If the models are more efficient, the tasks that still need a server will get the same result at a lower cost. OpenAI can also pivot to building more local models and license them to device makers, if it wants.

The finances of big tech companies isn’t really relevant anyway, except to point out that Ed Zitron’s arguments are not based in reality. Whether or not investors are getting stiffed, the bad outcomes of AI would still be bad, and the good outcomes would still be good.

I wrote the article, Ed said that in the linked blog post: “There Is No Real AI Adoption, Nor Is There Any Significant Revenue - As I wrote earlier in the year, there is really no significant adoption of generative AI services or products.”

There is a pretty clear path to profitability, or at least much lower losses. A lot more phones, tablets, computers, etc now have GPUs or other hardware optimized for running small LLMs/SLMs, and both the large and small LLMs/SLMs are becoming more efficient. With both of those those happening, a lot of the current uses for AI will move to on-device processing (this is already a thing with Apple Intelligence and Gemini Nano), and the tasks that still need a cloud server will be more efficient and consume less power.

I wouldn’t really trust Ed Zitron’s math analysis when he gets a very simple thing like “there is no real AI adoption” plainly wrong. The financials of OpenAI and other AI-heavy companies are murky, but most tech startups run at a loss for a long time before they either turn a profit or get acquired. It took Uber over a decade to stop losing money every quarter.

OpenAI keeps getting more funding capital because (A) venture capital guys are pretty dumb, and (B) they can easily ramp up advertisements once the free money runs out. Microsoft has already experimented with ads and sponsored products in chatbot messages, ChatGPT will probably do something like that.