Two Curves Moving Opposite Directions Inside One Company

Here's the deal: Baidu reported Q2 2026 on August 18. Total revenue came in at RMB 31.3 billion (about $4.6 billion), down 4% year over year — a fifth consecutive quarter of decline, and a miss against consensus.

In the same release: GPU cloud revenue up 283%, accelerating from 184% the prior quarter.

Seeing numbers point that hard in opposite directions inside one earnings report is unusual. This isn't a story about Baidu doing well or badly. It's a story about two different companies inside one shell, one of them dying while the other grows.

The dying one is online marketing — search advertising. Q2 revenue of RMB 13.1 billion, down 19%. The business that funded this company for over twenty years.

The growing one is AI cloud. AI cloud infrastructure revenue reached RMB 7.3 billion, up 50%, with GPU cloud inside it up 283%. The full "Core AI-powered Business" segment — cloud infrastructure plus AI applications plus AI-native marketing — came in at RMB 12.5 billion, up 25%.

A quick note on the company sharpens the contrast. Founded in 2000, China's largest search engine, once called "China's Google." Most of its revenue came from ads attached to search results, and that cash funded autonomous driving (Apollo), smart speakers, and foundation models (ERNIE). What's happening now is the phase where the cash-generating business dries up and the businesses it funded start generating cash. The problem is that the handoff is running faster than planned.

Why Advertising Is Falling This Hard

Nineteen percent isn't a macro-slowdown number. There's structure behind it.

First, search itself has shifted position. People typing keywords and clicking blue links is declining in favor of asking a chatbot in sentences and reading a summarized answer. That's not specific to China or to Baidu — but it hits harder at a company that draws nearly half its revenue from search ads.

Second, a dilemma Baidu created for itself. It put AI summaries at the top of search results. User experience improved; a user satisfied by the summary doesn't click the ad link. Improving the product cuts the revenue. Google carries the same problem, but Google has thick cushions in cloud and YouTube.

Third, competition. In China's ad market, content platforms like Douyin and Xiaohongshu are absorbing budget. The migration from search ads to feed ads has been running for years.

Fourth, advertiser mix. Baidu's search advertising leaned on education, healthcare, gaming, and e-commerce, and regulatory tightening over recent years sharply reduced spend in the first two. Macro, regulatory, and technological pressure are all pushing the same direction. Fixing any one of them wouldn't produce a rebound.

The Numbers

Item Q2 2026 YoY
Total revenue RMB 31.3B (~$4.6B) -4%
Online marketing (search ads) RMB 13.1B -19%
Core AI-powered business RMB 12.5B +25%
AI cloud infrastructure RMB 7.3B +50%
GPU cloud — +283%
Net income attributable to Baidu RMB 2.3B 7% net margin
Diluted EPS per ADS RMB 5.74 —

Look at the gap between advertising (13.1) and the core AI business (12.5): RMB 0.6 billion. One more quarter at this trajectory and AI overtakes advertising. The point where the company's identity changes is visible in the numbers now.

Net income was RMB 2.3 billion at a 7% margin. Not a loss, but nowhere near what search advertising used to produce. Advertising has near-zero marginal cost; cloud means buying GPUs and burning electricity. Change the revenue mix and the margin structure changes with it.

Dou Shen, executive vice president of Baidu AI Cloud Group, pushed back on that framing in an interesting way: GPU cloud carries a more attractive margin profile than traditional CPU cloud, and it's becoming a larger share of cloud revenue. That contradicts the received wisdom, but it makes sense once you account for high GPU instance pricing and tight supply putting negotiating power on the seller's side.

Unpack the margin argument a bit further. CPU cloud is a standardized commodity with fierce price competition, and China's market has been especially brutal on that front. GPU instances are supply-constrained, so there's no reason to discount, and utilization stays high because idle time is short. All of which assumes scarcity. Loosen export controls or scale up domestic chip supply and the margin advantage can evaporate quickly.

Who Gets What Out of This

Baidu gets time. If cloud grows faster than advertising shrinks, the company survives. The current numbers barely satisfy that condition: total revenue fell only 4% because a 25% AI gain nearly offset a 19% advertising loss.

Chinese AI companies get a domestic GPU source. With US export controls making the latest Nvidia silicon hard to procure directly, renting GPUs a domestic cloud operator already secured became the practical option. A large share of that 283% is explained by exactly this demand. If you can't buy the chip, rent it by the hour.

Nvidia gets indirect demand. Even without selling current-generation silicon into China directly, a growing resale-by-the-hour market raises utilization on chips already sold — and produces a continuous stream of data on how large Chinese GPU demand actually is. It's the kind of number both sides cite in export control debates.

Investors get a muddled signal. Revenue has declined five quarters running while the growth segment's growth rate accelerates. Which one you weight determines your conclusion entirely, which is why the post-earnings stock reaction was mixed.

Apollo Go took a hit this quarter. Ride volume dropped temporarily after operational adjustments in certain domestic cities for regulatory reasons. Autonomous driving is a long-running Baidu bet, and this is a reminder that regulatory variables land directly in quarterly results.

