A Tenth of the Revenue at Five Times the Multiple: Rhodium X-Rays China's AI Financing

The race between the United States and China in artificial intelligence is usually narrated through model releases, benchmark scores, and chip export controls. A report published Thursday by the Rhodium Group, “Examining China’s AI Financing,” shifts the lens to something less glamorous but arguably more decisive: where the money comes from, and whether it can keep coming. The picture that emerges is one of extraordinary momentum built on a strikingly thin revenue base — and a financing structure whose two load-bearing walls, equity issuance and state money, are both showing cracks.

The headline numbers

Rhodium’s survey of 13 listed Chinese companies — hyperscalers, telecom operators, frontier AI labs, and independent data center operators — projects China’s AI capital expenditure will roughly double this year, up 103% to 932 billion yuan (about $139 billion), and top 1.2 trillion yuan ($193 billion) in 2027. That is very strong growth. It is also, by Rhodium’s estimate, only around 15–20% of what the United States is investing in data centers this year — roughly $800 billion.

On the revenue side, the gap is even more dramatic. Total annual recurring revenue for all of China’s AI models comes to around $10.7 billion on the latest available data, which Rhodium places at roughly 10% of the recently reported levels of OpenAI and Anthropic. The individual figures inside that aggregate are sobering:

• ByteDance: ~$4 billion ARR — the largest single figure in the Chinese ecosystem
• Alibaba: ~$2.4 billion
• Z.ai (formerly Zhipu): $1.8 billion, per a Wednesday investor call; the company raised its year-end forecast to $3 billion, up from $2.4 billion
• Moonshot AI: ~$1 billion
• MiniMax: ~$800 million, up from $150 million in February
• DeepSeek: ~$500 million — the lowest among major Chinese labs, a striking fact for the lab whose open-source releases convulsed global markets in early 2025

The multiple problem

Valuations tell the more uncomfortable story. Rhodium calculates valuation-to-ARR ratios of roughly 46x for Z.ai, 50x for Moonshot, and a towering 163x for DeepSeek — against approximately 34x for OpenAI and 21x for Anthropic. Moonshot’s reported valuation has rocketed from $4 billion at the end of 2025 to $50 billion by August. “Valuations relative to revenue appear exorbitant for Moonshot and DeepSeek at present,” the report states flatly.

The timing matters. Anthropic is expected to go public on a U.S. exchange next month; OpenAI has pushed its own IPO into next year. Moonshot has reportedly filed confidentially for a Hong Kong listing, and DeepSeek is said to be preparing one as well. Public markets have already delivered a preview of how this can go: Z.ai’s shares briefly more than tripled over the summer before tumbling back to spring levels, and MiniMax has struggled to hold its IPO-day gains.

How China’s AI buildout is actually financed

The structural core of the report is a comparison of financing channels. In the U.S., the great enabler of the AI buildout has been debt: the five big U.S. hyperscalers saw combined free cash flow collapse from $191 billion last year to $14 billion in the first half of 2026, and responded by raising $163 billion through debt in the first half, mostly long-term bonds — a shift that has made bond markets and private credit central to the sustainability of the American boom.

China’s AI sector is financed differently. Ranked by scale and importance, Rhodium puts the channels in this order: operating cash flows, then equity issuance, then loans and bonds, then asset-backed securities, REITs, and financial leases. Bond issuance — the American workhorse — plays a comparatively small role, partly because China’s bond market still heavily favors state-owned issuers, which make up over 60% of it since 2022.

Equity is where the action is. Announced equity investment in China’s AI sector reached 282 billion yuan through August 21 — a historical record, more than double the 2023–2025 lows. Frontier labs alone tapped 179 billion yuan in equity financing in the first eight months of 2026, against just 9 billion yuan of PE/VC money in all of 2025. Zhipu (now Z.ai) raised about 4 billion yuan in its January Hong Kong IPO, followed by a 27 billion yuan private H-share placement in July, and this month closed another roughly $5 billion raise — its second mega-round in two months.

