China’s AI Models Near U.S. Level as Valuations Diverge

At the “Investing in the AI Economy” forum during the fifth Global Financial Leaders’ Investment Summit on Wednesday, Duane Kuang, founding managing partner of Qiming Venture Partners, said China’s AI model capabilities were very close to those of the United States.

At the same forum, Silver Lake co-chairman Kenneth Hao said: “Frontier models can help the top 20 percent of employees on the team increase productivity by 100 times in research and development.” He said the remaining staff are better suited to affordable models.

Kuang said the valuation gap between large AI model companies in China and the United States reflects the business, revenue and profitability of Chinese AI companies, not meaningful differentiation among their open-source models. Similar model performance does not create similar company value, because revenue quality, operating economics and profitability determine what investors can discount.

What explains the valuation gap?

Kuang therefore placed the valuation difference in Chinese AI businesses rather than in the models themselves. The forum account supplied no company revenue, funding or profitability figures, so this remains an investor judgment rather than a measured comparison. Chinese model developers still need to show how technical parity produces durable cash flow.

Comparable models do not guarantee comparable company economics.

Why does edge AI matter to chip demand?

As AI models become smaller and edge computing grows more powerful, more computation can run offline on devices. Kuang said this makes edge AI chips a focus of attention, shifting some AI processing away from dependence on remote computing.

For semiconductor suppliers, that change moves demand closer to finished products. It raises the importance of chips able to perform AI tasks within devices rather than sending every task to a distant data center. The commercial test is whether smaller models and local processing create enough demand across devices to support a wider silicon market.

Kuang offered no process nodes, wafer sizes, capacity figures or shipment figures for that opportunity. His point described a direction of travel, not a quantified production roadmap, and the scale of any chip demand remains unstated.

Edge computing can expand the deployment base, but the source did not quantify the market.

How do returns and token pricing develop?

Hao expressed confidence that AI will bring revolutionary changes, while separating two levels of use. His productivity claim makes model quality an allocation choice: frontier systems address the highest-value work, while lower-cost models serve a broader workforce.

He expects more rational allocation and differentiated pricing for token economics this year. In his view, return on AI should naturally follow “all-in” investments, making the pace of deployment urgent rather than something companies can defer indefinitely.

Premium models target specialized gains, while affordable models determine wider adoption.

Why is overnight business transformation unrealistic?

Two Sigma co-founder and co-chairman David Siegel said expecting large language models to transform business overnight is unrealistic because people deploy the technology. Model performance establishes potential, but people and operating processes determine how much value an organization captures.

Insight Partners managing director Deven J. Parekh sharpened that implementation point. He said management ought to restructure existing product architectures for a product-driven world, because AI replaces not only other software technology budgets but also part of human resource budgets.

This changes the commercial pitch for AI suppliers. A deployment can draw on both software technology budgets and human resource budgets, which enlarges the potential budget pool but also raises the standard of proof. AI must work inside products and workflows before enterprises can convert capability into recurring revenue.

For chip suppliers, adoption will depend on those product decisions as much as model benchmarks. The speakers provided no rollout timetable, and Hao’s 100-times productivity claim remains a forecast from a technology investor rather than a measured industry result.

Model capability attracts investment; deployment and budgets determine revenue.

Who gains as China closes the model gap?

The forum connected three shifts: Chinese models approaching U.S. capability, smaller systems moving more AI work onto devices, and pricing that separates frontier use from affordable mass deployment. Each shift favors a different part of the market.

Model developers must close the profitability gap Kuang identified. Edge chip suppliers gain attention if offline processing expands device demand, although the discussion offered no capacity or shipment target. Enterprise buyers will determine the pace by restructuring products and deciding which work receives premium models.

The near-term contest should center on who converts comparable model capability into revenue at sustainable prices, rather than who alone publishes the strongest system.