Sam Altman himself just flagged a real risk to the AI trade but the data says that risk is still years away (Save this).
Milk Road AI Twitter · Milk Road AI (@MilkRoadAI) · 2026-07-28
Milk Road AI counters Sam Altman's AI compute oversupply warning with supply chain data—including $1 trillion in Nvidia GPU orders, sold-out HBM memory for 2026, TSMC capacity booked through 2027, and 1% data center vacancy—and recommends infrastructure stocks positioned to benefit from sustained scarcity.
Appears in
Extraction
Topics: ai-compute-supplygpu-shortagedata-center-infrastructureai-infrastructure-investment
Claims
- Sam Altman identified compute oversupply as a potential risk if AI becomes so efficient that human attention becomes the bottleneck, placing the window at roughly two to six years out.
- Nvidia GPU orders have grown to $1 trillion through 2027 with lead times stretching to nearly a year, and all three HBM memory suppliers are sold out for 2026.
- TSMC's advanced node capacity is sold out through at least 2027, while spot DRAM prices are up roughly 8x since early 2025.
- Power infrastructure is the tightest constraint, with natural gas capacity taking five to seven years and nuclear over ten years to bring online, and grid interconnection queues already stretching beyond five years in many US regions.
- Data center vacancy sits at just 1% for a second straight year, with 92% of capacity currently under construction already pre-leased before completion.
Key quotes
A data center can be built in two to three years, but the power to run it can take five to seven years for natural gas, over ten years for nuclear, and grid interconnection queues already stretch beyond five years in many US regions.
Data center vacancy sits at just 1% for a second straight year, with 92% of capacity currently under construction already pre-leased before it's even finished.
Altman explains that compute oversupply could eventually emerge if AI models become so efficient that human attention becomes the bottleneck, or if hardware costs stop falling because the industry hits a scaling wall.