More on that: the weakness wasn't demand, it was a compliance-driven pullback in specific cities. Robotaxi is licensed municipality by municipality, so one city's policy change can zero out that region's revenue. Earnings volatility is structurally high regardless of how mature the technology is — which is why, after years of investment, this still isn't a stable pillar of results.

AI applications are folded into these figures too. The RMB 12.5 billion core AI business includes AI applications and AI-native marketing services on top of the RMB 7.3 billion in cloud infrastructure. Which means part of the advertising business has been reclassified into the growth line under the "AI-native marketing" label. Segment definitions changing alters what a growth rate means, so it's worth reading the definitions in the release directly.

Companies That Went Through This Transition

Microsoft is the successful case: shifting weight from Windows licensing to Azure. The key was that the two businesses didn't overlap. Windows declined gently and Azure grew in an entirely new market, so total revenue never broke during the handoff.

IBM is the opposite. Moving from mainframes and services toward cloud and AI, the new business never caught up to how fast the old one shrank. Revenue stalled or fell for the better part of a decade, and the portfolio had to be rebuilt through spinoffs and acquisitions. Transitions are decided by relative speed, not direction.

Yahoo is the extreme ending. Search and portal lost to Google while no replacement growth axis materialized, and the company was dismantled. That's precisely the scenario Baidu is trying to avoid.

Netflix's DVD-to-streaming pivot offers a different angle: it killed its own business to move, and the stock swung hard doing it. But Netflix started the transition voluntarily and controlled the timing. Baidu's advertising decline isn't voluntary — the market forced it — which makes the conditions worse.

The question running through all four is the same: which is faster, the decline of the old business or the growth of the new one. Microsoft's new business was faster; IBM's old one was; Yahoo had no new one. Baidu's quarter shows the two speeds running nearly even — a 19% advertising hole almost filled by 25% AI growth, with -4% left over at the top line. Which way that balance tips next quarter plausibly decides the company's next five years.

How the Competitive Board Moves

Alibaba Cloud leads China's cloud market, invests far more in AI infrastructure, and has the Qwen open-weight line. Baidu's GPU cloud growth rate is high, but the absolute gap is still wide.

Tencent Cloud and Huawei Cloud are in the same market. Huawei in particular has its own Ascend silicon, which becomes relatively more valuable the tighter Nvidia supply gets. Baidu has developed its own Kunlun AI chip, but trails on ecosystem maturity.

ByteDance squeezes Baidu on both fronts: Douyin taking ad budget, Volcano Engine pushing into cloud. One competitor on two lines at once.

Chinese regulators cut both ways. As Apollo Go's quarter shows, regulation can halt a business outright. Conversely, policy support for domestic AI infrastructure favors local operators like Baidu.

Foreign hyperscalers are effectively absent. AWS and Azure operate in China under constrained partnership structures, but they aren't real variables in mainland competition. Every competitor Baidu faces is domestic — which limits upside and also insulates it from external shocks.

So What Actually Changes

If you follow Chinese AI markets, watch mix rather than total. RMB 31.3 billion says stagnation; RMB 12.5 billion growing 25% inside it is the more informative number. Whether AI overtakes advertising next quarter is the thing to watch.

If you're in the cloud business, the claim that GPU cloud beats CPU cloud on margin deserves attention. If true, it justifies reordering investment priorities. But it's a claim conditioned on today's scarcity, and scarcity ends.

If your business depends on search advertising, treat 19% as a reference point. It's rare public data on the revenue-scale impact of AI summaries on search traffic.

If you're evaluating enterprise cloud, this quarter tells you about negotiating conditions. GPU instances are a seller's market right now, so long-term contract discounts are hard to extract. CPU workloads are fiercely contested and negotiable. Split the two lines even when you're negotiating with the same vendor.

If you procure AI infrastructure, these numbers show how large the rent-instead-of-buy demand is in China. A 283% growth rate isn't a normal demand curve — it's distortion produced by supply constraints.

If you're an individual investor, get in the habit of checking whether segment definitions match the prior quarter. Companies in transition sometimes widen the boundaries of the growth segment, which flatters the growth rate. Baidu's "core AI business" line mixes cloud and advertising-derived revenue, so reading it as a single clean metric invites misinterpretation.

🥄 Three Things You're Probably Wondering

— Isn't five straight quarters of decline dangerous? It's a warning sign, yes. But the decline narrowed to 4% and the growth segment's rate accelerated. The question is when the top line turns, and on this trajectory that could arrive around the same time AI overtakes advertising. Too early to call.

— Is 283% GPU cloud growth sustainable? Base effects are doing a lot of work there — small starting points produce big percentages. That said, accelerating from 184% the prior quarter isn't explained by base effects alone. As long as US chip export controls hold, the underlying demand likely persists.

— Should I buy Baidu stock? Not something to answer here. What this quarter establishes is two facts: the core business keeps shrinking, and the new business grows fast but hasn't passed it yet. When those two curves cross is the judgment call, and the raw material for making it is now on the table.

References

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!