden age, a company's valuation depended on how many daily active users its applications could reach, how much user time they occupied, and how much GMV they generated. Building technical stacks, designing organizational structures, and setting KPIs all revolved around getting those lines of code in front of more fingers.

This financial report reveals a fundamental shift.

When Alibaba's engineers are no longer racing day and night to optimize recommendation algorithms for Taobao and Tmall but instead focusing their core efforts on iterating the Tongyi Qianwen model, adapting Pingtouge chips, and deploying models to BMW and China Unicom's production lines, the locus of value creation has physically migrated.

Software itself is being downgraded; it is no longer the source of value but has become an interface for invoking model capabilities.

A clear signal: This year, Alibaba Cloud serves approximately 63% of China's A-share listed companies. These enterprises are no longer access (Note: This Chinese term seems out of context here and may require verification) a standardized software system but a digital hub with continuous evolutionary capabilities. In the past, you bought software to manage inventory and generate reports; now you access (Note: This Chinese term seems out of context here and may require verification) a model to read contracts, write code, and make decisions. The divide between the two is like that between an abacus and a brain.

The underlying logic of this shift is that software solves efficiency problems, while models solve capability problems.

Efficiency improvements have a ceiling; capability expansion has no bounds. When a company discovers that invoking a model can replace an entire customer service team, a legal department, or even a portion of R&D functions, its willingness to pay for traditional software plummets.

The triple-digit growth in AI revenue in Alibaba's financial report is essentially harvesting this substitution demand.

And this is just the beginning.

The famous slogan 'software eats the world' is being replaced by 'models eat software.' This financial report adds the most solid annotation to this slogan with the flow of real money. Software has not disappeared; it has merely become a shell, its soul extracted by models.

The End of Standardized Replication

The core business model of traditional software is standardization and scalable replication.

Diminishing marginal costs are the golden rule of this industry—developing a SaaS product and selling it to a hundred customers versus ten thousand customers involves vastly different cost structures. When marginal costs approach zero, profits begin to explode. This is the underlying logic behind the rise of Salesforce and the capital stories repeatedly told by Chinese SaaS entrepreneurs over the past decade. The entire industry's imagination was built on the narrative of 'create a hit product and then sit back and collect money.'

The service logic of the AI era is dismantling this rule.

Deep customization, private deployment, and industry-specific models each entail significant computing power consumption and engineering tuning costs. You may think you're selling an API, but in reality, every client demands that you integrate the model into their private cloud, fine-tune it with their own data, and ensure data remains within their domain.

This is not a business of rolling out a standard product and sitting back to collect money but a heavy-service model requiring continuous investment and repeated refinement. The sharp decline in profits and the surge in capital expenditures to 120 billion yuan in Alibaba's financial report are precisely the price paid for this 'anti-scale' nature.

This price reveals a harsh reality: the era of making easy money by selling standardized APIs is far from over.

The process of models devouring software inevitably involves high costs and painful business restructuring.

The efficiency narratives that once made the SaaS model shine are rendered feeble in the face of models' computing power black holes. A troubling inference is that if the golden rule of SaaS—diminishing marginal costs—fails in the model era, then the entire SaaS industry's valuation model needs to be rewritten. The unicorns in the primary market valued at 'software subscription revenue × multiple' may be undergoing a quiet asset revaluation.

Profits no longer concentrate in the most standardized segments but precipitate (Note: This Chinese term seems out of context here and may require verification) in the underlying infrastructure layers with the deepest computing power reserves and the most advanced model capabilities.

Cloud vendors capture the largest share of the pie, while application-layer players struggle to survive in the quagmire of customized services. The twilight of the SaaS empire may arrive sooner than we imagine.

A New Order in the Digital World

When a trillion-yuan revenue giant begins to bet heavily on a technology with massive profit-table contractions, it is gambling on a structural fracture (Note: This Chinese term seems out of context here and may require verification) in the era.

This financial report clearly presents the path of power transfer after this fracture (Note: This Chinese term seems out of context here and may require verification): the voice in the digital world is shifting from the software application layer, which captures user attention, to the underlying infrastructure layer, which controls core computing power and foundational models.

The 170,000 models derived from Tongyi Qianwen represent a deep binding to Alibaba's technology stack with each derivation and fine-tuning.

