This quarter, Alibaba delivered a report card that left the market stunned: revenue of 268.953 billion yuan, up 9% year-on-year, yet net profit attributable to shareholders plunged 75.6% to just 10.537 billion yuan. Free cash flow flipped from positive to negative, with a net outflow of 44.67 billion yuan, while capital expenditures surged 75% to 67.678 billion yuan. The immediate consensus from public opinion was almost uniform: AI has burned through Alibaba's profits.
Wall Street's skepticism even overshadowed the earnings report itself—with a 380-billion-yuan AI investment plan only halfway executed, 190 billion yuan in real money sunk in, and no decent return yet visible on the books. Just three days after the earnings release, Alibaba announced a placement of 80 billion Hong Kong dollars to further double down on AI. The pace of cash burn is faster than the criticism itself.
Is this explanation correct? Yes, but only on paper. If "betting heavily on AI necessarily destroys profits" holds true, what about Industrial Fulian? Also betting on AI computing power, Industrial Fulian's net profit grew 96% year-on-year, with AI server revenue surging 230%. Tuojing Technology saw net profit multiply 13-fold. In the same track, with equally heavy bets, some have turned investment into growth, while others have turned it into losses. The missing link in between isn't capital—it's organizational capability, organizational capability, and again, organizational capability.
So the real question worth asking isn't "how much has Alibaba spent on AI," but rather "has the organization—caught the money or not." To see whether it caught it, look at the actions. Alibaba has actually moved aggressively over the past two years, so aggressively that it can be summarized as "five steps in two years," with only one direction: forcing AI under a single baton.
In 2023, Eddie Wu took over as Group CEO, and his first move was to retract the decentralized "1+6+N" structure—canceling the Alibaba Cloud spinoff and taking on the role of Alibaba Cloud CEO himself. This marked the shift from "decentralization" back to "centralization," with a simple logic: AI must be fought collectively, not independently.
In August 2025, business operations were reorganized into four major segments, further strengthening group-level coordination. In March 2026, the ATH business group was established, bundling Tongyi Lab, MaaS, Qwen, Wukong, and five AI innovation units into one team, directly managed by Eddie Wu. The official mandate was summarized in three verbs: create tokens, deliver tokens, and apply tokens. Industry insiders called this the key step in Alibaba's shift from "product-centric" to "token-centric."
In June 2026, ATH was upgraded to Token Foundry, still under direct CEO management, with Jingren Zhou transitioning to Group Chief Scientist for frontier research, and Bo Zheng bringing the generative-side product lines closer to practical implementation. In August 2026, the earnings report segment was restructured, for the first time listing "AI Labs and Applications" as a standalone segment—Alibaba chose to openly disclose its most cash-burning ledger to Wall Street.
Across these five steps in two years, behind every change in the org chart, what truly shifted was the reporting lines: core AI businesses now report directly to the CEO. For a large company with complex operations, "straightening the reporting chain" is the highest-level declaration of priority—resource allocation, investment decisions, and cross-business coordination are all decided at the group's top level, without waiting for layer-by-layer approval.
After consolidating power came breaking down silos. Before ATH was established, Alibaba had over a dozen internal AI agent product lines, with QoderWork, Wukong, and MuleRun each fighting separately—inconsistent underlying architectures, data logic, and security standards. It wasn't until July 2026 that integration into "Qwen Office" began, and it's still being refined today—one employee openly said on a community forum, "I basically only use QoderWork now, rarely open Wukong anymore." The group-level technical committee established in April 2026 coordinates talent, computing power, and project prioritization, aiming to prevent "top algorithm engineers from being drowned in trivial business requests" and "GPUs alternating between idle and queued."
Eddie Wu also opened up internal networks across business units, adjusted cross-department mobility mechanisms, and explicitly set a goal of making post-85s and post-90s the backbone of management teams within four years. On the process level, what was done can be summed up in one sentence: shifting the way of fighting from "various warlords" to a "central army."
