This month felt like AI moved into the machinery.

Government. Defense. Hiring. Advertising. Startup funding. Data infrastructure. Diplomacy. Speech. Cybersecurity. Football ads, somehow.

All the normal places.

Compute became currency

OpenAI offering Y Combinator startups up to $2M in API credits in exchange for equity was one of the cleanest tells of the month.

Sam Altman has worked out how to invest using his own fuel. Not cash. Compute.

In 2026, that might be close enough.

For OpenAI, it is brilliant. Equity upside, distribution, dependency, and a front-row seat to the next generation of companies being built.

For startups, it is tempting too. $2M in credits is real money when your product is basically a nice interface and a terrifying token bill.

The platforms are getting closer to the cap table.

Which feels important.

Also faintly medieval.

YC quietly changed the startup religion

Y Combinator also put out its latest Requests for Startups, and the really interesting bit was the push toward AI-native service companies.

For years, the religion was SaaS. High margin. Low touch. No messy humans.

Now YC is basically asking: what if AI lets service businesses scale like software?

Quiet announcement. Big signal.

The services layer may be where a lot of the real value lives. Which is funny, because the pitch was that AI would do everything.

Turns out “doing everything” might itself be a service business.

Advertising got a new payment model

JLR hiring WPP on an outcome-based remuneration model feels like a big one for advertising.

Agencies have talked about moving away from hours forever. Somewhere, a strategy director in 1998 is still saying “we should be paid for value, not time” while making a deck about brand love.

Now JLR and WPP are actually doing it.

If machines do more of the execution, clients are going to care less about how many people touched the work and more about what actually happened.

Did sales move?

Did demand grow?

Did the brand get stronger?

Did the work do the thing?

Terrifying for the old agency model. Quite exciting for people who can actually deliver.

And the holding companies seem to know it.

Publicis bought LiveRamp for $2.2B, which is basically a giant bet that in the AI marketing era, the cleanest data pipes build the smartest systems.

Less “who has the best deck.”

More “who has the best data.”

Slightly less romantic. Probably more valuable.

The labs moved into implementation

Anthropic’s enterprise services venture acquiring Fractional AI, hot on the heels of OpenAI launching its Deployment Company and buying Tomoro, felt like another big tell.

For a while, the labs said: here is the model.

Now they are saying: here is the model, the implementation team, the workflow playbook, the engineers, the platform, and eventually, presumably, the invoice for your entire operating model.

They are coming for everything.

Quietly at first.

Then all at once.

The market found religion in data centers

The Dow hit all-time highs again this month, helped along by AI optimism, chip stocks, and the market’s continuing belief that if we simply build enough data centers, everything will probably be fine.

Very comforting.

Nothing says “healthy economy” like looking under the hood and finding Nvidia, Micron, hyperscaler capex, and a cooling bill the size of a mid-tier nation.

The economy is three data centers in a trench coat remains one of the more accurate things I have said recently.

I’m sure the trench coat is structurally sound.

Context windows got silly

Subquadratic came out with SubQ, claiming a 12 million-token context window and an architecture that makes long-context analysis dramatically cheaper.

If true, that is a big deal.

Whole codebases.

Whole legal archives.

Whole financial histories.

Finally, an AI that can read the entire mess instead of asking you to upload it in sad little chunks.

Psst… 12 million tokens.

China credentialed the influencer economy

China tightened rules around influencers giving advice on finance, health, law and education, requiring verified qualifications before people can cosplay as experts online.

On one hand, fewer fake finance gurus in rented Lamborghinis feels like a civilizational win.

On the other hand, giving the state a big “qualified to speak” stamp has… range.

Still, if this prevents one more 23-year-old from explaining passive income in front of a rented Lamborghini, civilization gets a small point on the board.

AI got a government badge

This month, AI also marched into the machinery of civilization wearing a suit that didn’t quite fit.

