Keep pulling the thread on Dylan Patel.
Dylan Patel's technology analysis firm increased its spending on AI from tens of thousands of dollars in the previous year to a current annualized run rate of $7 million.
Dylan Patel's firm signed an enterprise contract with Anthropic and is spending $7 million per year on Claude Code, which represents over 25% of the firm's $25 million salary expense.
Anthropic's annualized revenue run rate has grown to between $35 billion and $40 billion.
Anthropic's gross margins are estimated to be at a floor of 72%, an increase from a leaked figure of around 30% at the beginning of the year.
Anthropic's unreleased model, Mythos, is considered potentially the biggest step-up in AI model capabilities in the last two years.
Anthropic has not broadly released its Mythos model due to concerns about its potential impact on the world, and its token cost is 5 to 10 times higher than other models.
Dylan Patel predicts that the cost to achieve the same quality as current Anthropic Opus models will be 100 times cheaper within a year.
Anthropic achieved its goal of creating a model with the capability of an L4 software engineer with Opus 4.6, a target originally set for the end of 2025.
Anthropic's unreleased Mythos model demonstrates the capabilities of an L6 software engineer, a significant advance from the L4-level capabilities of Opus 4.6.
Dylan Patel predicts that significant breakthroughs in robotics enabling few-shot learning will occur within the next 6 to 18 months.
OpenAI has secured large amounts of compute from providers including Oracle, CoreWeave, SoftBank, Microsoft, and is now also getting Tranium chips from Amazon.
The economic value created by the best AI models is growing faster than the industry's ability to supply the necessary compute infrastructure.