NextFin News - Baidu's artificial-intelligence business is on track to generate profits and cash payback equal to its legacy search operation "soon," Chief Financial Officer Haijian He said, a milestone that would validate the company's painful pivot away from the advertising model that built it. The finance chief's comments, made as AI-driven revenue already accounts for more than half of sales, mark the clearest signal yet that China's search giant is becoming an AI company with an internet foundation - not the other way around.
The stakes are concrete. Baidu's Core AI-powered Business - cloud infrastructure, AI applications and AI-native marketing services - generated RMB 12.5 billion in the second quarter of 2026, up 25% from a year earlier and equal to exactly half of Baidu General Business revenue. Within that, AI Cloud Infrastructure revenue rose 50% to RMB 7.3 billion, and GPU Cloud revenue surged 283% year over year, accelerating from 184% growth in the prior quarter. Meanwhile, the old engine is stalling: online marketing services fell 19% to RMB 13.1 billion, and the broader legacy business dropped 23% to RMB 13.6 billion. Net income attributable to Baidu fell 68% year over year to RMB 2.3 billion, a 7% net margin, as capital expenditures nearly doubled sequentially to RMB 11.39 billion and free cash flow sank to negative RMB 7.95 billion, about US$1.18 billion.
He's claim reframes that pain. If AI revenue margins converge with search margins in the near term, and if new GPU clusters pay for themselves in two to three years, then the current cash burn is a bridge, not a leak. But the timing of that convergence - and whether the ad decline can be held at bay long enough for AI margins to catch up - is the question the market has not yet decided.
The Numbers Behind the Pivot: Half the Business Is Now AI
The mix shift is no longer a narrative; it is a reported line item. In the quarter ended June 30, 2026, revenue from Baidu Core AI-powered Business reached RMB 12.5 billion and accounted for 50% of Baidu General Business revenue - up from 38% a year earlier. In the first quarter, the AI-powered core exceeded half of general business revenue for the first time, at more than RMB 13.6 billion and up 49% year over year; the second quarter's RMB 12.5 billion represents a 25% annual gain, a deceleration that shows the transition is real but not yet frictionless. The company's own revenue table put the AI share at 52% for the quarter. That is the fastest structural revenue rotation any major Chinese internet platform has disclosed, and it happened while total revenue declined 4% year over year to RMB 31.3 billion.
The growth is concentrated in infrastructure, not just models. AI Cloud Infra revenue of RMB 7.3 billion grew 50% year over year, with GPU Cloud - public cloud-based AI computing - expanding 283%. AI Applications contributed RMB 2.5 billion, up 3%, and AI-native marketing services were roughly flat at RMB 2.6 billion. The contrast with the legacy stack is stark: Legacy Business revenue of RMB 13.6 billion fell 23%, and the online marketing services line - the search ads that funded Baidu for two decades - dropped 19%.
Profitability tells the other side of the story. Net income attributable to Baidu was RMB 2.3 billion with a 7% net margin; non-GAAP net income was RMB 2.6 billion, and non-GAAP diluted earnings per ADS of RMB 7.22 missed the consensus estimate of RMB 9.84 by a wide margin. Cost of revenue rose 4% year over year to RMB 19.1 billion, driven by AI cloud-related costs, while operating expenses fell 17% to RMB 9.2 billion. Operating cash flow remained positive for the fourth consecutive quarter at RMB 3.4 billion - the metric management points to when investors ask about cash discipline - but free cash flow went negative as capex nearly doubled in a single quarter.
CEO Robin Li framed the trade-off plainly in the company's earnings release:
While our online marketing business remains under pressure, the growing momentum in our core AI-powered Business reaffirms Baidu's transition from an internet-centric company to an AI-first company, and strengthens our confidence in our long-term growth potential.
The CFO anchored the cash story in the same release.
CFO Haijian He, who joined Baidu from Kingsoft Cloud in July 2025, anchored the cash story:
Operating cash flow for Baidu remained positive for the fourth consecutive quarter, reaching RMB 3.4 billion in the second quarter.
He also confirmed the company is moving forward with its conversion to a dual-primary listing in Hong Kong, expected to take effect within the year.
Why Margins Could Converge - and Why That Is Not the Same as Profits Catching Up
He's statement that AI revenue margins could match search margins "in the near term" deserves careful parsing. Search advertising is one of the highest-margin businesses ever invented - once the index is built, each additional query costs almost nothing. Cloud infrastructure, by contrast, is capital-intensive by design: every unit of revenue carries depreciation on GPUs, power, and data-center space. For AI margins to converge with search margins, Baidu needs one of two things, or both: a dramatic rise in utilization of its GPU clusters, and a shift in the AI revenue mix toward higher-margin software and model services rather than raw compute rental.
The 283% growth in GPU Cloud revenue suggests the utilization argument is working. A GPU cluster that sits half-empty is a margin sink; one that runs near capacity spreads its fixed depreciation across more billable tokens and inference calls. He's two-to-three-year payback window for new GPU clusters implies an internal hurdle rate that only makes sense if utilization is already high and rising. That is the mechanism: capex today buys capacity, utilization fills it, and the depreciation burden per unit of revenue falls faster than the revenue itself grows.
