Exponential View Has an Informed View of the Money Flows in the Generative-AI Economy: CHARTS OF THE DAY
The $30 Billion Token Tab vs. the $100 Billion GPU Binge: sers spend $30 billion on roiling boils of linear algebra while GPU makers and TSMC feast on an $100‑billion build‑out—and nobody really...
The $30 Billion Token Tab vs. the $100 Billion GPU Binge: sers spend $30 billion on roiling boils of linear algebra while GPU makers and TSMC feast on an $100‑billion build‑out—and nobody really knows if it’s for genuine productivity or just option value on the First Coming of Digital God…
In the first quarter of 2026, customers paying $30 billion and tokens delivered to users using “Generative AI” for something: Of that $30 billion, Exponetial View thinks $2.5 billion net of immediate payments to datacenters and foundation modeler goes to app providers, $5 billion net of immediate payments to datacenters goes to foundation modelers (including Google’s Gemini and xAI’s Grok), and the other $22.5 billion to datacenter operators.
But a great deal of the within generative AI sector payments right now are very funny money of various flavors.
Of course, this $30 billion is small potatoes compared to the $81.6 billion paid to NVIDIA for NVIDIA-designed GPU chips, plus another $15 billion paid to the other “fabless” designers. Spending $100 billion (plus a lot more on memory) on the build-out but so far collecting $30 billion in revenue is an interesting deal.
NVIDIA (and AMD, and Broadcom) then turned around and immediately paid $22 billion to TSMC for actually making the GPU chips.
Are users spending that $30 billion right now because they are finding these roiling boils of linear algebra genuinely useful, or are they spending it because they think they may well need to build the engineering capabilities to use the subsequent generations of these things when they do become truly useful? I do not know. And neither does anybody else.
The Exponential View team has a nice take on what this looks like from the corporate offices:
Azeem Azhar, William Gildea, Hannah Petrovic, Nathan Warren, & Marija Gavrilov: 2 | Economy <https://intelligence.exponentialview.co/assets/ev-state-of-ai-economy-2026.pdf>: ‘Big is still small, and early. Even for the highest corporate spenders, AI is a rounding error in the P&L. It still looks early. Initiatives have focused on efficiency & cost savings, although the mix is changing. And measured revenue may understate the social gains, as consumers report benefits that don’t yet show up in the data. Against GDP, AI revenue is still a rounding error. At a company level, AI spending is still relatively small: e.g. Uber’s $1.5k per engineer barely dents the P&L…
See also: <https://www.exponentialview.co/p/the-state-of-the-ai-economy> <https://intelligence.exponentialview.co/>
Is Anthropic planning on launching its IPO this summer because they have tapped out the private market of informed investors and need to draw on the skeptical in public markets who do not know the industry, in which case this is a sign that they have private information that this is really for real—or alternatively, that they are cultic votaries of Digital God? Or is Anthropic planning on launching its IPO this summer because they have the opportunity to draw on the gullible in public markets who do not know the industry, in which case this is a sign that they have private information that right now is the peak of the bubble—or at least that they can unwind a substantial part of the bet so that they are plutocrats in the future no matter how this turns out?
I have found nobody credible who claims to know.