6. AGI needs less new infrastructure than you'd think.
You've rightly emphasized the massive barriers to entry in AGI research—it's not just about GPUs, but also data centers, power, land, regulatory approvals. The scale required rules out almost everyone.
But a handful of companies have already cleared that hurdle. Google, for example, consumed around 2.74 GW of power in 2023 ( https://en.wikipedia.org/wiki/Google ) and likely surpassed 3 GW in 2024. Microsoft, Amazon, and Meta aren't far behind. These companies have a track record of success solving infrastructure problems and they're not slowing down.
So don't worry about needing vast infrastructure upgrades for AGI; the top companies are already there, they're just wasting most of their capacity on non-AGI. The moment an internal prototype convinces a Pichai or Nadella that the finish line is in sight, expect a sharp pivot. Cloud prices will spike, GPUs and hardware accelerators vanish from the market, global compute shortages will hit as these companies redeploy their enormous infrastructure towards the ultimate business advantage.
Thanks Tedd, really sharp comment. I actually think we agree more than it might seem.
You're absolutely right that the major players—Google, Microsoft, Amazon, Meta—already command vast infrastructure footprints, with multi-gigawatt scale. But that doesn’t negate the infrastructure bottleneck thesis. If anything, it reinforces it.
Yes, these companies have already built the necessary substrate. But most of that capacity is currently allocated to other revenue-generating priorities: ad targeting, search, cloud VMs, media delivery. Repurposing that footprint for AGI isn’t as simple as flipping a switch. It involves:
--Cannibalizing existing businesses. Redirecting 1–2 GW from Azure or AWS to AGI training means displacing paying customers, which comes with major opportunity costs.
--Dealing with architectural mismatches. A lot of deployed infrastructure isn't optimized for training massive models. Edge zones, general-purpose VM clusters, and legacy interconnects don’t readily convert into dense, bandwidth-heavy training nodes.
--Retrofits and rewiring. Even if you wanted to reallocate existing infra, the bandwidth density, power delivery, and thermal constraints of many facilities would require substantial retrofitting.
--Permitting and external constraints. Even these giants are still bound by local zoning fights, water rights issues, and multi-year interconnect queues when trying to expand beyond their current footprints.
In other words: yes, they have the infra. But to reallocate and scale it toward AGI is nontrivial. And once they do make that pivot—if a prototype convinces a Pichai or Nadella, as you put it—then we’re looking at a global strategic shock: cloud price spikes, GPU shortages, scramble for power, geopolitical energy conflicts. That pivot doesn’t eliminate the infrastructure bottleneck; it just activates it.
So I’d put it this way:
AGI won’t require fundamentally new categories of infrastructure. But it will require a radical reallocation of existing infrastructure. And that reallocation is going to hurt.
Appreciate the push—your point adds important nuance.
Agreed. My comment is framed as counter-argument for your list but it is almost certainly an apocalyptic outcome. Could any hyper-scalar CEO be so tempted? One won't act alone, but if they think the others are moving that way...?
Let me add one more point:
6. AGI needs less new infrastructure than you'd think.
You've rightly emphasized the massive barriers to entry in AGI research—it's not just about GPUs, but also data centers, power, land, regulatory approvals. The scale required rules out almost everyone.
But a handful of companies have already cleared that hurdle. Google, for example, consumed around 2.74 GW of power in 2023 ( https://en.wikipedia.org/wiki/Google ) and likely surpassed 3 GW in 2024. Microsoft, Amazon, and Meta aren't far behind. These companies have a track record of success solving infrastructure problems and they're not slowing down.
So don't worry about needing vast infrastructure upgrades for AGI; the top companies are already there, they're just wasting most of their capacity on non-AGI. The moment an internal prototype convinces a Pichai or Nadella that the finish line is in sight, expect a sharp pivot. Cloud prices will spike, GPUs and hardware accelerators vanish from the market, global compute shortages will hit as these companies redeploy their enormous infrastructure towards the ultimate business advantage.
Thanks Tedd, really sharp comment. I actually think we agree more than it might seem.
You're absolutely right that the major players—Google, Microsoft, Amazon, Meta—already command vast infrastructure footprints, with multi-gigawatt scale. But that doesn’t negate the infrastructure bottleneck thesis. If anything, it reinforces it.
Yes, these companies have already built the necessary substrate. But most of that capacity is currently allocated to other revenue-generating priorities: ad targeting, search, cloud VMs, media delivery. Repurposing that footprint for AGI isn’t as simple as flipping a switch. It involves:
--Cannibalizing existing businesses. Redirecting 1–2 GW from Azure or AWS to AGI training means displacing paying customers, which comes with major opportunity costs.
--Dealing with architectural mismatches. A lot of deployed infrastructure isn't optimized for training massive models. Edge zones, general-purpose VM clusters, and legacy interconnects don’t readily convert into dense, bandwidth-heavy training nodes.
--Retrofits and rewiring. Even if you wanted to reallocate existing infra, the bandwidth density, power delivery, and thermal constraints of many facilities would require substantial retrofitting.
--Permitting and external constraints. Even these giants are still bound by local zoning fights, water rights issues, and multi-year interconnect queues when trying to expand beyond their current footprints.
In other words: yes, they have the infra. But to reallocate and scale it toward AGI is nontrivial. And once they do make that pivot—if a prototype convinces a Pichai or Nadella, as you put it—then we’re looking at a global strategic shock: cloud price spikes, GPU shortages, scramble for power, geopolitical energy conflicts. That pivot doesn’t eliminate the infrastructure bottleneck; it just activates it.
So I’d put it this way:
AGI won’t require fundamentally new categories of infrastructure. But it will require a radical reallocation of existing infrastructure. And that reallocation is going to hurt.
Appreciate the push—your point adds important nuance.
Agreed. My comment is framed as counter-argument for your list but it is almost certainly an apocalyptic outcome. Could any hyper-scalar CEO be so tempted? One won't act alone, but if they think the others are moving that way...?
Yeah, in some sense the game theory is very similar to Cold War-era mutually assured destruction.