Digital AdvertisingAI AdoptionAI boomAI Infrastructure

Tokenomics 2: Electric Boogaloo 😉😂

Last week we saw how Nvidia Blackwell will reduce the cost of tokens by 35X to 50X (per dollar, per MW, respectively) and Vera Ruben (launching in Q3) will reduce that by another 10X.
There are plenty of studies showing the explosive growth of AI in the economy including 315% CAGR growth in Target's online site (now 1.7% of all sales are AI lead gen).
In fact almost 5% of the top 5 online retailers is AI lead gen (growing 145% to 315% CAGR) indicating that by 2027-2032 50+% of online retail sales will be AI generated.
Other studies (including from Princeton) show why the big expensive models are MUCH better at operating businesses and doing complex tasks. This explains why frontier labs are thriving.
The RPO cloud backlog is $2.1 trillion and growing 185% per year BUT 50% of that is Anthropic and OpenAI so what's the proof that those companies can ever become profitable?
Ramp Capital is plugged into 70,000+ companies and tracks real time spending and its monthly AI spending report is the gold standard of actual AI spending.
Median company AI revenue spending is up 12X in 2.5 years, 111% CAGR growth, and Top 10% ($611 per month per employee) is growing 220% and top 1% ($7,450 per employee) is growing 310% CAGR.
At current growth rates in 5 years the median company will be spending $464 per month per employee (8% of salary up from 0.23% today). The question of "is AI profitable for companies?" The answer is a definitive "YES!"

Adam Galas

Jun 22, 20265:55 PM ET3450

Last week, we looked at Tokenomics and the exciting upcoming decline in token prices, which is expected to lead to continued AI adoption and AI revenue (and even margin expansion) for AI companies like Anthropic and OpenAI.

Tokenomics: The Economy Of AI Part 1

Today, we’re answering the question that a lot of investors have about the incredible amount of AI spending.

OK, the math seems logical, but aren’t All Models Wrong?

Kudos to anyone who remembered one of the smartest quotes to ever be uttered by a British Statistician😉

Source: Business Insider, Chat GPT 5.5 Pro

OK, so Blackwell is able to generate 1 mega token (1 million tokens) for just $0.12, 35X cheaper than Hopper. And it’s 65X faster at generating tokens AND most importantly for companies like Microsoft, Amazon, and Alphabet, where power is the single biggest constraint, the same 1 Megawatt can now generate 50X more tokens.

That is the supply growth that Jensen and his team of delightful engineering nerds have created.

BUT what evidence is there that all the extra capacity can be soaked up!? I mean, if we’re about to crank up the supply of tokens by 50X, then even in the age of AI agents and power users, what if we saturate demand!?

Source: The Information, SEC filings, Chat GPT 5.5

The RPO backlog is the contracted revenue backlog that hyperscalers have secured for their AI data centers and cloud businesses.

“OK but isn’t like 50% of that backlog Anthropic and OpenAI?! The unprofitable companies that, if they grow their way into bankruptcy, could disappear and thus take half of the RPO backlog with them?”

Yes indeed, about 50% of the RPO backlog is from Anthropic (profitable in Q2 for the first time but still negative on an FCF basis) and OpenAI, which is expected to lose $212 billion before becoming profitable in 2030.

So remember first principles investing. If Anthropic and OpenAI are 50% of the RPO backlog (which is growing at 185%, 2X as fast as capex spending), then let’s analyze the health of those companies (as best as we can before the S1 forms come out).

Almost 5% of the top retailer sales are now coming from AI recs, and that is growing at 145% to 315% CAGR

So 5% of the top 5 retailers are now AI, but when would that reach 50% at current growth rates?

How We Know Anthropic And OpenAI Are (Likely) Healthy

OK, so companies pay OpenAI and Anthropic for their API tokens (the power users), though OpenAI also has 54 million paying members (5.4% of total users).

Those 50 million paying members (and 4 million Pro) are generating $1.8 billion per month in revenue = $17.28 billion per year out of $25 billion ARR, so the subscription business is super important to OpenAI for now, while Anthropic, the coding-focused company, is where the enterprise dollars are the key growth driver.

BUT given that OpenAI’s paying subs have been relatively stable over time, let’s focus on the Enterprise clients, because this is the most important thing for both Anthropic and OpenAI.

A $17.3 billion ARR business that is growing at a very modest rate is not going to get OpenAI to profitability. Advertising MIGHT help a bit, but it’s going to be enterprise API revenue that makes OpenAI profitable (or the lack of it that will kill the company).

So this brings us to Ramp Capital, the most important enterprise SAAS company that no one has ever heard of.

Ramp Capital: The Way We Track Real AI Spend By Enterprise Customers

Keep Reading

Create a free account to access more content.