The Return of the S Curve

Cognitive dissonance made manifest

This is the moment.

The second derivative is clear.

Hypothetical line of best fit to the frontier. Probably won’t flatten out this quickly…

GPT5 is here and the rate of change is clearly slowing.

The market wasn’t sure what was more disappointing, the confirmation of what we’ve been saying on this topic for a year…or this chart crime.

Models are getting better, of course, but just as the scaling law promised and foretold, we've reached the point where the the marginal extra OOM investment in compute, data and energy required to hill climb the intelligence curve is uneconomical.

As a reminder, this is a natural consequence of the math we’ve been highlighting for years.

The S Curve is Here

Exponentially bigger models need exponentially more compute to hold and run them.

@ramez

They need exponentially more data to train them.

@ramez

@ramez

Which means exponentially more energy to train and run.

The Energy Cost of Reasoning: Analyzing Energy Usage in LLMs with Test-time Compute

The Compute Burden

Last time I checked, electricity wasn’t free and the US wasn’t making exponentially more of it.

Remind me, what happens to the price of something when the demand goes up exponentially and the supply stays constant?

(Seems like a good moment to remind you that every solar panel has silver in it…)

Seethe, Cope and Opportunity

What you will see over the next 12m will be a combination of “seethe”, “cope” and, yes opportunity.

Seethe from the hype machine linkedin posters, who now have to make breathless threads of "You will never believe what the new LLM can do!" even though their audience *knows* that the returns aren't infinitely positive and increasing.

Cope from the dooomers, who now have to move the goalposts to preserve the eschatological narratives that finance their NGOs and group homes.

Opportunities to the wise merchants. Who understand there are still insanely positive and convex economic returns in the gap opened up by the concave returns to scaling models.

Invest (and live) accordingly.

Till next time.

Disclaimers

Charts and graphs included in these materials are intended for educational purposes only and should not function as the sole basis for any investment decision. As noted the views in this portfolio are not intended to maximize return and any use in a portfolio ought be considered in the broader context of your portfolio and investment needs and obligations. Please consult with a registered investment advisor if you would like financial advice, that’s their job.

This letter does not constitute an offer to sell, a solicitation of an offer to buy, or a recommendation of any security or any other product or service by Rose, Campbell or any other third party regardless of whether such security, product or service is referenced in this brochure. Furthermore, nothing in this website is intended to provide tax, legal, or investment advice and nothing in this website should be construed as a recommendation to buy, sell, or hold any investment or security or to engage in any investment strategy or transaction. You are solely responsible for determining whether any investment, investment strategy, security or related transaction is appropriate for you based on your personal investment objectives, financial circumstances and risk tolerance. You should consult your business advisor, attorney, or tax and accounting advisor regarding your specific business, legal or tax situation.