---
title: "Baidu: Compute windfall in one pocket, out the other; when will Kunlun Chip come to the rescue?"
type: "Topics"
locale: "en"
url: "https://longbridge.com/en/dolphin/post/45093246.md"
description: "Compute profits are being plowed back into infrastructure. Ads no longer backstop earnings."
datetime: "2026-11-16T12:56:25.000Z"
locales:
  - [en](https://longbridge.com/en/dolphin/post/45093246.md)
  - [zh-CN](https://longbridge.com/zh-CN/dolphin/post/45093246.md)
  - [zh-HK](https://longbridge.com/zh-HK/dolphin/post/45093246.md)
author: "[Dolphin Research](https://longbridge.com/en/dolphin.md)"
generator: "portal-rs"
---

# Baidu: Compute windfall in one pocket, out the other; when will Kunlun Chip come to the rescue?

Pre-mkt on Nov 16 ET, BIDU (BIDU.O) reported Q3 results. The print was mixed but broadly in line with expectations, suggesting ample pre-earnings communication. In short, BIDU is still riding the compute boom, but incremental gains are being reinvested into new infra and foundation models. $Baidu(BIDU.US) , $BIDU-W(09888.HK)

As the village’s only true hope, Kunlun Chip’s spin-off IPO progress remains investors’ top focus. The discussion below refers to BIDU Core only, excluding iQIYI.

**1\. Ads remain weak: hit by both macro and AI.**

Under the prior scope, Q3 total ad revenue fell 23%, deteriorating QoQ. Traditional search was down 25%, and AI-native marketing also struggled, down 11% YoY.

Beyond the macro backdrop, AI-driven search cannibalization is increasingly evident, and even Alphabet’s Google is not immune. BIDU App MAU fell by 40 mn QoQ, underscoring the trend.

**2\. Cloud compute stays hot as expected: watch Kunlun Chip’s IPO**

Q3 AI infra (including bare-metal GPU rentals, MaaS model distribution, and Kunlun Chip’s external sales, etc.) grew 91%, with QoQ acceleration. Momentum remains robust.

With overseas compute restricted, onshore shortages have intensified, which marginally benefits early capacity accumulators like BIDU. While pricing sustainability is debatable, demand looks solid.

AI infra is now BIDU’s primary swing factor, including the Kunlun Chip IPO catalyst. Given the urgency of domestic compute substitution, a smooth listing could underpin group valuation.

In an early-Sep note, Dolphin Research valued Kunlun Chip’s equity within BIDU at $13 bn. Even with sector sentiment weaker and the P/S multiple cut from 30x to 20x–25x, a ~$10 bn handle still looks achievable.

**Since Oct, reports surfaced that Kunlun Chip may delay its IPO (Bloomberg: still awaiting mainland regulatory approval), and the broader backdrop has turned cautious. Sentiment has therefore become more guarded.**

**The company denied any delay and said the confidential filing is progressing, but it offered no clearer timeline.** That makes it hard to fully allay investor concerns. The post-Oct drawdown can be seen as a higher discount applied to a later IPO.

On the call, expect institutions to focus on Kunlun Chip’s listing timeline. Capacity planning will also be in the spotlight.

**3\. Margin slippage: ad drag, higher depreciation, and more model spend**

In Q2, management flagged Q3 as the trough quarter for Core OP. On GAAP, Q3 OP was only RMB 670 mn, pressured by higher costs (mainly depreciation), slightly higher R&D headcount for large models, and ECL provisioning.

Ex one-offs, non-GAAP OP was RMB 1.7 bn. This remains below historical quarters.

Mid-year, BIDU reorganized its large-model team and brought in talent from Fudan MOSS (Sun Tianxiang), DeepSeek (Wei Haoran), and a North America frontier AI Lab. The goal was to boost R&D vitality.

In Nov at its world conference, BIDU launched ERNIE 5.5 Flash (280 bn parameters), optimized for P800 performance at a lower price. It also disclosed that the Tianchi 1,024-card supernode will go live in 2027.

Q3 capex was RMB 7 bn, down QoQ as guided. Q4 should be flat to slightly up QoQ, taking FY capex to RMB 30–35 bn.

Because BIDU stocked cards early at lower cost, cloud margins are still enjoying a tailwind-driven improvement phase near term. That cushion may not last indefinitely.

To support expanding compute, BIDU is scaling base-layer infra. Right-of-use assets rose from RMB 9.3 bn at start of year to RMB 25.3 bn, while interest-bearing debt increased from ~RMB 89.5 bn to ~RMB 105.1 bn to stabilize short-term liquidity.

**With foreign cards restricted and higher marginal compute costs, margin improvement could slow, in Dolphin Research’s view.** The trajectory likely hinges on Kunlun Chip’s performance and capacity (M300 targeted for mass production in 2027).

Management previously guided confidence in lifting cloud GPM to 35%–40% within two years. Given the changed environment, watch for any update to this outlook.

