Currently, AI models are trained using GPUs. In the future though, Generative AI will probably require its own specialized ASICs to achieve the best performance. This happened with bitcoin mining a few years ago and is also the reason big tech companies are making their own CPUs now.
Since there are only a few companies on the planet capable of producing these chips in bulk, the government could easily place restrictions on the purchase of AI hardware. This would control who has access to the best AI.
Only the government and a few permitted parties have access to the best AI. Everyone else would use worse AI that, while still good enough for most people, could be detected by the government. The government could use their superior models to easily detect whether a post is AI-generated, for example, and provide that insight as a service to citizens.
Effectively, the government becomes the sole purveyor of truth, as opposed to that power being in the hands of whoever can afford the biggest computer.
It’s more a matter of cost efficiency than performance. That is especially critical for cryptomining where your only means for profit is by competing against the cost of electricity on an hour to hour level. That’s much less the case for training AI. They’d still use GPUs, they’d just spend more money on electricity and cooling than is ideal, furthering climate change and water insecurity. If you want to regulate AI it would be a lot more efficient to just regulate AI.