I see a lot of people dunking on this and comparing it to 8 years’ worth of Claude subs spent to get inferior AI models and slower inference.
But I think the people doing these kinds of investments and experiments will be the long-term winners, for four reasons:
1- local AI will continue to get better. You’ll get a lot more intelligence for a fixed amount of compute and memory. We already have GPT-4-level LLMs running on laptops, so it’s only a matter of time before local AI catches up with the current frontier.
2- AI subsidies will come to an end. AI labs are currently burning cash to maintain market share (just look at Anthropic’s S1). They will eventually charge you the full price of the LLM tokens you spend. So if you’re a power user, don’t count on the $200 Claude subscription giving you thousands of dollars’ worth of tokens every month.
3- If you’re a startup, you should be wary of pumping your data into AI companies. They are incentivized to look for and clone successful applications. Even if they don’t directly look at your data, they have ways to mine useful information from your usage.
4- You don’t need the absolute frontier of AI for most tasks. Owning your own AI stack helps you explore the Pareto frontier and find the best per-task model. That knowledge and experience will be very valuable as AI continues to evolve.