AI Shortages Through 2028
Dave Spillane@davespillane
· 1
−12.8%
since Oct 9 · after money added
The idea
I believe AI scaling keeps running into physical bottlenecks through 2027-28, so money flows to whatever runs short first. The basket leans into power for AI data centers, memory chips, and the operators running those centers, with a smaller slice in chip makers since those are already widely owned. I added a slice for the plumbing behind AI coding tools: the monitoring, storage, and security layers that get busier as more code gets written.
What it owns
- Data centers running AI consume enormous amounts of electricity, and grid capacity and generation are major constraints on expansion. · Bloom EnergyIts fuel cells enable data centers to generate power on site, bypassing the years-long wait for grid connections that slows AI infrastructure buildout., VistraIt owns power generation assets including nuclear plants and sells electricity that AI data centers are competing for at higher prices., GE VernovaIt manufactures gas turbines and grid equipment needed to add power generation for data centers, with orders backlogged years in advance., VertivIt supplies cooling and power management equipment installed inside every AI data center, and construction demand continues to grow., Constellation EnergyIt operates the largest US nuclear power fleet, providing the constant baseload power that AI data centers require and that large technology companies are contracting for long term.24%
- Advanced AI models require vast amounts of memory and storage, and supply has not kept pace with demand. · Micron TechnologyIt makes high-bandwidth memory that AI chips require, and the take assumes memory supply stays constrained through the end of the decade., SandiskIt manufactures flash storage, which AI data centers need in far greater quantities as model training and inference scales., Western DigitalIt manufactures storage drives used to persist the large datasets that AI systems train on, with cloud customers ordering faster than production can supply.18%
- Companies that own and lease data center capacity benefit directly as demand for AI computing outpaces available supply. · CoreWeaveIt leases AI computing power to AI labs and technology companies, growing as demand for rented capacity outpaces available supply., Nebius GroupIt rents cloud computing capacity to companies training and deploying AI models, growing as demand fills its expanding data center footprint., Core ScientificIt operates data center sites with substantial existing power connections, converting former cryptocurrency mining facilities into AI computing space., IRENIt owns data center facilities with substantial power supplies and rents that capacity for AI computing as customers sign contracts.17%
- As AI generates more software code, tools for monitoring, storing, testing and securing that code see increased usage and revenue. · DatadogIt monitors the health and performance of running software, and AI-generated code creates far more systems that need continuous observation., CrowdStrikeIt protects computer systems and cloud infrastructure from threats, and more AI-written code and autonomous agents create expanded security needs., JFrogIt stores and validates software components before deployment, and AI-generated code creates greater volume of code to manage and check., MicrosoftIt owns GitHub Copilot, one of the most widely used AI coding tools, and operates Azure cloud infrastructure running significant AI workloads.17%
- Semiconductor manufacturing capacity for advanced chips is a critical bottleneck in the AI buildout. · Taiwan SemiconductorIt produces nearly all advanced chips used for AI training and inference, making its factories a critical constraint on capacity., NvidiaIt designs and sells most of the processors on which AI is trained and executed, with demand continuing to grow among major technology companies., BroadcomIt designs custom silicon chips for major technology companies and the networking processors that connect data center infrastructure together., IntelIt is a major US-based manufacturer of advanced semiconductor processors and is opening factory capacity for outside customers to reduce reliance on foreign chip production.14%
- Cash10%