The author analyzes the current AI race, arguing that big corporations are using high costs to outlast smaller competitors, but predicts a shift to flat-fee pricing and locally-run AI in the long term, advocating for government co-built data centers for sovereign AI infrastructure.
I made this image (with nanobanana 2.0) to illustrate what I think is happening in the current AI race. Right now, there's a heavy decline in quality and access to AI tools. Things like thinking budgets and usage quotas are getting heavily restricted and models are getting severely quantized in the native web UIs, coding and agentic usage. The big corporations are using the cost of running AI, in a vacuum, to starve out smaller competitors in the world's most expensive game of "loss-leader musical chairs". The truth is, the big corporations aren't bleeding cash overall - their record-breaking profits easily cover the bill - while smaller companies whose only product is AI can't keep up with them. But honestly? I think the big companies have the right strategy (*though I think switching to token-based usage on top of a flat-fee pricing while reducing quotas feels like the wrong move, it is likely a brief window of opportunity for a possible short-sighted profit - gambling the trust of loyal users*). Much like early SMS, internet access and phone calls, I believe the final price model will inevitably become a "flat fee only" for private users due to fierce market competition. However, we need diversity as well as room for smaller companies, upstarts and specialists, etc., with different kinds of innovations, approaches and models. We are early in the development of both the models and hardware. We desperately need these tech giants to spend billions building serious infrastructure so future AI doesn’t end up trained on a potato-powered toaster. Eventually, as technology matures and compute becomes cheaper, powerful AIs will run locally on our private computers - just like the tech from the first moon landing can now run inside a modern refrigerator. Unless, of course, the market is artificially held back to control access. In the long run, I hope to see governments (like the EU or individual nations within) co-build massive data centers with giants like Nvidia or Google. National states should own the physical buildings and the data for full sovereignty, while tech partners supply the hardware and pre-trained AI models. This infrastructure could then be used in tandem to both provide big tech models and to train open-source models for universities and public development, among other things, ensuring that the power of AI doesn't just stay behind corporate walls.
The article compares the current AI landscape to a musical chairs game, warning that rising costs and limited access are concentrating AI benefits among wealthy entities, and urging non-wealthy participants to act now before it's too late.
An opinion piece analyzing the potential economic and political consequences of AI replacing 20% of jobs, warning of a massive wealth transfer to AI companies and the creation of an oligarchy.
The article discusses the massive $725 billion capital spending battle in AI cloud infrastructure, questioning whether the market is becoming unmanageable and highlighting inductive analysis as the next major battleground.
The author reflects on mixed signals in the AI industry, noting high spending without proportional productivity gains and Anthropic's move to restrict Claude Code access while raising massive funding, questioning the direction of AI's revolutionary claims.