The Pentagon reportedly signed classified AI agreements with major tech companies, while Anthropic was left outside the room because it would not loosen restrictions around surveillance and autonomous weapons.

Anthropic spends all this time trying to be the sensible one, and the reward is apparently being told it can’t come to the really serious meetings where sensible people might have been useful.

I honestly don’t know what to feel.

Proud of Anthropic.

Safer.

Also somehow less safe.

Since Nagasaki, humans have done the one thing that makes us the most human: agreed to stop using a technology that could destroy us all.

Will the AIs be so human?

OpenAI also got FedRAMP Moderate authorization, meaning U.S. government agencies can more easily use ChatGPT Enterprise and its API platform.

So ChatGPT has a government badge now.

I assume it will soon be attending a mandatory compliance webinar and eating a depressing sandwich under fluorescent lights.

AI started interviewing people

Amazon rolled out agentic software for hiring and supply chains, including AI-led interviews and recruiter notes.

I understand the appeal. Hiring is slow. Supply chains are complicated. Humans are expensive and occasionally want chairs.

But there is something very modern about being assessed for “culture fit” by a machine that has never once endured a company offsite.

The future of work is getting very efficient.

Possibly too efficient.

The CEOs went to China

Trump’s trip to China was one of the biggest stories of the month.

The interesting bit was who followed him there.

America’s most powerful CEOs were in Beijing meeting Chinese regulators and ministers around market access, regulatory barriers and deals.

And I know this is how business has always worked to some extent. But the mask feels thinner now.

We used to pretend these trips were about diplomacy, with business somewhere politely downstream of the statecraft. This felt more like the CEOs were the delegation, and the diplomats were doing continuity paperwork in the background.

China, I’m quite sure, had its diplomats on the other side. People who have spent lifetimes learning how to think in decades.

Meanwhile, democracies keep changing leadership every few years, and the tech elite just keep compounding.

There’s that line in Succession where Logan says: “I’ve been president. Lot longer than he has.”

No idea why that came to mind.

Adidas saved the mood

And then, thank God, Adidas released Backyard Legends.

A proper big football ad.

Starry. Silly. Majestic. World Cup energy.

Messi, Bellingham, Beckham, Zidane, Lamine Yamal, Trinity Rodman, Bad Bunny, and somehow even Timothée Chalamet works.

It reminded me of those old World Cup ads like The Other Game and Good vs Evil, where advertising remembered it was allowed to be grand, funny and slightly ridiculous.

Honestly, it felt right.

A useful reminder that amid all the capital allocation, compute deals, defense contracts, and machine-led interviews, creativity can still just make the room feel better.

And then, Milla Jovovich gave us the cherry on the cake!

My random inclusion that makes me unexpectedly happy: Milla Jovovich of The Fifth Element fame released an open-source AI memory tool based on the ancient Greek memory palace technique.

Sci-fi.

Models.

Ancient Greek memory theory.

So yes, it ticks all my boxes.

In a month full of compute-for-equity deals, government badges, AI hacking, machine-led interviews and CEOs doing diplomacy, there was something genuinely delightful about Leeloo helping improve RAG.

Honestly, fair enough.

And I’d revise the final section like this:

So what was the month?

Compute became startup currency.

YC rediscovered services.

Agencies moved toward outcomes.

Holding companies bought data pipes.

Labs moved into implementation.

Markets found comfort in data centers.

Context windows got ridiculous.

China credentialed the influencer economy.

AI got a government badge.

Hiring got more automated.

CEOs became diplomats.

AI hacking became industrial.

Adidas remembered how to make a football ad feel like an event.

And Milla Jovovich somehow wandered in with ancient Greek memory theory.

The connective tissue is pretty clear.

AI is changing how work gets sold, bought, funded, staffed, measured and paid for.

It is also changing who gets a seat at the table.

And this month, apparently, the answer was: the platforms, the labs, the CEOs, the holding companies, the governments, the investors, and Leeloo with a memory palace.