The AI safety debate just got its real deadline — and it's being set in Beijing, not San Francisco.
The Summary
- Eclipse Ventures CEO Lior Susan argues the AI risk debate needs to shift from retreat to collaboration if the US wants to maintain its lead over China
- Data center expansion and "physical AI" could become the next major driver of US economic growth, but only if the industry can win over local communities
- The real risk isn't moving too fast on AI — it's moving too slowly while competitors sprint ahead
The Signal
The safety-vs-speed debate in AI just ran headfirst into geopolitics. Eclipse's Lior Susan is making the case that the US can't afford to pump the brakes on advanced AI development while China keeps its foot on the gas. His thesis: collaboration between industry, government, and communities is the only path that balances safety concerns with competitive necessity.
This isn't the usual Silicon Valley chest-thumping about innovation. Susan is pointing at something concrete: physical AI as the next major economic engine. Not chatbots. Not content generation. Robots, autonomous systems, and AI that moves atoms, not just bits. That requires infrastructure — specifically, the data centers that local communities are increasingly pushing back against.
"The tech industry needs to do a better job explaining the benefits of data centers to local communities."
The data center fight is where the AI race gets real. Every advanced model, every training run, every agent deployment needs compute. Massive amounts of it. But NIMBYism meets AI infrastructure in town halls from Virginia to Arizona, where residents see power draws and cooling towers, not economic opportunity. Susan's argument is that the industry has botched the messaging. Here's what's actually at stake:
- Data centers create high-paying technical jobs in regions starved for them
- They anchor supply chains for chips, cooling systems, and grid infrastructure
- Physical AI built on that compute could reshore manufacturing and logistics jobs
The China framing isn't fear-mongering. It's clock-setting. Chinese AI labs don't have to navigate the same regulatory gauntlet or community approval processes. They have state backing, streamlined permitting, and a national mandate to win the AI race. The US lead in frontier models is real but narrowing. The lead in physical AI infrastructure — the unsexy stuff that determines who can actually deploy agents at scale — is up for grabs.
Susan is calling for a reframe: stop treating AI development as reckless and start treating it as strategic. That doesn't mean ignoring safety. It means building safety frameworks that don't hand China a structural advantage. The debate over whether to slow AI progress is effectively a debate over whether the US wants to set the rules for the agent economy or follow someone else's.
The Implication
Watch where data center projects get approved or blocked over the next 12 months. That's the real leading indicator of whether the US can sustain its AI infrastructure advantage. If local opposition wins out repeatedly, expect more frontier AI deployment to shift offshore — not to China directly, but to friendlier jurisdictions with fewer roadblocks.
For builders in the agent space, this is a forcing function. If you're designing agents that need heavy compute or physical presence, start thinking about where that infrastructure will actually exist. The US has the technical lead today. Whether it keeps it depends less on algorithms and more on whether we can build the unsexy stuff fast enough.