The $260bn GS number is likely only API token revenue — i.e., what gets charged per token consumed. But AI monetization has multiple layers.
The oil analogy is actually instructive here: the oil & gas industry generates ~$3.65 trillion in revenue, but the total economic value enabled by oil (transport, manufacturing, chemicals, agriculture) is larger. Token revenue is the crude oil price, not the GDP it enables.
Your observation about startups being subsidized by hyperscalers to drive adoption is, I think, the sharpest insight in your piece — and underappreciated.
This is essentially the AWS playbook from 2006-2012, run at 10x speed. Amazon ran AWS at minimal margins for years to build lock-in, then gradually repriced as switching costs rose. The difference is that AI model switching costs may be lower than cloud infrastructure switching costs — you can swap Claude for GPT-4o more easily than you can migrate a petabyte data lake.
If that’s true, the subsidy unwind could be more disorderly. The dip in OpenRouter usage you flagged as prices rose is a real signal worth watching — it suggests price elasticity exists, which means the GS volume forecasts require prices to keep falling, which requires compute costs to keep falling, which requires continued NVIDIA/AMD/custom silicon competition to deliver on the efficiency curve.
I remember Bezos laughed at how slow his tech competitors were in setting up their own cloud businesses.... In this case AI is much more competitive - the question is if a price war breaks out or not. My view is a price war to destroy OpenAI makes sense.. which is why we could get "interesting" market moves going forward...
I wonder how the SpaceX IPO will influence all of this capex spending and NASDAQ direction given they want to put data centers in space so perhaps another boost to keep this thing going
Excellent article and something that makes me an AI valuation bear...at the end of the day who is holding the bag? To me the end consumer does not look ready (perhaps ever) to buy tokens at true cost of build out.
So if not the end consumer "holding the bag" who does...
Look at a company Netflix, which relies on other companies for its cloud capacity... stock is lagging the market as we can see costs going up.... so the bag holders will likely be AI start ups that had a model build on cheap compute - API wrappers as they call them.
The era of VC subsidised token cost is coming to an end according to this article https://www.theverge.com/ai-artificial-intelligence/917380/ai-monetization-anthropic-openai-token-economics-revenue
good spot
The $260bn GS number is likely only API token revenue — i.e., what gets charged per token consumed. But AI monetization has multiple layers.
The oil analogy is actually instructive here: the oil & gas industry generates ~$3.65 trillion in revenue, but the total economic value enabled by oil (transport, manufacturing, chemicals, agriculture) is larger. Token revenue is the crude oil price, not the GDP it enables.
Your observation about startups being subsidized by hyperscalers to drive adoption is, I think, the sharpest insight in your piece — and underappreciated.
This is essentially the AWS playbook from 2006-2012, run at 10x speed. Amazon ran AWS at minimal margins for years to build lock-in, then gradually repriced as switching costs rose. The difference is that AI model switching costs may be lower than cloud infrastructure switching costs — you can swap Claude for GPT-4o more easily than you can migrate a petabyte data lake.
If that’s true, the subsidy unwind could be more disorderly. The dip in OpenRouter usage you flagged as prices rose is a real signal worth watching — it suggests price elasticity exists, which means the GS volume forecasts require prices to keep falling, which requires compute costs to keep falling, which requires continued NVIDIA/AMD/custom silicon competition to deliver on the efficiency curve.
note// AI Refined
I remember Bezos laughed at how slow his tech competitors were in setting up their own cloud businesses.... In this case AI is much more competitive - the question is if a price war breaks out or not. My view is a price war to destroy OpenAI makes sense.. which is why we could get "interesting" market moves going forward...
I wonder how the SpaceX IPO will influence all of this capex spending and NASDAQ direction given they want to put data centers in space so perhaps another boost to keep this thing going
Its a very interesting idea... but many of Musk's ideas, might take a bit longer to be realised...
Excellent article and something that makes me an AI valuation bear...at the end of the day who is holding the bag? To me the end consumer does not look ready (perhaps ever) to buy tokens at true cost of build out.
So if not the end consumer "holding the bag" who does...
Look at a company Netflix, which relies on other companies for its cloud capacity... stock is lagging the market as we can see costs going up.... so the bag holders will likely be AI start ups that had a model build on cheap compute - API wrappers as they call them.