**4\. Key metrics at a glance**

## Dolphin Research view:

Of the two catalysts highlighted last quarter, Stock Connect inclusion and the Kunlun Chip IPO, one has played out. The IPO delay rumor was denied by the company, but without greater certainty on the call, confidence will be hard to restore and the Kunlun discount hard to close.

Beyond catalysts, the Q3 surprise was a worse-than-expected QoQ ad performance, rather than a tentative bottoming. Management likely pre-briefed the market, and Alphabet’s print also showed AI erosion in search. Expectations had already been reset pre-earnings.

The issue is that with an unfavorable beta and limited alpha, legacy search makes it hard to call the bottom with conviction. Ads no longer fund cash flow, and AI-native marketing has yet to backfill the gap.

To keep cloud growing fast, BIDU needs fresh compute infra. The company has ~1 GW today and plans to double to 2 GW in 1–2 years.

**Net-net, BIDU’s ‘fire and ice’ pattern stands: hotter fire, thicker ice.** Cash flow can no longer backstop the business, and neither AI apps nor AI-native marketing is plugging the search hole (AI penetration is rising in Docs and Drive, but revenue is flat QoQ). In other words, AI products beyond selling compute still aren’t turning into cash.

**For now, compute windfalls are being plowed into upfront investment for future growth.** This is why Dolphin Research has stressed that a low headline multiple alone is not a buy thesis. Instead, we look for opportunities around the Kunlun Chip catalyst window.

With few updates on Kunlun’s listing progress, we assumed last quarter that the catalyst window might land by year-end. Ahead of that, the Q3 call is the key touchpoint. Meanwhile, with foreign card channels abruptly cut off, future compute deployment and planning have become focal points, making this call especially important.

On valuation, at the Nov 13 close of $78.14, BIDU’s mkt cap was ~$26.8 bn. Net cash stood at $19.2 bn at Q3-end, but unlike most, we prefer to exclude it.

The reason: during a visible investment phase, ad cash flow won’t backstop capex, and cash balances may be consumed. Despite a $5 bn 3-year buyback authorization, the slow pace speaks for itself. In other words, that ~$20 bn net cash isn’t effectively at shareholders’ disposal.

**For biz. value, we maintain an SOTP: ads + cloud ex-Kunlun + Kunlun Chip equity.** The framework remains unchanged.

**Our neutral valuation is $37.2 bn (slightly below the prior note), still implying ~40% upside.** This is our base case.

**We also run a more cautious bear case —** AI cloud plus ad revenue is fully reinvested into compute infra, leaving the P&L roughly breakeven. Under that scenario, the compute business ceases to create value for shareholders.

BIDU would then be valued on a breakup basis as $19.5 bn net cash plus proceeds from selling Kunlun: 195\*90% (assuming 10% operating wind-down loss) + 90\*70% (quick-disposal discount) = $23.9 bn. The current ~$26 bn mkt cap is already very close to this bearish construct.

\<End here>

Risk disclosure and statements: [Dolphin Research disclaimer and general disclosure](https://support.longbridge.global/topics/misc/dolphin-disclaimer)

**Dolphin Research articles on BIDU:**

**Recent earnings season**

Sep 7, 2026 Hot Take: [AI Cloud + Kunlun Chip: Does BIDU deserve a re-rating?](https://longbridge.cn/topics/43787192?channel=SH000001&invite-code=355628&app_id=longbridge&utm_source=longbridge_app_share&locale=zh-CN&share_track_id=cb39c134-9e38-41e8-b9e3-4cd5ebd1c51c)

Aug 18, 2026 Earnings note: [BIDU: Legacy biz still bottoming; AI driven by compute windfalls](https://longbridge.cn/topics/43459818?channel=SH000001&invite-code=355628&app_id=longbridge&utm_source=longbridge_app_share&locale=zh-CN&share_track_id=3d360983-73ec-47a2-9903-76e8595bbb42)

May 18, 2026 Trans: [BIDU (Transcript): Full-stack AI advantages in cost efficiency and deployment stability](https://longbridge.cn/topics/40885468)

May 18, 2026 Earnings note: [BIDU: Nothing else to count on, all-in on Kunlun Chip](https://longbridge.cn/topics/40881952)

### Related Stocks

- [09888.HK](https://longbridge.com/en/quote/09888.HK.md)
- [BIDU.US](https://longbridge.com/en/quote/BIDU.US.md)
- [89888.HK](https://longbridge.com/en/quote/89888.HK.md)
- [K3SD.SG](https://longbridge.com/en/quote/K3SD.SG.md)
- [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md)
- [HBUD.SG](https://longbridge.com/en/quote/HBUD.SG.md)
- [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md)
- [GOOGM.US](https://longbridge.com/en/quote/GOOGM.US.md)
- [GOOGN.US](https://longbridge.com/en/quote/GOOGN.US.md)
- [GGLL.US](https://longbridge.com/en/quote/GGLL.US.md)

## Comments (1)

- **Leslie Lim_vDcUH2 · 2026-11-17T01:07:04.000Z**: ...


---
> **Disclaimer: This article is for reference only and does not constitute any investment advice.**