# AI capex — X 热门讨论 (2026-09-26 19:33 UTC)
## @0xcatuchiha (0x) · 09-26 03:20 · ♥31 ↻0 💬41 the more i look at @vangrid_io, the more the “zero capex sensor network” idea makes sense.
you don’t need to deploy a fleet of specialised mapping hardware to keep collecting ground level data.
the phones are already out there.
people can capture locations through the vangrid app, those different views can be turned into spatial data, and the network can attach provenance to the capture.
that changes the economics of collecting real world data.
instead of building the sensor network first, vangrid is using the devices already sitting in people’s hands.
the hardware is already everywhere.
the interesting part is turning all those cameras into a coordinated data layer for physical ai. https://x.com/0xcatuchiha/status/2103685994214219898
## @harshmadhusudan (Harsh Gupta Madhusudan) · 09-26 08:44 · ♥30 ↻7 💬3 The two prop/closed mega frontier labs keep on delaying their IPOs, even as US markets are near all-time highs and the tech is no doubt transformative. The reason is very simple: at these multi-trillion dollar valuations, the unit economics and business model just does not add up despite very rapid growth.
And the entire AI trade is ultimately predicated upon the frontier labs, closed and open, being viable.
Since India has been one of the losers of the AI trade (not AI tech), investors in India should consider their priors about whether this level of capex (underwritten by such a parabolic trade) can continue given that cash flows, reserves, debt and equity issuance all are being dipped into. https://x.com/harshmadhusudan/status/2103767557597696455
## @rohanpaul_ai (Rohan Paul) · 09-26 06:48 · ♥32 ↻3 💬8 The $800B–$1T AI capex wave is a beautiful choreography of chips, power, credit, and capital.
Hyperscalers bring the credit quality, outside investors bring the capital, and frontier AI firms get access to far more compute than they could finance on their own.
From Columbia Business School report "Financing the AI Buildout" ans that. > 引用 @rohanpaul_ai: How much revenue each unit of deployed compute would need to generate for 182.7 GW of additional U.S. data-center capacity between 2025 and 2032
This Columbia Business School report "Financing the AI Buildout" ans that.
182.7 GW is its central buildout scenario which is the U.S. data-center capacity between 2025 and 2032.
Using the GB300 NVL72 configuration, that capacity works out to roughly 1.07 million racks / 77 million GPUs, which lets the author express the required $3.725 trillion of mature annual revenue as about $5.5 per installed GPU-hour at full utilization.
This required unit revenue sits within the $6–$10+ per GPU-hour for current high-end NVIDIA capacity, so the central scenario does not require some radically higher compute price to make the arithmetic work. https://x.com/rohanpaul_ai/status/2103738379913552123
## @LoganJastremski (Logan Jastremski) · 09-25 19:24 · ♥33 ↻4 💬5 Dropping a podcast with @pequityresearch
Mr. P has been doing some great work breaking down the AI buildout and following where hyperscaler capex actually goes. In this episode I wanted to walk through the full stack with logic, memory, power, and networking.
We get into why memory could become the largest line item in the AI bill, what old GPU rental prices tell us about compute demand, and where value is going to accrue as the physical constraints get harder to solve.
At the center of the conversation is his view that AI spending cannot grow forever. Long-term contracts might change the shape of the next memory downturn, but they don’t eliminate the cycle. And signing a 10-year contract does not mean anyone can actually see 10 years of demand.
We also get into why he’s excited about optics, where NAND and HBF fit as agents use more memory, and his views on Chinese open source and the future of US model development.
We discuss: - Why he thinks 10-year demand visibility is bullshit - Memory’s growing share of hyperscaler capex, and why estimates vary so much - Why older GPUs are still renting and what that says about compute demand - How long-term agreements, pricing floors, and prepayments actually work - Power as a bottleneck, and why identifying a constraint isn’t the same as finding an investment - Copper vs optics, and where networking value accrues - Agents, NAND, and where HBF fits in the memory hierarchy - CXMT, Chinese open source, and the risks of slowing frontier model development
Timestamps: 0:00 – Why AI Spending Can’t Grow Forever 1:13 – P Equity Research’s Background 5:48 – Where Hyperscaler Capex Actually Goes 8:00 – Is Compute Still Tight? 12:08 – Memory’s Share of the AI Bill 17:20 – Why the Memory Cycle Isn’t Dead 18:37 – Inside a Long-Term Agreement 28:30 – What Happens If Customers Cancel? 32:17 – The Rising Cost of the AI Buildout 33:43 – Copper vs Optics 38:49 – Power and Gas Turbines 39:29 – US Models and Chinese Open Source 42:36 – Agents and NAND 47:00 – Where HBF Fits 52:23 – What P Is Most Excited About 56:27 – CXMT and China’s Memory Industry 1:04:00 – Closing Thoughts
Enjoy! https://x.com/LoganJastremski/status/2103566276723613791