Got tokens?
Brent Donnelly summed up the situation in the market in his most recent podcast: “It seems like the market is more worried about the shortage of tokens rather than the shortage in oil.” I agree. There isn’t much edge anymore in trading the war in Iran - there is too much contradictory noise, Washington insiders are trading ahead of the market before information goes public, and weekend gap risk is significant. The euphoria from Friday’s headlines quickly reversed this weekend as Iran fired on two merchant ships flying the Indian flag while the US boarded an Iranian flagged vessel.
My base case is that the US and Iran remain in a state of perpetual “talks”. A ground invasion of Iran is an ax hanging over the head of the market, but I believe Trump will do everything to avoid this path as it would be suicide for the midterms and drag the US into a quagmire.
The demand for tokens, on the other hand, is an inevitable force propelled by Jevon’s Paradox, the lure of productivity gains, and the fear of being disrupted or left behind. It is immune to the level of interest rates, Fed liquidity, and the price of oil. Six months ago, the world wondered how much more powerful AI models could get, and how many more users would be willing to pay $20/month to get access to apps like ChatGPT. Today we have the answer - AI agents are now being left overnight to build products and apps, and the number of superusers paying over $100/month for Claude, ChatGPT, or Perplexity Computer has exploded.
Azeem Azhar is one of my favorite writers on AI wrote about the demand for tokens in his recent post:
That relentless load is why Google’s TPUs across all seven generations, some now seven and eight years old, are running at full utilization. Hardware that was expected to have depreciated years ago is still earning its keep.
The pressure shows up in revenue. For instance, Anthropic’s total revenue is growing, but the price per token is falling – and it’s falling faster than overall revenue is rising. This means the business is becoming increasingly dependent on volume to sustain growth.
The scary thing is that Anthropic’s Mythos is only the first model trained on Nvidia’s Blackwell GPUs to get released. OpenAI’s latest model Spud may be released as early as this week (based on Polymarket odds), kicking off the next leg of the arms race between the latest generation of models.
Paid subscriber section: Current positions, latest trades, the sector in AI I’m most bullish on
