The Plumbing Under the Apps
Cat McGee@cat
−4.6%
since Sep 29
The idea
I believe building apps has gotten so cheap and easy that the real money moves down the stack, especially as more usage shifts from people to AI agents. So I own the plumbing everyone depends on regardless of which app wins: chips and hardware, cloud platforms and models, the power and data centers running it all, plus smaller bets on the rails agents will use to transact and the data pipes connecting it.
What it owns
- Every application, built by humans or agents, runs on the same processors, so chipmakers collect revenue regardless of who wins at the software layer. · NVIDIAIt supplies the processors that nearly all AI workloads run on today, collecting revenue as long as agent and model demand for computing power keeps growing., Taiwan SemiconductorIt manufactures the advanced chips for almost every major chip designer, so it collects revenue no matter which competitor wins the design race., BroadcomIt builds custom AI chips for major cloud providers and the networking equipment that connects data centers, benefiting from cloud giants' expansion efforts., Advanced Micro DevicesIt provides the alternative compute chip option for data centers, filling the role of a credible second supplier to keep the market competitive., Marvell TechnologyIt manufactures chips that move data within and between AI data centers, a toll that every application pays regardless of whether it was built by a person or deployed automatically.26%
- Agents operate on cloud platforms and pay per API call for models, so usage grows even as individual applications become cheaper or more abundant. · MicrosoftIt rents both cloud infrastructure and model access through Azure, so metered usage revenue continues regardless of whether applications above become commoditized., AlphabetIt controls the full stack of chips, models, and cloud infrastructure, collecting revenue at multiple layers as long as its AI services gain adoption., AmazonAWS dominates the cloud infrastructure layer and generates metered revenue whenever agent applications call APIs all day., OracleIt has signed major multi-year contracts to rent AI compute, a direct bet on the principle that compute infrastructure captures value regardless of who builds applications., CoreWeaveIt rents GPU compute capacity, expressing the core belief that value concentrates in infrastructure beneath software applications.24%
- Agent activity requires electricity and physical infrastructure, bottlenecks that get paid before any software layer above them. · VertivIt supplies the cooling and power systems that every data center must install before running any compute, getting paid on the infrastructure buildout., EatonIt makes the electrical distribution equipment that data centers require, one of the physical bottlenecks that limits buildout speed., Constellation EnergyIt supplies nuclear power directly to data center operators running applications around the clock, benefiting from long-term power contracts at premium rates., VistraIt sells electricity that data centers cannot operate without, sitting at the foundation of the stack as power demand from AI outpaces new supply., Digital RealtyIt owns the physical buildings housing the AI infrastructure and collects rent from every layer of the stack that sits above it., GE VernovaIt builds the turbines and electrical grid equipment needed to deliver more power to data centers, working at the deepest layer of physical infrastructure.22%
- Autonomous agents transacting independently will need payment and money-movement infrastructure, a layer forming at the intersection of finance and crypto. · CoinbaseIt operates the largest US crypto payment rails, which would carry transactions if autonomous agents conduct business with each other using stablecoins., VisaIt is building infrastructure for AI agents to spend money independently, positioning the card network as the mainstream layer for agent payments., Circle Internet GroupIt issues USDC stablecoin, which agents would likely use to transact with each other if autonomous agent payments become a significant usage pattern.13%
- Agent traffic flows through networks and databases that charge based on usage volume, not on the number of human users. · DatadogIt charges customers based on usage volume rather than per-seat, so a world of agent-built applications generating continuous machine activity accelerates its growth., CloudflareIt controls traffic routing for a large portion of internet traffic and is building gateways that determine which agents and crawlers gain access, functioning as a tollbooth.8%
- Cash7%