The state’s share of that equity is falling, but from a high base: direct government investment accounts for about 25% of announced AI equity this year, down from 30% in 2024–2025 but double the pre-pandemic average of 13%. And it is highly concentrated: more than 60% of equity investment in AI chips and servers comes from state-affiliated sources — government guidance funds plus mostly state-owned banks. Rhodium’s co-author Logan Wright argues this is by design: “Government funding has been helpful on the hardware side of the buildout of compute capacity, but similarly will probably balk at direct funding for the frontier labs.”

Cash flow stress runs through the system

China’s hyperscalers face the same arithmetic as their U.S. peers, with less room to maneuver. Combined free cash flow for Alibaba, Tencent, and Baidu swung from positive 170 billion yuan in 2025 to negative 16 billion yuan in the first half of 2026. Tencent booked about 51 billion yuan of AI-related prepayments inside Q2 operating cash flow — nearly matching its 59 billion yuan of quarterly capex. Alibaba, in its first new share placement since its 2019 Hong Kong listing, is raising HK$80 billion with 100% of proceeds earmarked for AI.

The cross-subsidy that makes this tolerable is weakening. Some 57% of hyperscalers’ 2025 revenues are linked to domestic consumption through e-commerce, entertainment, and advertising — exactly the part of the Chinese economy under the most strain. Alibaba’s e-commerce adjusted EBITA plunged 44% in fiscal 2026 amid the price war with JD.com and Meituan. Its newly separated AI app segment generated 3 billion yuan in Q2 revenue against a 14 billion yuan adjusted-EBITA loss. ByteDance’s Doubao app draws over 200 million daily users but under one million yuan of daily revenue, mostly e-commerce commissions — while Douyin converts 750–800 million daily users into over one billion yuan of daily ad revenue.

Why Chinese AI revenue is structurally low

Rhodium identifies a familiar cluster of causes: structurally lower pricing power, customers historically unwilling to pay for software, intense domestic competition, and — most consequentially — the open-weight strategy most Chinese frontier labs have embraced. When a third party downloads a model and serves it themselves, the lab that built it earns nothing. That choice turbocharged global adoption of Chinese models but hollowed out monetization. Moonshot and Alibaba are now pushing revenue-sharing agreements with major hosts of their open-weight models; Moonshot is in discussions with Microsoft, Amazon, and Google, with its revenue share reportedly running as high as 30%.

The pricing data underscores the gap. Per Artificial Analysis estimates cited by Rhodium, Chinese frontier models generally price at $0.04 to $0.50 per task, with only the best models reaching $0.50 to $1 — while top-tier Claude models command $2 to $4 per task and higher-end GPT models $1 to $2. DeepSeek, notably, achieved a reported gross margin on its API business by July close to Anthropic’s, through aggressive optimization of models and inference infrastructure — evidence that the problem is not unit economics at the API level but scale, pricing, and mix. Company-wide, Zhipu’s and MiniMax’s gross margins fell from 41% and 25% in 2025 to 26% and 18% in H1 2026 even as ARR surged.

The bottom line

Rhodium’s synthesis is that China’s AI expansion now hangs on two contingencies: continued strength in the equity market, and state support that is likely to remain focused on chips rather than frontier labs. “The financing gap means it will be far more difficult for Chinese frontier AI labs to scale sustainably,” Wright said. “They will be heavily dependent upon a favorable climate in the equity market — historically that’s not an easy bet in China.”

There are genuine offsetting signals. Chinese model usage has grown explosively from a low base; Z.ai just lifted its year-end ARR forecast to $3 billion; MiniMax grew ARR more than fivefold in six months; ByteDance reportedly secured a $29.6 billion offshore syndicated loan from around 30 banks in September to underwrite the single largest marginal capex increase in the ecosystem. The direction of travel on revenue is clearly up.

But the report’s core observation stands: China has built an AI industrial base that spends like a rival to the United States while earning like a tenth of one. Whether usage converts into revenue fast enough — before public-market patience, equity-market conditions, or state priorities shift — is now one of the defining questions of the next phase of the AI race.