It’s happening. We’re officially in our AI IPO era.

OpenAI, Anthropic, and SpaceX all filed in the last week — which means the race for compute dominance, user attention, and capital market bragging rights is very much on.

And there’s a term I’ve been hearing a lot of lately that figures into it all: token maxxing.

Translation: the AI industry has spent the last few years throwing more tokens at everything — bigger context windows, longer prompts, more agents, more data, more compute — because more input often meant better output.

But now the bill is arriving.

Enter, the new term on the block: tokenomics.

As in: how do companies manage compute efficiently when every prompt, every agent, every automated workflow, and every “just ask the AI” moment comes with a price tag?

Beneath all the promises of efficiency, this is fundamentally an economics problem.

How do we get more output from fewer inputs? How do we maximize value from labor? How do we decide when to use an AI agent, when to use a human, and when to outsource to the intern? 

And with compute constraints only getting worse (see below on the Google + SpaceX deal) part of me wonders...

…at what point do humans cost less than compute?

Something to brew on ☕️

1. Anthropic and OpenAI both file for historic IPOs

My Tea: The AI IPO race has officially commenced. Anthropic currently leads on timing and valuation.

2. Apple announces revamped Siri AI at WWDC

My Tea: We’ve all been waiting for the Siri glow-up. Two questions: will it actually work? And is it too little, too late?

3. Google and SpaceX sign $920M/month compute deal

My Tea: This one’s particularly interesting given that Google has its own TPUs — but clearly needs even more compute. Timing is key, too: right before SpaceX’s IPO.

4. Alphabet raises an unprecedented $85B through public markets to fund its AI buildout

My Tea: Further proof IMO that the AI race is really a compute race... which is really a capital race.

5. NVIDIA launches Cosmos 3, a world model for physical AI

My Tea: NVIDIA doesn’t just want to power chatbots. It wants to power robots, cars, factories... everything that moves.

I have a hell of a few weeks coming up. I’m currently writing from Vegas, where I’m covering the Pegaworld conference. Lots of talk here about agentic systems, orchestration, and how enterprises can effectively use tokens.

Tomorrow I’m off to New York (the DAY of the NBA finals…wish me luck) for a Healthcare AI event. Then, Boston for my HBS 10-year reunion! Then, back to NYC for a very exciting partnership with IBM. Then Ohio. I’ll explain later. If you want to follow along, with the adventures, I’ll be storying on IG.

I’ve seen a lot of growth these past few weeks. Thanks to all the new folks here for subscribing. Here’s a little more about me.

(Crowd favorite fun fact: I have lived several places, including Japan, and I speak Japanese)

If you already think San Francisco is a bubble, try imagining it after the biggest AI IPO wave in history. A $3M home listing says it’s accepting Anthropic or OpenAI stock as payment. For the on-paper billionaires. Ofc.

I had a fascinating conversation with Okta’s CPO Ely Kahn last week about identity — and why it matters even more in an agentic world.

How do you prove you’re really you online? How do you give your AI agent access without handing over control?

We got into all of that. But first, because we are still human… here’s the blooper reel.

—

AI can code, but it’ll never brew tech tea like I do. Stay steeped my friends. -M