Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
IndustryInsight · On RadarTNW→ original

Chinese Tech Giants Double AI Spending to $139B, While Labs Recover Less Than 8%

Chinese tech giants are building AI infrastructure faster than they earn from it. According to Rhodium Group, 13 companies will invest $139B in it this year (103% more), while the revenue of their AI labs covers less than 8% of that sum. All Chinese models combined bring in about $10.7B a year, roughly ten times less than OpenAI and Anthropic. That means model providers face risk not only in chips but also in money.

AI-processed from TNW; edited by Hamidun News
Chinese Tech Giants Double AI Spending to $139B, While Labs Recover Less Than 8%
Source: TNW. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTNW ↗
Editor-in-Chief
2027 forecastmore than $193B (1.2 trillion yuan)
Key metric #2The STAR 50 index, China's Nasdaq equivalent, has fallen 43% since late July
Key metric #3DeepSeek needs four Huawei 950 chips to match one Nvidia GB300

Imagine a company that doubles its server spending every year, while the revenue from the service running on those servers barely covers a small fraction of it. That is how the AI race among China's largest tech companies looks, according to consulting firm Rhodium Group. Rhodium surveyed 13 public companies. This year they will invest 103% more in AI infrastructure (data centers, chips, servers) than last year: 932 billion yuan, or about $139B. Yet the revenue of their AI labs, meaning the units that build and sell models, covers less than 8% of that sum. The New York Times reported this on June 13, and we are examining what it means for those building products on Chinese models.

The conversation about the US-China AI race usually boils down to chips: who can buy what, and which export restrictions Washington imposes. The new figures show a second problem: China lacks not only hardware but also the money that would justify buying it.

Spending Is Outpacing Revenue

According to Rhodium's forecast, AI infrastructure spending will exceed 1.2 trillion yuan ($193B) in 2027. For scale, China's AI infrastructure is only 15–20% of America's. The annual recurring revenue (ARR, meaning yearly income from subscriptions and paid access) of all Chinese AI models combined is about $10.7B. That is roughly one-tenth of what OpenAI and Anthropic report together.

Kyle Chan of the Brookings Institution compared the situation to a treadmill that keeps speeding up, making each step harder. Alibaba, Baidu and Tencent, which back several leading Chinese labs, now have negative cash flow according to the Times: they spend more than they take in. Investors have also cooled. The STAR 50 index, China's equivalent of the Nasdaq, has fallen 43% since late July.

Open Models and a Hardware Shortage

China's bet on open models complicates monetization. In August, Z.ai released GLM-5.3 Flash and halved prices for two weeks, because the company wanted users to run the model on its servers. By November, as the Times reports citing OpenRouter data, about 20 competing providers were selling access to the same model for less than Z.ai itself. If the model weights are open, anyone can offer the model more cheaply than its author.

Compute remains scarce and expensive. On the Chinese secondhand marketplace Goofish, sellers listed an Nvidia DGX B300 server for almost $3 million, roughly eight times the price in the US. This week Washington moved to close one loophole: renting overseas data centers equipped with Nvidia hardware. Tencent and Alibaba have already warned about the consequences of the new restrictions.

DeepSeek's head, Liang Wenfeng, reportedly said in a leaked recording of a conversation with investors that Huawei chips lag Nvidia by about two years. To match a single Nvidia GB300, DeepSeek needs four Huawei 950 chips. "We simply cannot afford to train the largest models right now," he said. DeepSeek has raised more than $20B, including from investors linked to Tencent and CATL, but few competitors have such reserves.

Where the Money Shortage Comes From

Private AI investment in China totaled $12.4B in 2025, against $285.9B in the US. This figure probably understates the real picture, because state funds also invest in AI. Foreign money has largely left, in part because the US bars American venture funds from investing in certain Chinese technologies.

China still has strengths. Its researchers have repeatedly found ways to train and run models more cheaply. Han Lin of Asia Group pointed to strong engineering talent and a state capable of mobilizing enormous resources. But, in his words, cutting-edge technology also needs venture capital willing to absorb failures.

What This Means for Developers and Product Managers

Cheap open Chinese models will likely stay cheap, since competition among providers is pushing prices down. But the stability of the labs themselves is not guaranteed. If costs keep rising faster than revenue, companies may cut back on training their largest models or change strategy, for example by closing their weights or raising prices. So do not tie a product to a single model from a single provider, even if it is cheap today. Keep an abstraction layer over the API, test a fallback option, and estimate in advance what a migration would cost.

Caveat: some of the figures are Rhodium's estimates across 13 companies, and many facts, including Liang Wenfeng's remarks from the leaked recording, come from the Times' account. Still, the picture is clear: China's AI bottleneck lies not only in chips but also in economics.

Story

Key Questions Answered

What are the key details on Spending Is Outpacing Revenue?

According to Rhodium's forecast, AI infrastructure spending will exceed 1.2 trillion yuan ($193B) in 2027. For scale, China's AI infrastructure is only 15–20% of America's. The annual recurring revenue (ARR, meaning yearly income from subscriptions and paid access) of all Chinese AI models combined is about $10.7B. That is roughly one-tenth of what OpenAI and Anthropic report together. Kyle Chan of the Brookings Institution compared the situation to a treadmill that keeps speeding up, making each step harder.

What are the key details on Open Models and a Hardware Shortage?

China's bet on open models complicates monetization. In August, Z.ai released GLM-5.3 Flash and halved prices for two weeks, because the company wanted users to run the model on its servers. By November, as the Times reports citing OpenRouter data, about 20 competing providers were selling access to the same model for less than Z.ai itself. If the model weights are open, anyone can offer the model more cheaply than its author. Compute remains scarce and expensive.

Where the Money Shortage Comes From?

Private AI investment in China totaled $12.4B in 2025, against $285.9B in the US. This figure probably understates the real picture, because state funds also invest in AI. Foreign money has largely left, in part because the US bars American venture funds from investing in certain Chinese technologies. China still has strengths. Its researchers have repeatedly found ways to train and run models more cheaply.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…