AI is no longer just something that happens in a cloud-connected data center. It is moving into the devices in your pocket and even onto spacecraft. As this technology spreads, it is hitting physical limits—both in the amount of heat it generates and in the communities where it is built.
Smartphone makers are now squeezing AI capabilities directly onto new chips. Qualcomm’s new Snapdragon 8 Elite chips allow phones to run a personal digital assistant or improve photos without sending your data to a remote server. Meanwhile, space startups like AstroForge are using similar AI models to help spacecraft navigate and solve technical problems autonomously, reducing their reliance on radio signals from Earth. But these advances rely on massive server farms—giant warehouses filled with computers—which have sparked significant local opposition. Data center projects worth $68 billion were disrupted by local opponents in just the second quarter of 2026 alone, as residents raise concerns about energy costs and transparency.
The physical limits of intelligence
An AI model is essentially a massive list of instructions that helps a computer guess what to do next. Inside the software are parameters—think of these as millions of tiny knobs that the system twists during its training phase to learn how to recognize a voice or steer a spacecraft. A neural network is the underlying architecture that organizes these knobs into layers, mimicking how a brain might process signals to make a decision.
Traditionally, these models were too large to run on a phone, so your device had to send information to a distant computer. Newer mobile chips, like those from Qualcomm, can now handle models with 200 million parameters locally. For more complex tasks, some chips use a mixture-of-experts approach, where the model is large—up to 30 billion parameters—but it only turns on the specific knobs needed for the current task to save energy. Meanwhile, in massive data centers, the density is so high that air cooling fails. Engineers now use liquid cooling, pumping fluid directly over components to absorb heat, allowing server racks to operate without fans even as they push toward 1 megawatt of power usage.
The push toward AI is creating a divide between what happens on our devices and the infrastructure required to build those experiences. While we gain smarter phones and more capable machines, we are also facing a bottleneck in how we physically build these systems. The resistance from communities suggests that the future of AI won't be decided only by who has the best technology, but by whether companies can build a foundation of trust with the public. We are moving toward a world where AI is everywhere, but we are still figuring out the environmental and social costs of keeping that intelligence powered and cooled.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy