XNews

Fin Argues Open-Source Models Won't Erode OpenAI and Anthropic Margins

Fin's AI semiconductor endgame analysis argues that cheaper open-source models such as DeepSeek will not prevent OpenAI and Anthropic from earning profits or buying GPUs. The analysis cites falling costs matching falling prices, a separate flagship market, and open-source models accounting for over 20% of token usage but only 5% to 10% of spending.

Event version 1.

Key insights

Costs Fall as Fast as Prices

Each new chip generation produces about twice the tokens from the same power, and software optimization cuts costs by half annually. A 1GW data center running flagship models can sell $80 billion per year while hardware amortization is only about $8 billion annually.

Cheap and Flagship Models Are Separate Markets

Open-source models compete on price only in the low-end segment. The flagship tier is contested only by OpenAI and Anthropic, where prices remain stable.

Open-Source Share Is Small by Spending

Open-source models account for more than 20% of token usage but only 5% to 10% of actual spending. Stronger models enable more long tasks, which will increase demand.

Source posts