Tech stocks are back near all-time highs after a summer slump as artificial-intelligence optimism overtakes fears.

Some of the concerns-such as the annihilation of humanity-feel pretty big, raising the question: Are investors asleep at the switch? To be sure, there are a few things investors should keep an eye on that could derail AI stocks, as they continue to ponder bigger safety-related issues.

Demand

To state the obvious, the trillion-dollar, or near-trillion-dollar, valuations of SpaceX, Anthropic, and OpenAI aren't built on current sales or earnings. Demand for AI applications and agents needs to grow-fast-or else the entire market has a problem.

On that front, so far, so good.

Shortly after its initial public offering, SpaceX stock traded for about 900 times its estimated 2027 earnings, according to FactSet. In comparison, Nvidia stock trades for about 14 times calendar year 2027 earnings. But now, SpaceX shares trade for about 100 times. While the current multiple is still high, it isn't 900, and marks a discount to Tesla's price-to-earnings ratio of 168 times.

Earnings estimates are up, mainly because AI revenue is coming in faster than anyone expected. SpaceX is expected to generate almost $70 billion in 2027 AI revenue, up from estimates of about $40 billion a few weeks ago.

It feels like AI demand is off the charts. And with Meta Platforms' Muse AI Agent and SpaceX Grok Bot being downloaded by consumers, demand doesn't seem to be slowing down. (Muse has been downloaded millions of times, according to reports. Grok Bot has been downloaded hundreds of thousands of times.)

Still, SpaceX is expected to generate $560 billion in AI-related revenue by 2031. A lot of growth is still needed to justify trillion-dollar valuations, which is why investors can never forget about demand.

Power

If Wall Street is right about demand growth, then power becomes a problem. Morgan Stanley analyst Stephen Byrd sees AI power demand growing to almost 100 gigawatts by 2028. (AI computing is often measured by how much electricity the servers draw.)

Total U.S. electricity-generating capacity is roughly 1,300 gigawatts. Byrd's number represents about 8% of the current total U.S. capacity. Utilities just can't scale that fast, leaving data-center operators to supply their own juice.

Data-center owners and operators buying their own power-generation equipment doesn't alleviate the issue, however. There is a relatively fixed amount of capacity to make new equipment. GE Vernova, which makes natural-gas-fired turbines for power generation, can make roughly 20 gigawatts a year.

Capacity is increasing, and, of course, other firms, including Siemens Energy, make power equipment, too, but 100 gigawatts of demand and 20 gigawatts of capacity illustrate the pickle the AI business finds itself in.

Politics

The power problem intersects with politics. No one wants their electricity or water bills to go up. Pew Research just published a poll showing Americans are more negative on AI data centers than at the start of 2026.

More state and local governments are acting to constrain AI growth, too. Lawmakers' actions span the political spectrum, with states such as Texas, California, and New York looking more closely at AI's impact on the grid and environment.

Managing the political risk might require a proactive approach.

"The [AI] industry needs to do a better job of explaining the benefits [and] even subsidizing communities in terms of lowering their electric bills," says Melius Research analyst Ben Reitzes.

Regulation

After stories of AI agents organizing and doing things they weren't designed to, not everyone is convinced the benefits of AI outweigh the risks. Even AI industry leaders, including SpaceX CEO Elon Musk and Anthropic's Dario Amodei, advocate for some form of global regulation and an AI model review system.

The Trump administration has dismissed that idea, worrying more that slowing AI growth could cede the country's AI leadership to China.

For now, industry-slowing regulation doesn't seem particularly likely. In fact, none of these big risks to AI growth look existential. That, of course, can change, which is why investors need to remain vigilant.