With the organization moved and processes changed, next came the money. According to employee leaks on Maimai—which Alibaba has not responded to—the year-end bonuses in August 2026 showed sharp divergence: the same P7 level employee doing model infrastructure in Alibaba Cloud with a 3.75 performance rating received 7.5 to 8.5 months of cash bonus plus 100,000 to 150,000 yuan in stock, totaling 700,000 to 900,000 yuan. Meanwhile, the same P7 with the same performance rating doing operations in local life services received only 3 to 4 months of cash, with no stock. The gap is two to three times. This isn't just about year-end bonuses—it's Alibaba using real money to draw a track for the entire company: the future belongs to AI, and the question is whether you're in it or not.
But the contradiction lies here too. The yardstick for performance evaluation hasn't kept pace with the flow of money. Officially, forced 361 assessments were abolished in December 2020, yet as of August 2026, employees on Maimai were still complaining about the "361 system," with 3.25 ratings still getting no year-end bonus and still being labeled as "needs improvement." The system changed, but employees don't even know it.
Even more contradictory, one self-media article pointed out sharply: employees who heavily relied on AI to generate 310,000 lines of code were rewarded, while those who spent time validating AI output and preventing major business risks for the company were instead penalized for "low efficiency." The performance yardstick hasn't anchored on the difference between "consuming tokens" and "creating value," so the results it measures are naturally absurd. Changing the yardstick is easy; changing people's minds is hard—and this is the toughest part of building organizational capability.
On the cultural front, Alibaba went through a forced course correction this year. In June, a former product manager at DingTalk published an exposé titled "Inside DingTalk," revealing high-pressure management and mechanical execution, with reports of 300 to 400 people leaving the core product and research teams. The Alibaba Partners Committee made a rare intervention, publishing an internal post titled "Compassion, Growth, and Meaning—That's Alibaba Culture," with one line carrying the most weight: "Innovation in the AI era has never relied on high pressure and mechanical execution, but on every colleague's passion and creativity." This was Alibaba's highest decision-making body publicly admitting that old management methods have failed in the AI era.
The public remarks from Alibaba Research Institute President Yuan Yuan were even more candid: "We cannot use the success of pioneers to mask the real pain that others endure during the transition buffer period. There aren't yet enough ways to support ordinary transformers." Even the official camp admits that the support mechanism for organizational transformation is blank.
So the question arises: with so many actions taken and so much money spent, why did profits still plunge 75.6%? The two standalone segments made the report look less favorable. AI Cloud and Computing Power—the most thoroughly reorganized segment—posted revenue of 48.437 billion yuan, up 45% year-on-year, with EBITA soaring 133% and a 12% profit margin. AI Labs and Applications—the part where the organization hasn't fully worked yet—posted revenue of 3.338 billion yuan, a loss of 13.861 billion yuan, and a loss margin of 415%. They share the same models, the same chips, and the same CEO; the only variable is the organization. One is an organization that has completed restructuring, and the other is still mid-restructuring. The gap in organizational capability building progress is clearly illuminated by the ledger.
This also explains the most easily misunderstood part of the whole story. Outsiders looking at Alibaba see "money spent, profits down," and attribute it to AI. But the true internal picture is: the structure has changed, the processes have changed, and the flow of money has changed, yet people's perceptions, the performance yardstick, and cultural inertia remain on the old track. Organizational transformation lags behind strategic announcements by half a beat, or even a year—Wall Street only looks at results, not processes, and it's precisely this time lag that it's voting against with its feet.
Moreover, Alibaba's real concern isn't just in AI's ledger: customer management revenue fell 7% year-on-year, with growth propped up by cash-burning instant retail. Market share is rented, and the rent gets more expensive every day—this is the symptom of organizational muscle atrophy. For Alibaba, its 380-billion-yuan foundation gives it the qualification to "trade profits for time," betting that the organizational capability gained in exchange will eventually balance the books. But for the vast majority of enterprises, there is no such qualification—between strategic correctness and successful transformation lies a chasm called organizational capability. Most companies don't die from wrong direction; they die from the organization failing to catch what's thrown at it.
Organizational capability has never been a supplementary clause to strategy; it is the only variable that determines whether strategy can land. The truth of transformation is this: proactive pain is called transformation; reactive pain is called rescue. Alibaba is currently experiencing proactive pain, but what this pain can buy doesn't depend on money—it depends on whether it can, beyond the ledger, complete that slowest of tasks—organizational capability—just a little faster.
Comments