losnappas blog https://losnappas.github.io/ A blog built with Quarto quarto-1.8.26 Mon, 16 Mar 2026 22:00:00 GMT The Taalas Proposition losnappas https://losnappas.github.io/posts/the-taalas-proposition/ Taalas is building “inference machines” — ASICs designed only to run models, not train them. They just take finished weights and execute inference, potentially much faster and cheaper than GPUs.

That could create a strange market dynamic. One company might spend heavily to train a model on GPUs, only for another to take the weights and run the model far more efficiently on dedicated inference hardware.

This matters because training is a huge upfront cost, but the payoff usually comes later, when the model is deployed and serving users. If someone else can take the trained model and offer the same capability at much lower cost, they may be able to undercut the original creator on price or simply earn better margins.

If you release the weights, you may be giving away the only asset that lets you recover the training cost. And if another company is better at serving those weights cheaply, the economic benefit may shift away from the team that built the model and toward the team that operates the inference infrastructure. Who’s left to pay for the training?

]]> opinion taalas https://losnappas.github.io/posts/the-taalas-proposition/ Mon, 16 Mar 2026 22:00:00 GMT The AI Company Win Condition losnappas https://losnappas.github.io/posts/the-ai-company-win-condition/ What do I mean by a win condition? I mean whatever makes an AI company succeed financially in the long run.

Condition 1: Fast Takeoff

The dream of any AI company, as well as anyone, is that AI starts recursive improvement, becoming smarter than human beings can imagine. Assuming it aligns (misalignment being the horrible loss condition), there is no reason to be against this outcome. It is technological singularity, in a nutshell.

Condition 2: “Go Make Money for Me”

The AI doesn’t become infinitely smart, but it is clever enough to make more money than it costs to run, by itself.

For example, the AI company tells their AI to “make money,” and it succeeds. Now, what happens to the worth of the AI company? This is a general money-making machine, so you can imagine using it as leverage for your loans. Who wouldn’t give you a loan, with knowledge you have access to a machine that can make more money than it consumes? The AI company becomes like the Federal Reserve, with the ability to print money. Money itself turns obsolete.

Condition 3: The Slow Grind

Currently, AI companies burn money on R&D and capital expenditure (CapEx, like buying GPUs). If the previous conditions seem out of reach, I would assume the AI company has to start reducing R&D and CapEx costs, improving token/$ cost, and expanding their market reach.

For an AI researcher, I imagine this would be the dullest outcome. Away from building toward singularity, and more toward pleasing market forces.

Conclusion

I’ve described three scenarios, and I’m sure there are more. It wouldn’t be surprising if condition two was the current playbook, as how else are the companies getting such big investments. There is hype, but is it unfounded, as “AI bubble” folks claim?

]]> opinion https://losnappas.github.io/posts/the-ai-company-win-condition/ Thu, 12 Feb 2026 22:00:00 GMT