Counting the OOMs
Dice@dice
−6.8%
since Oct 6
The idea
I believe we're on a trendline toward AGI by roughly 2027 and superintelligence shortly after, with a trillion-dollar compute buildout along the way. This basket owns every layer of that argument: the chips and memory enabling it, the frontier labs racing to build it, the power and datacenter companies that become the real bottleneck, and the security and government-adjacent names the market still underprices. It's weighted toward those bottlenecks since they're the binding constraint on the timeline, with a small cash reserve to add if the middle years get volatile.
What it owns
- The orders for chips, memory, and manufacturing equipment that make each leap in compute possible are the physical trendline itself. · NVIDIAIt manufactures the processors at the center of the computing advances underlying this take, so its growth depends on sustained spending by the largest buyers., Taiwan SemiconductorIt manufactures every leading-edge AI chip at scale, so the compute arc depends on its fabs staying full and Taiwan remaining investable for years., BroadcomIt designs custom AI chips for the labs and builds the networking fabric that binds massive clusters together, so custom silicon orders must grow alongside the buildout., ASML HoldingIt makes the extreme ultraviolet lithography machines that leading-edge fabs need, so sustained fab expansion drives demand for its irreplaceable equipment., Micron TechnologyIt manufactures high-bandwidth memory that every AI accelerator requires, so memory staying supply-constrained through the buildout matters to the margin story., Advanced Micro DevicesIt competes as a second GPU supplier to prevent lock-in and meet explosive training demand, so it must keep landing real AI chip orders from major labs.33%
- Power generation and datacenter infrastructure become the constraint long before chips do, and the market has priced this layer lightly. · GE VernovaIt builds the gas turbines and grid equipment that new datacenters require, so the power buildout staying on pace keeps its backlog growing., Vertiv HoldingsIt makes liquid cooling systems for dense GPU racks that would otherwise overheat, so power density keeps climbing and it holds its market share., EQT CorporationIt operates the largest natural gas field in the country and new datacenters will burn significant gas, so AI demand must translate into real gas consumption., EatonIts switchgear, transformers, and power distribution systems go into every new datacenter before any GPU turns on, so construction activity stays steady., Constellation EnergyIts nuclear plants provide clean, dispatchable power that hyperscalers are already contracting for, so long-term datacenter power agreements keep materializing., VistraIt sells power in markets where datacenters are clustering, so compute demand outpacing new generation keeps electricity prices elevated in its regions., CoreWeaveIt builds GPU datacenters and rents them to labs, making it the most direct play on the buildout itself, though its success depends entirely on sustained rental demand and debt management.29%
- The frontier labs are private, so the public companies that own stakes in them or provide their compute are the closest access to the race. · AlphabetIt owns DeepMind and manufactures its own TPU chips, mattering if compute density and in-house silicon prove critical to staying at the frontier., MicrosoftIt owns a stake in OpenAI and operates Azure, positioning it as the public way to own the leading frontier lab and its compute platform., Meta PlatformsIt is investing at frontier scale in models and computing infrastructure while operating a cash-generative business, which matters for staying competitive in this take., AmazonIt holds a stake in Anthropic and operates AWS, positioning it to profit from both a frontier lab and the compute infrastructure others train on.19%
- If advanced AI becomes a national security priority, the security and defense contractors selling to government and labs capture that spending. · Palantir TechnologiesIt provides software for government and defense AI systems, so spending on national security AI projects keeps growing and justifies its valuation., CrowdStrikeIt provides security infrastructure for model weights and classified compute, so government and lab spending on AI security must become a real budget line., Kratos DefenseIt makes autonomous defense systems and target platforms for government spending on AI-era military capability, so defense budgets must shift toward autonomous procurement.11%
- Cash8%