The Inevitable Death of Anthropic and OpenAI
Tired of annoying my colleagues I come here. I think OpenAI and Anthropic are destined to fail if they continue on the path they're on and I don't see them changing.
TL;DR
AI will converge to local machines and that's a good thing for everyone, except for OpenAI, Anthropic and their investors.
Long version
When OpenAI released GPT-3 everyone thought that the model was this huge moat, that OpenAI was going to dethrone Google. Very quickly we realized that models were closer to a commodity rather than a moat. Models like Alpaca and relatives started coming out and leaking. People started distilling models from the frontier ones getting good results. Still they continue to compete on this front. Why? The previously unheard size of demand, the step changes of increasing the number of parameters, the push from investors are some reasons.
Remember the jump from GPT-3 to GPT-3.5 or from GPT-3.5 to GPT-4? Those changes were enablers of new applications before not possible. Now the changes from Opus 4.6 to 4.7 to 4.8 feel more going from iPhone 16 to 17 rather than from iPhone 4 to 5.
Meanwhile, Gemma 4 is a huge step change from Gemma 3. DGX Spark and Mac minis enable reasonable local models. Nvidia seems to be going in a direction of running trillion-parameter models on your workstation. Extrapolating this, to me it seems like companies and consumers will weigh the two options and find out that the local model sounds really good. I don't know if this has been your experience, but personally I only use the frontier models for coding. For my agentic workflows I've been using Gemini-3-flash (not even the latest version). It's a ver good balance between cost, stability and performance. On the last WWDC Apple showed some stuff already running on device. This already happened with mainframes and terminals to commodity hardware and personal computers.
The fact is that it's unconvenient to use an API you don't control and it's risky from a business point of view to depend fully on another company's wellbeing for your company or productivity to be healthy. The exorbitant prices of tokens.
Now they are ripe for an IPO, just on the brink (in my opinion) of the last year of explosive growth for these companies. This strikes me as extremely opportunistic, given the slowing pace of improvement in these models. I think Anthropic and OpenAI are on the peak just before the void. They want to cash out before this becomes a reality. They didn't invest on other product lines because worse models meant losing customers and investors support. We're all the way back on early 2000 weighing clicks as the metric of success for companies that lead to a bubble. Tokenmaxxing sounds exactly like that. Hearing Jensen Huang say that people should be spending 250k of tokens is not only ridiculous but just blatantly obvious considering they're the only ones profiting from AI so far including the energy companies (see the last link).