- The Digital Asset Daily
- Posts
- The AI Spending War Just Hit a New Stage
The AI Spending War Just Hit a New Stage
A New Chart Shows Exactly Who’s Winning the AI Spending War
Teeka Tiwari
July 15, 2026
A week ago, I warned you the AI compute business looked like it was about to enter a price war.
I said compute would become commoditized – just like coal and natural gas – as the hyperscalers cut prices against each other to secure market share.
Meta just proved my point.
Last week, the company announced its new AI model, called Muse Spark 1.1.
Meta priced access to that model at $1.25 per million input tokens and $4.25 per million output tokens. Compare that to Anthropic, which charges $5 and $25, respectively, for its top model, Opus 4.8.
That’s 4-6x cheaper.
Tokens are how AI companies measure usage. A token is usually a piece of a word, a full word, or a small phrase. Input tokens are what you feed into the model — your questions, prompts, commands, or instructions. Output tokens are what the model gives back — the answer, analysis, code, image instructions, or response.
So when Meta charges 4-6x less per million tokens, it means developers can ask more, build more, and run more AI tasks for a lot less money.
That’s Meta using price as a weapon, exactly as I wrote it would on July 8:
[AI compute is] one of the most commoditized services a company can offer... Which means the only lever left to pull is price. And that's exactly what's coming, a price war for AI compute among Meta, Amazon, Microsoft, and Google. When four giants compete to sell the same commodity, none of them wins on price. They all lose margin.
I’m not bringing this up to take a victory lap. I’m bringing it up because this commoditization is happening faster than most people realize.
And if you’re holding these popular AI hyperscalers, you could see their valuations compress as this trend plays out.
Meta Shows Its Hand