The Watson Weekly Weekend edition is sponsored by Avalara - the agentic AI platform automating global tax and compliance for leading eCommerce brands.

I listened to Sundar Pichai’s sit-down with John Collison and Elad Gil recently, and I walked away thinking: most people are still not taking the right lessons from this conversation.

Everyone wants to debate whether Gemini is better than GPT or whether Search is dead.

Meanwhile, the actual interesting story is hiding in plain sight. Here’s what people are sleeping on.

Google has obsessed over latency since the beginning — famously displaying the millisecond query time right in the search results as a kind of flex. That was not an accident. It was a statement of values.

What Sundar described in this conversation takes that further than most people realize. Search teams at Google operate with formal millisecond-level latency budgets. If you shave 3ms off a query, you earn 1.5ms of budget credit. This is simple Google’s culture, and it’s been built over decades.

Now apply that same discipline to Gemini on Tensor Processing Units (TPUs) which train its LLMs. Flash models running at 90% of Pro capability but dramatically faster and cheaper to serve. This is not a product decision. It is a compounding competitive advantage that will be very hard to replicate.

And in a compute-constrained world — and we are absolutely in one right now — being fast and efficient is not just good UX. It is survival.

Pichai was remarkably candid here. $175B–$185B in CapEx in 2026, and Google still cannot spend as much as it wants. The constraints are real and physical — not software problems, and not money problems.

Here is the practical implication most executives are missing: the companies that secured compute early are pulling away from everyone else right now, and the gap cannot simply be closed by writing a check. There is not enough memory in the world to satisfy demand in 2026 or 2027. Some players may not get what they need. (Amazon has said two SEPARATE companies attempted to secure all of its compute for 2026.)

If your AI strategy assumes that compute will just be available when you need it, you should rethink that assumption immediately.

Also, I noted that Sundar is one of the only executives talking seriously about data centers in space. Elon Musk gets all the attention for thinking in civilizational timeframes. But Sundar quietly mentioned that Google already has a small team working on the concept of data centers in space as a long-term answer to the physical constraint problem. That’s interesting to me.

Think about what it takes to even have that conversation inside a company. You need leadership that genuinely believes the compute buildout curve extends far enough into the future to make orbital infrastructure relevant. That is a different kind of conviction from the one most CEOs operate with.

This is not a press release. It is a small team with a small budget working toward a first milestone. Which, if you have been paying attention to how Google bets on things, is exactly how Waymo started.

One of the most important things Sundar said was almost a throwaway line: the curve is shifting pretty dramatically this year.

He is talking about the gap between what the models can do and what most organizations are actually using them for. That gap is enormous right now. The models are capable of far more than most companies have figured out how to deploy.

But that gap is closing. The tooling is getting better. The interfaces are getting simpler. The agentic workflows that feel like science fiction today will feel obvious by the end of 2026.

If you are sitting on the sidelines because implementing AI feels too complex or too unstable, you are making a timing bet. And Sundar’s view — which I share — is that the window for that excuse is closing faster than most people expect.

This one stuck with me more than anything else in the conversation.

The CEO of Alphabet — one of the most complex organizations on earth — blocks a specific hour every week to review compute allocation at a project-by-project level. He knows which teams are using which compute units. He is making active prioritization decisions based on that information.

Most executives I talk to have no idea how their AI spend is allocated or what return it generates. They are treating it like a line item rather than a strategic resource.

Compute is the scarce resource right now. If you are not actively managing it, you are not actively managing your AI strategy. Those two things are the same thing.

The Google story in 2026 is not about whether any single model benchmark beats OpenAI this week. It is about a company that has been building the physical and intellectual infrastructure for this moment since 2016, run by a CEO who sweats the milliseconds and knows exactly where every TPU is going.

That is a different kind of company than the narrative suggested a year

Tony Bignell is out as Chief Innovation Officer after less than a year. Andy Caine takes the role effective this week. This is the third person to hold that position in under three years.

Elliott Hill put Bignell in the job last June as one of his first moves as CEO. Bignell delivered where it counted — the Vomero Premium shipped in eight months, half Nike's standard development window. That is exactly the kind of execution Hill said he needed.

It was not enough to keep him.

The backdrop matters. Nike shares just hit their lowest level in more than a decade. Revenue guidance disappointed. China — the second-largest market — is forecast to decline by as much as 20% this quarter, marking the seventh consecutive quarterly decline. Local Chinese competitors routinely ship seasonal product in the time it takes Nike to finalize a brief.

Hill's stated priority is speed. His innovation org has had three leaders in 36 months.

Those two facts do not coexist comfortably.

Caine has 20-plus years at Nike and real product credentials — Mercurial cleats, Air iterations. The pipeline is being rebuilt. The competitive gap, particularly in China, is not waiting for the org chart to stabilize.

Three innovation chiefs in three years is not a sequencing problem. It is a verdict on whether the structure can support the strategy.

Nike is not losing innovation chiefs. Nike is losing the argument that its org structure can support the speed its CEO is demanding.

Hill has been explicit: Nike needs to move faster. Three innovation leaders in three years means three people have now stood between that mandate and the reality of executing it inside one of the most complex product organizations in the world.