Brief IA
›

AI: Chinese Groups to Spend $139 Billion

Models & LLM·Tom Levy·

AI: Chinese Groups to Spend $139 Billion

AI: Chinese Groups to Spend $139 Billion
Key Takeaways
1Rhodium forecasts $139 billion in investments in AI infrastructure from 13 publicly traded Chinese companies this year.
2According to Stanford, Chinese private investments in AI reached $12.4 billion in 2025.
3According to OpenRouter, around 20 providers offered the Z.ai model at a lower price than its creator in November.
Why it matters : revenues from the affected labs would account for less than 8% of the projected investments this year, according to Rhodium.

Investments in AI infrastructure from 13 publicly listed Chinese companies are expected to reach $139 billion this year, a 103% increase, according to Rhodium Group. The revenues from their labs would account for less than 8% of this amount. Funding poses a challenge alongside access to chips.

Chinese Private Capital Remains Limited

In 2025, private investments in AI totaled $12.4 billion in China, compared to $285.9 billion in the United States, according to Stanford. However, the university notes that the Chinese figure is likely underestimated: public-oriented funds also contribute capital. Foreign funding has dried up, particularly because the United States prohibits its venture capital investors from financing certain Chinese technologies.

DeepSeek has reportedly raised over $20 billion, including $15 billion in a round supported by Tencent and CATL. Few of its competitors have comparable resources. In the markets, the Chinese STAR 50 index, akin to the Nasdaq, has fallen by 43% since the end of July.

Han Lin, director for China at the Asia Group, highlights the exceptional skills of the country's engineers and the government's ability to mobilize substantial resources. He reminds, however, that advanced technologies also require investors who can bear failures.

Revenues Struggle to Keep Up with Investments

By 2027, Rhodium Group projects investments exceeding 1,200 billion yuan, or approximately $193 billion. Its estimates, published in September, focus on the AI infrastructure of the companies in its sample. The expected amount this year corresponds to 932 billion yuan.

The annual recurring revenues of all Chinese models are estimated at around $10.7 billion, roughly one-tenth of the combined revenues of OpenAI and Anthropic. Rhodium also assesses the development of Chinese AI at 15% to 20% of the American level. Alibaba, Baidu, and Tencent, which fund several major labs in the country, are now showing negative cash flows.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Kyle Chan, a researcher at the Brookings Institution, describes a financial pressure that is intensifying as these companies continue their efforts. According to Logan Wright, a partner at Rhodium and co-author of the study, funding constraints are likely to weigh more heavily in the long term than those related to chips.

Z.ai Faces Cheaper Competitors

After launching GLM-5.3 Flash in August, Z.ai offered a 50% discount for two weeks to attract users to its servers. In November, data from OpenRouter indicated that about 20 providers were selling access to the same model at a lower price than its creator.

Chinese researchers have, however, repeatedly found ways to reduce the cost of training models. They have also developed less expensive methods to operate them.

Restrictions and Chip Costs Weigh on AI

Access to advanced semiconductors, central to discussions about Chinese AI, remains constrained. This week, Washington took measures against a way to circumvent its restrictions: renting overseas data centers equipped by Nvidia. According to Tencent and Alibaba, these new measures would slow down AI development.

Computing also represents a high expense: sellers have offered the Nvidia DGX B300 server for nearly $3 million on Goofish, a Chinese second-hand sales platform. This price is about eight times the rate in the United States.

In a conversation with investors, a transcript of which leaked, Liang Wenfeng, CEO of DeepSeek, places Huawei's chips about two years behind those of Nvidia. He specifies, however, that DeepSeek would need four Huawei 950 chips to match one Nvidia GB300. He also states that the company cannot afford to finance the training of the largest models currently available. DeepSeek did not respond to a request for comment.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.