This is a ritual of ecological allegiance—training models on your base locks in my future technology roadmap.

When a startup chooses to fine-tune its industry model on Tongyi's base, its future model iterations, computing power expansions, and deployment optimizations all become dependent on Alibaba's infrastructure. This locking effect is far stronger than the bundling of Windows operating systems to PC manufacturers in the past.

Wu Yongming's statement that 'not a single card is idle on Alibaba's servers' depicts this new form of power. Computing power has become the era's scarcest hard currency, defining who can set the underlying rules in the next era.

A deeper change is that this power not only manifests at the commercial level but also permeates the industrial level.

BMW optimizes its production processes with Alibaba's models, and China Unicom reconstructs its customer service with Alibaba's models. These cases bid farewell to the clichés of 'digital transformation' and announce a new industrial division of labor: those who control models begin to define the operating standards of manufacturing and services.

After all, when a car manufacturer's core production decisions begin to rely on a large model, the model provider substantially gains pricing power over the supply chain.

The mass production of Pingtouge's self-developed GPU chips pushes this power game to a deeper dimension. This is no longer a competition of who can write more sophisticated algorithms but a hardcore war centered around the physical world's computing power supply, wafer capacity, and energy consumption. Alibaba attempts to complete a vertical integration closed loop (Note: This Chinese term seems out of context here and may require verification) from chips to models to applications. Once this closed loop (Note: This Chinese term seems out of context here and may require verification) is formed, the competitive barriers it constructs will be exponential, making it nearly impossible for latecomers to catch up.

In the past, when we talked about 'technical barriers,' we referred to a line of code or a patent; now, when we talk about barriers, we refer to hundreds of thousands of GPU cards, self-developed chip architectures, and data centers spread globally. The materiality of barriers has increased, and the difficulty of breaking them has risen exponentially.

Looking back at this financial report, the 38% growth rate in cloud business and the triple-digit growth in AI revenue for eleven consecutive quarters send a sufficiently clear signal: a new power center is forming, and its expansion speed exceeds most people's expectations.

Companies still measuring competitiveness with old dimensions like 'customer acquisition cost' and 'user retention' may not realize that the battlefield has changed.

Is Your Moat Still in Place?

On the back of this financial report lies a question that all tech companies must face: Is your moat still in place?

For Alibaba itself, this is a life-or-death transformation.

Actively tearing off the e-commerce label and exchanging it for a ticket to the next era at the cost of massive profit fluctuations is a gamble as risky as any. Historically, few giants have successfully navigated cyclical transitions; more often, the cash flow from old businesses is dragged down by investments in new ones, while the new businesses fail to establish absolute advantages within the window of opportunity. Alibaba's financial report has provided a phased answer—the commercialization of AI-related products is accelerating, and the ecosystem scale of Tongyi Qianwen is expanding, but massive capital expenditures will continue to suppress the profit table for the long term.

This gamble is far from reaching its conclusion; the only certainty is that Alibaba has pushed all its chips onto the table.

For the entire industry, this financial report serves more like a coordinate origin.

It forces everyone to re-examine the true location of their barriers. When your competitors are already using computing power clusters and foundational models to forge new power foundations, are the application barriers you've built with code still as impregnable as before?

In the model era, the migration cost of the application layer is rapidly declining. If a competitor's model is powerful enough, users can switch from one e-commerce platform to another or from one office suite to another almost seamlessly. Because the real value no longer resides in the interaction interface but in the intelligence provided by the model.

Those still immersed in the organizational structures, talent pools, and technical path dependencies of the last 'software golden age' may find themselves more fragile than imagined in the face of this 'models devour everything' wave.

R&D teams built around software engineering face an insurmountable skills gap when confronted with the algorithm engineers, data annotators, and computing power scheduling experts needed in the large model era. This gap cannot be bridged by hiring a few people; it requires a complete reorganization of the organizational structure, a redesign of the KPI system, and even a fundamental transformation of corporate culture.

This financial report is like a sharp whistle. It does not pause for anyone's hesitation or take responsibility for anyone's nostalgia. It faithfully records a fact: the era of software defining everything is over, and models are taking over the world.

Those who react slowly will ultimately be left behind on the platform of the old era. And those left on the platform may not realize that the train has already left long ago.

*All images in this article are sourced from the internet