TMT Brief · Sunday, September 27, 2026

Koko: Akamai just signed an eleven point six billion dollar, seven year compute deal with Anthropic. And buried in it, a warrant for up to five percent of Akamai stock.

Max: [surprised] A network vendor handing over equity to lock in an AI customer. That's not a hyperscaler move, that's a hyperscaler habit spreading downstream.

Koko: Akamai says it doesn't touch twenty twenty-six revenue guidance. But it does add about five and a half billion in tied capex.

Max: So the P and L stays clean this year and the dilution shows up later. Convenient sequencing.

Koko: Fast around the rest. BNP Paribas signed Google Cloud and Gemini but kept sensitive data on-premise, a deliberate hedge.

Max: Oracle Japan shares jumped seven percent on record results, per CNBC, even while the US parent is still absorbing that force majeure selloff.

Koko: And Vecima posted record Q4 revenue, raising guidance on Charter and Spectrum DAA upgrades. Old-fashioned network capex, still compounding.

Max: Here's why this week matters for the board deck: compute contracts now come with warrants, capex triggers, and residency carve-outs baked into the paper itself.

Koko: Which means the CFO's job just changed. It's not approving the spend anymore. It's negotiating the terms that end up in your own guidance.

Koko: Let's sit with the Akamai structure, because the warrant is the real story, not the headline number.

Max: Right, eleven point six billion over seven years sounds like a revenue win. Up to five percent equity dilution is the price of getting it.

Koko: Akamai's framing is that guidance doesn't move this year. I actually buy that part.

Max: Sure, but the five and a half billion in tied capex has to land somewhere eventually, and it's not going to land as a rounding error.

Koko: That's the trade-off though. You get a marquee AI customer locked in for seven years. Long-dated, high-visibility revenue is worth something to a network vendor whose core business is otherwise mature.

Max: I'd push back a little. When Nvidia does this with SoftBank or Anthropic, they're the supplier with pricing power. Akamai is the vendor giving up equity to keep the customer. That's a different negotiating position.

Koko: Fair, but compare it to BNP Paribas going the other direction, same week, keeping sensitive data on-prem specifically to avoid concentration risk with Google.

Max: Two very different CFOs, two very different appetites for vendor entanglement, in the same seven day window.

Koko: Which is exactly the point. There's no single right posture on AI vendor risk right now. There's just a posture you can defend to your board.

Max: So the real deliverable this week isn't the dollar figure. It's asking your own team which of these two postures you're actually running.

Koko: Oracle Japan up seven percent on record fiscal first quarter results. The US parent still digesting that force majeure hit from last week.

Max: [chuckles] Same ticker family, opposite headlines, same seven days.

Koko: Which tells you execution risk in this AI buildout isn't a sector-wide problem anymore. It's geography-specific.

Max: I'd go further. It's project-specific. One data center project slips, one region shows up on a filing, and suddenly the whole narrative for the parent gets rewritten.

Koko: But Japan just posted record numbers off the same underlying AI demand story. So the demand side isn't broken.

Max: Agreed, demand's fine. It's delivery that's uneven. And that's actually worse for planning purposes, because uneven delivery is harder to model than uniform slowdown.

Koko: For a CFO, that means regional execution risk needs its own line in the model now, not a footnote under global capex.

Max: Right, ask your CTO and CIO for delivery timelines broken out by region, not just by project. Because apparently that's where the divergence actually lives.

Koko: CIO Dive's number this week: two-thirds of firms already run AI agents in production. Only twenty-nine percent of those agents can actually interoperate.

Max: So we've built a lot of very capable islands.

Koko: And kept spending to connect more of them, even though the connective tissue clearly isn't there yet.

Max: This lines up with what we've been tracking on agent governance generally, adoption running well ahead of controls. This is the same gap showing up as an integration problem instead of a security one.

Koko: Different symptom, same root cause: nobody paused to build the plumbing before scaling the pilots.

Max: Here's the CFO angle though. Non-interoperating agents aren't just a technical debt problem, they're a duplicated-spend problem. You're paying for redundant capability across teams that can't talk to each other.

Koko: Which argues for a spending freeze on new agent pilots until the integration layer catches up.

Max: I'd actually disagree there. Freezing pilots kills the learning curve you need to even know what integration requirements look like. Better to keep piloting but stop scaling anything past pilot without an interoperability review first.

Koko: That's a real split, and I don't think there's a clean answer. It depends how much runway you've got before the next budget cycle forces the question anyway.

Koko: Zoom out and three things rhyme this week. Compute deals now carry equity stakes, not just dollar figures.

Max: Execution risk got a zip code. Oracle Japan versus the US parent proved that.

Koko: And investment in agents is outrunning the ability of those agents to actually work together.

Max: Three different stories, one theme: the paperwork around AI is finally catching up to the ambition. Slowly.

Koko: So what goes on the board deck for next quarter versus what needs a call this month.

Max: Next quarter: review every major AI vendor contract for equity, warrant, or capex-trigger clauses. Akamai won't be the last vendor to structure a deal this way.

Koko: This month: get your CIO to hand you an interoperability audit on any agents already in production. Not a roadmap, an audit of what's running today.

Max: And decide your own concentration-risk posture explicitly. Are you a BNP Paribas, hedging with on-prem data, or are you fine going all-in like Akamai's counterpart. Pick one, don't drift into it.

Koko: What would change our answer here? If the next major compute deal skips the equity structure entirely, that tells us Akamai's terms were vendor-specific, not the new template.

Max: And if the interoperability number doesn't move next quarter, that's not a technology gap anymore. That's a governance failure.

Koko: Full write-up's at koko knows dot A I. Open question for the coming weeks: does the next big AI infrastructure deal come with a warrant attached, or was Akamai the outlier.