But there is a second mechanism at work, and it is less discussed. Baidu is not just renting GPUs - it is selling its proprietary Kunlun chips and its Ernie models as a stack. The company's chip unit, Kunlunxin, is pursuing a Hong Kong listing, with reported target valuations ranging from about US$14.7 billion to US$50 billion depending on the report, and Baidu will retain a controlling stake. If Kunlun chips replace imported accelerators in Baidu's own data centers, the effective cost of compute falls; if Kunlun sells to external customers such as Tencent and ByteDance, the chip business becomes a margin contributor of its own. The CFO's margin-convergence comment and the chip spin-off are two halves of the same vertical-integration thesis.
This is where the cyclical-versus-structural call matters. The revenue mix shift - AI crossing 50% of the core business - is structural. It is a regime change in what Baidu sells, and it will not revert on its own. But the margin convergence is partly cyclical: it depends on the AI capex supercycle delivering utilization before the next generation of hardware makes today's GPUs obsolete. If the capex cycle turns and utilization stalls, margins revert. The right read is a structural revenue transition riding a cyclical profitability wave - and the two can point in opposite directions for several quarters.
The Counter-Thesis: The Ad Decline May Outrun the AI Build
The strongest case against Baidu's convergence story is simple arithmetic, and it is the view already priced into the shares. Online marketing revenue fell 19% year over year to RMB 13.1 billion, and management expects the business to remain under pressure in the second half. Search advertising is not just declining - it is declining while Baidu is spending heavily to replace it. If the ad business falls another 15% to 20% over the next two quarters while AI cloud gross margin stays below search margins, total profit will keep shrinking even as the AI narrative improves. The market's reaction says it is worried about exactly this: shares sank more than 12% after the August earnings release, and Wall Street analysts responded swiftly. Morgan Stanley's Gary Yu downgraded the stock to Underweight from Equal-weight and cut his price target to US$80 from US$130; Barclays lowered its target to US$96, citing an advertising decline that has now run for eight consecutive quarters since the second quarter of 2024 and has accelerated to roughly 20% year over year in recent periods.
The cash-flow evidence feeds the bear case. Capital expenditures nearly doubled sequentially to RMB 11.39 billion, pushing free cash flow to negative RMB 7.95 billion. Baidu is not alone - Bank of America estimates hyperscaler capital spending will reach roughly US$860 billion in 2026 and could approach US$1.2 trillion in 2027, with aggregate free cash flow set to finish the year negative. But Baidu has less room than Microsoft or Alphabet: its advertising cash cow is shrinking, not funding the build from a position of strength. Alibaba, the larger rival, reported a 75% drop in quarterly net profit after ramping AI capex 75% to RMB 67.68 billion, even as AI-related products reached 35% of Alibaba Cloud's external revenue. The entire Chinese cloud sector is making the same bet; the winner will be the one whose utilization improves fastest, not the one that spends the most.
The counter-thesis has a nameable falsifying signal. If online marketing revenue falls more than 20% year over year for two consecutive quarters while AI Cloud gross margin remains below the search business margin, the convergence timeline breaks and the stock's multiple has no earnings anchor. A second signal: if quarterly capex stays above RMB 11 billion through 2027 without a corresponding rise in GPU Cloud utilization, the two-to-three-year payback window stretches and the CFO's milestone slips.
What Comes Next: Three Horizons for the AI Transition
In the short term - the next two to three quarters - the stock will be driven by the advertising trend and the cadence of capex, not by AI revenue growth. Investors will watch whether the 19% ad decline stabilizes, whether operating cash flow stays positive, and whether the dual-primary Hong Kong listing completes within the year as management expects. Sentiment is fragile: the same AI capex that promises future margins is depressing current free cash flow, and the market punishes that gap. The analyst community is split along the same fault line - Morgan Stanley sees core operating profit falling to RMB 2.3 billion from RMB 3.8 billion, while bulls point to GPU Cloud growth of 283% as proof the AI engine is only now reaching speed.
Over the medium term - 2027 to 2028 - the base case is that AI Cloud Infrastructure keeps growing at 40% to 50% annually as GPU Cloud demand persists, GPU Cloud growth moderates from triple digits but stays strong, and AI applications scale as Ernie Assistant, which surpassed 200 million monthly active users in early 2026, converts users into paying subscribers. In this scenario, AI revenue margins approach search margins by late 2027, and the Kunlunxin listing unlocks sum-of-the-parts value. The upside case adds faster ad stabilization and higher chip pricing power; the downside case is a deeper ad recession that forces Baidu to choose between defending search profitability and funding AI capex.
In the long term, the structural question is whether Baidu becomes a full-stack AI infrastructure provider or remains a search company with a cloud division. The answer hinges on three things: whether Kunlun chips can compete with imported accelerators in a market cut off from the most advanced foreign supply, whether Ernie retains its lead against domestic models, and whether Apollo Go's robotaxi expansion - now in 28 cities with over 350 million autonomous kilometers - becomes a meaningful third revenue pillar. These are regime-level questions, and they will not be answered by a single quarter.
The market is not wrong to be cautious. Baidu's AI transition is real, but it is being funded by a shrinking cash cow at a difficult moment in the capex cycle. The CFO's promise that AI profits will match search profits soon is credible only if utilization keeps rising and the ad decline does not accelerate. The next two quarters will tell which force is stronger.
The verdict: Baidu has crossed the revenue Rubicon - half its core business is now AI - but the profit Rubicon is still ahead, and the river is widest exactly where the ad business is eroding fastest.
Explore more exclusive insights at nextfin.ai.
