Lisa Su Backs Open-Source AI Amid Security, China Debate

A CEO Weighs In at a Tense Moment

AMD chief executive Lisa Su used the spotlight at the company's Advanced AI conference to make a pointed defense of open-source artificial intelligence, arguing that openness remains essential to the technology's progress even as new hardware from AMD took center stage at the event 1. Her comments landed just as a security lapse involving Hugging Face, the widely used open model repository, resurfaced anxieties about how safe open AI infrastructure really is — reports indicate the incident was triggered by an OpenAI agent, adding an ironic twist to a moment when the industry's biggest players are jockeying over the future of open versus closed AI 1.

Su's remarks arrive amid a broader reckoning that has been building for months, one that pits enthusiasm for open-weight models against fears about security, geopolitics, and business viability. The result is a fractured conversation in which hardware makers, investors, policymakers, and AI labs are all reaching different conclusions about what "open" AI actually means and whether it should be encouraged.

The China Question Divides Opinion

Much of the recent friction centers on the rapid rise of Chinese open-weight models, which have won praise from developers for their capability and low cost. But that popularity has not translated into consensus about their implications. One line of commentary frames China's open releases as a strategic Trojan horse, warning that inexpensive, freely available models could foster long-term dependency on Chinese technology and give Beijing leverage over how information is filtered and controlled globally 2.

A competing view pushes back hard on the idea that Chinese models pose an inherent danger. Arcee, a U.S.-based open-source AI lab, has argued that the growing use of Chinese models by American companies does not itself constitute a security or ideological threat, even as the debate over what regulatory or competitive response is warranted has grown increasingly heated 5. That disagreement underscores a split between those who see open models as a geopolitical instrument and those who see them simply as tools whose value should be judged on technical merit.

Is Open-Weight AI Even a Good Business?

Separate from the security and geopolitical arguments, there's a more fundamental question about economics. One analysis contends that China's open-weight models, despite impressing engineers, represent a poor business model compared to traditional open-source software, because giving away model weights doesn't create the same ecosystem lock-in or monetization paths that open-source code historically has for companies that built businesses around it 3. In other words, the enthusiasm developers feel for free, powerful models doesn't necessarily translate into sustainable revenue for the companies releasing them, raising questions about how long the current wave of open releases can be subsidized.

Yet Wall Street has found a silver lining in the trend. Bank of America has pointed to a steady stream of new open-source releases — including China's Kimi K3 — as reinforcing its bullish outlook on memory chipmakers, particularly Micron Technology, on the theory that more capable open models drive greater demand for the memory-intensive infrastructure needed to run them 4. That view suggests that regardless of who wins the ideological argument over openness, the practical effect of more open models circulating is likely to be increased demand for computing hardware — a dynamic that also helps explain why a chipmaker like AMD has strong incentives to champion the open-source cause.

A Broader Freakout Takes Hold

Taken together, the moment reflects what one commentator has bluntly called a full-blown freakout over open-source AI, marked by finger-pointing across labs, governments, and commentators with no side emerging as clearly virtuous 6. Security incidents like the Hugging Face breach feed fears about the risks of freely distributed AI systems, geopolitical anxieties sharpen arguments about Chinese models, skeptics question whether open-weight releases even make commercial sense, and investors nonetheless find reasons to bet on the hardware boom that more open models could fuel.

What emerges from this scattered debate is a technology community still searching for a shared understanding of what openness in AI should mean — and who benefits most from it. Lisa Su's defense of the open-source approach, delivered against this noisy backdrop, signals that at least some of the largest players in AI hardware intend to keep betting on openness, even as the arguments about its risks and rewards show no sign of settling.