AI executive warns Chinese labs’ use of free open models could feed a new round of chip competition

Reflection AI CEO Misha Laskin is warning that the growing spread of open, freely available AI models may accelerate competition in hardware, not just software. In remarks reported by Yahoo Finance, Laskin argued that Chinese AI labs can use free models to start generating revenue abroad, and that this path can eventually translate into building and selling their own chips and data-center systems.

The concern hinges on a specific technology dynamic. Open models, especially those that can be downloaded and run with fewer licensing constraints than proprietary systems, allow smaller teams and research groups to customize and deploy AI capabilities faster. Laskin’s view, as characterized in the report, is that this speed can shorten the distance between “using the model” and “funding a broader stack,” including compute chips and the supporting infrastructure needed to serve customers.

In the same report, the discussion aligns with how NVIDIA describes its market outlook. NVIDIA’s public guidance, as characterized by Yahoo Finance, already “leaves China out” of its forward-looking expectations. That framing matters because it implies NVIDIA has been assessing the competitive and demand environment in the region separately from the rest of its business.

NVIDIA’s broader market depends on high-end AI accelerators used to train and run large language models. Those accelerators are often paired with large-scale data center deployments that require power, cooling, and networking designed for heavy computation. If rivals gain scale by selling AI applications based on open models, Laskin’s argument suggests they may eventually push downstream into the hardware layer, where switching costs and long-term procurement commitments can be significant.

The report also underscores the strategic tension around model openness. Open-model ecosystems can expand the availability of AI capabilities and reduce barriers to entry for new companies and developers. But they can also provide a training ground for engineering talent and operational learning that can later be applied to hardware-software optimization and manufacturing decisions.

Laskin’s comments also arrive at a time when investors have focused closely on how quickly AI demand translates into new silicon investment. The path he describes is not simply that open models undercut proprietary model sales, but that they help create customers and revenue flows that can finance chips, co-designed systems, and data centers intended to serve those customers efficiently.

Even if NVIDIA does not dispute that chip competition exists globally, the immediate takeaway from the report is about trajectory. By highlighting an endgame that moves from open model usage to hardware commercialization, Laskin is effectively warning that the competitive landscape could widen beyond the firms that currently dominate training and inference platforms.

Still, the public record behind this specific exchange leaves gaps. The Yahoo Finance write-up, based on the information provided here, does not lay out evidence about which particular labs are planning chip products, timelines for any such launches, or whether those labs would rely on existing manufacturing partners or pursue new production routes. NVIDIA itself was not shown, in the materials included here, responding directly to the claim in a quoted statement or filing.