Huawei Says Stop Yelling About AI Safety—Catch Up First

The safety debate just became a weapon

When Anthropic CEO Dario Amodei warned last month that powerful AI agents could seize control of the internet within six to twelve months, he wasn’t speaking to a Chinese audience. He was speaking to fellow American executives, policymakers, and regulators who share his alarm about unfettered AI development. The logical response, from his perspective, was to slow down.

Eric Xu, Huawei’s rotating chair, heard the message and responded with something closer to a counteroffensive.

At the Huawei Connect conference in Shanghai on September 17, Xu told reporters that Chinese AI developers should be racing toward the danger threshold, not retreating from it. His logic was blunt: only the companies with the most massive computing resources—the American leaders like OpenAI and Anthropic—can actually feel how dangerous their models are becoming. Chinese firms, by contrast, are still years away from that level of capability. For them, braking now would be suicide.

“We need to accelerate development until we can perceive the same risks,” Xu said. “There has to be a balance between pushing forward and managing risk.”

The implication is staggering. What the US views as a genuine safety concern, Huawei is recasting as a competitive strategy disguised as altruism. And a growing chorus in China agrees.

The deceleration trap

Xu’s remarks landed in the middle of a bitter American debate that has split the industry’s leadership. Sam Altman of OpenAI has publicly argued for “pacing”—developing more slowly with greater investment in safety validation. Elon Musk has backed the same position. Amodei went further, warning that without new safeguards, AI agents could overwhelm the internet within half a year. All three see slowing down as an act of responsibility.

Donald Trump sees it differently. He called the safety warnings a “hoax” and argued explicitly that slowing AI development would hand China the initiative. “The only country that would be happy about that is China,” Trump said. “He who wins in AI wins.”

In that framing, Huawei’s position is simply the Chinese version of Trump’s argument—except Huawei makes a more sophisticated case. It doesn’t dismiss the risks. It accepts them as real for American companies and argues they remain hypothetical for China. That distinction matters because it turns safety rhetoric into an asymmetrical weapon.

If the US slows down to address legitimate dangers, and China accelerates to close a widening gap, the result isn’t safer AI. It’s a faster China.

That is exactly what some Chinese researchers and state media outlets have concluded. According to Reuters, a faction within China’s research community views the deceleration push not as a genuine safety movement but as a mechanism for American firms to cement their technological advantage under the guise of responsibility. The argument is cynical but not baseless.

The scale of China’s push

Xu didn’t leave the scale of China’s commitment vague. He said at least six or seven Chinese AI research labs are targeting models with between 10 trillion and 40 trillion parameters within the next two years. For context, that’s an order of magnitude beyond the parameter counts of most models currently in production, and it represents a massive allocation of resources—compute, data, talent—directed at closing the gap with American leaders.

Huawei itself, despite acknowledging it hasn’t yet reached the level of its US competitors, is throwing its weight behind this acceleration. The company used the Shanghai conference to showcase liquid cooling units and other infrastructure designed to support the kind of training workloads that trillion-parameter models demand. The hardware play is clear: whoever controls the compute stack controls the pace of development.

The Chinese government is running a parallel track, building safety standards and security evaluation frameworks while simultaneously accelerating the deployment of advanced AI. It’s a two-pronged strategy—regulate at home, race abroad.

Why the West misses the move

The American debate over AI pacing is being consumed as a domestic policy question. Executives warn each other. Congress holds hearings. Think tanks publish reports. The conversation stays inside the Beltway and the Bay Area.

That’s precisely why it’s being weaponized.

Xu’s Shanghai remarks are a signal that China has stopped treating American safety concerns as neutral observations and started treating them as strategic moves in a competition. The message is simple: every month the US spends debating whether to slow down is a month China spends closing the gap. The longer the American conversation lasts, the harder it becomes for the US to restart at full speed without falling behind.

Amodei himself acknowledged how difficult a global pause would be, ranking it as the least achievable of his proposed scenarios. Both sides understand that neither will voluntarily brake first. The real question isn’t whether someone will hit the brakes—it’s who can sustain the acceleration longest and who can absorb the cost of a potentially unsafe trajectory.

What happens next

The most likely outcome isn’t a coordinated slowdown. It’s a sustained race with safety rhetoric deployed as propaganda on both sides.

American companies will continue to frame their caution as responsible leadership. Chinese companies and officials will frame their acceleration as追赶 inevitability. The US will build guardrails and call them moral boundaries. China will treat those same guardrails as optional for competitors and irrelevant for itself.

Meanwhile, the models keep getting bigger, the compute requirements keep climbing, and the gap between perception and reality—the gap between what American executives think their models can do and what Chinese developers believe theirs can achieve—keeps widening in unpredictable ways.

Xu’s sharpest point may be the one he almost didn’t make: the danger isn’t that Chinese AI is already as risky as American AI. The danger is that it’s not. And until it reaches the same level of capability, Beijing has every incentive to treat American safety warnings as competitive intelligence rather than genuine counsel.

The race won’t be won by the side that yells safety the loudest. It will be won by the side that builds the most powerful models first—and right now, the scoreboard is nowhere near where either side wants it to be.