1/
Google just raised its CapEx to $85 billion.
Two-thirds of it is going into servers not buildings.
This isn’t about flashy new models.
It’s about controlling the rails of the AI economy.
Here’s what that really means 🧵
Google just raised its CapEx to $85 billion.
Two-thirds of it is going into servers not buildings.
This isn’t about flashy new models.
It’s about controlling the rails of the AI economy.
Here’s what that really means 🧵
2/
In 2023, Google spent 10.5% of revenue on CapEx.
In 2025, it’ll be 21.5%.
That’s not normal.
That’s wartime spending.
Because in AI, compute ≠ cost.
Compute = control.
In 2023, Google spent 10.5% of revenue on CapEx.
In 2025, it’ll be 21.5%.
That’s not normal.
That’s wartime spending.
Because in AI, compute ≠ cost.
Compute = control.
3/
The Jevons Paradox is back:
As inference becomes more efficient, usage skyrockets.
Efficiency doesn’t save you money in AI.
It just pushes demand further into the red.
That’s why the GPU arms race keeps accelerating.
The Jevons Paradox is back:
As inference becomes more efficient, usage skyrockets.
Efficiency doesn’t save you money in AI.
It just pushes demand further into the red.
That’s why the GPU arms race keeps accelerating.
4/
Google isn’t just stockpiling chips.
It’s vertically integrating the entire stack:
• TPUs it controls
• YouTube + Android for distribution
• Ads + Cloud to cross-subsidize it all
That’s not a product strategy.
That’s a war strategy.
Google isn’t just stockpiling chips.
It’s vertically integrating the entire stack:
• TPUs it controls
• YouTube + Android for distribution
• Ads + Cloud to cross-subsidize it all
That’s not a product strategy.
That’s a war strategy.
5/
Even OpenAI Microsoft’s golden child runs overflow workloads on Google Cloud.
That’s not just hedging.
That’s a signal.
When the most advanced AI labs need backup compute,
Google becomes not a rival but infrastructure.
Even OpenAI Microsoft’s golden child runs overflow workloads on Google Cloud.
That’s not just hedging.
That’s a signal.
When the most advanced AI labs need backup compute,
Google becomes not a rival but infrastructure.
6/
CapEx this big comes with risks:
• Q2 free cash flow dropped 61%
• GPU lead times >12 months
• Power constraints could break timelines
• ROI pressures are mounting fast
Alphabet is walking a tightrope.
CapEx this big comes with risks:
• Q2 free cash flow dropped 61%
• GPU lead times >12 months
• Power constraints could break timelines
• ROI pressures are mounting fast
Alphabet is walking a tightrope.
7/
But they’re not alone.
Amazon’s Andy Jassy is thinking the same way.
Microsoft has committed billions to power deals.
CoreWeave is locking up GPU clusters like oil barons in the 70s.
Infrastructure is the game.
Models are just the players.
But they’re not alone.
Amazon’s Andy Jassy is thinking the same way.
Microsoft has committed billions to power deals.
CoreWeave is locking up GPU clusters like oil barons in the 70s.
Infrastructure is the game.
Models are just the players.
8/
Most AI startups won’t own their future.
They’ll rent it from Alphabet, Amazon, or CoreWeave.
In 2020, everyone was building.
In 2025, most are tenants.
Infrastructure ≠ cost center anymore.
It’s the moat.
Most AI startups won’t own their future.
They’ll rent it from Alphabet, Amazon, or CoreWeave.
In 2020, everyone was building.
In 2025, most are tenants.
Infrastructure ≠ cost center anymore.
It’s the moat.
9/
This might be #Google’s AWS moment.
Or it could be another CapEx panic waiting to happen.
If they can monetize the infrastructure fast enough,
they win the whole stack.
If not, the moat might collapse under its own weight.
This might be #Google’s AWS moment.
Or it could be another CapEx panic waiting to happen.
If they can monetize the infrastructure fast enough,
they win the whole stack.
If not, the moat might collapse under its own weight.
10/
#Alphabet’s $85B bet isn’t about AI hype.
It’s about building the rails that others will ride.
Control compute.
Control power.
Control distribution.
In this war, infrastructure is destiny.
#datacenters
#Alphabet’s $85B bet isn’t about AI hype.
It’s about building the rails that others will ride.
Control compute.
Control power.
Control distribution.
In this war, infrastructure is destiny.
#datacenters
• • •
Missing some Tweet in this thread? You can try to
force a refresh
