The Full Stack Squeeze
ishan@jpegmogul
−13.4%
since Oct 1
The idea
I believe AI infrastructure and everyday consumer demand are converging on the same computing bottleneck, so I built exposure across the whole stack: memory and chips get the largest weight since that squeeze looks tightest, while content delivery and cybersecurity provide steadier growth tied to usage regardless of which chipmaker leads. I avoid concentrating in any single name within a layer and keep a small cash reserve to add to the more volatile chip and memory names on pullbacks.
What it owns
- A handful of companies control the design and manufacturing of advanced AI silicon, creating a concentrated pipeline that all AI demand flows through. · NVIDIAIts AI accelerators are the default choice for data center training and inference, making it the processor most AI demand flows through., Taiwan SemiconductorIt manufactures nearly every advanced chip in this take, from NVIDIA to AMD to Qualcomm, making it the single tightest point in the plumbing., BroadcomIt builds custom AI chips and networking silicon for hyperscalers that want alternatives to relying only on NVIDIA., ASML HoldingIt is the only company making the EUV machines needed to build advanced chips, so every fab expansion requires its equipment.24%
- AI chips are starved for high-bandwidth memory and storage, making memory the tightest bottleneck in the entire computing stack. · Micron TechnologyIt is one of only three companies in the world making high-bandwidth memory, the component AI chips are most starved for., Western DigitalIt sells the mass storage that holds data for AI training and inference, the capacity that grows with data center buildouts., SanDiskIt makes NAND flash storage inside phones and data centers, catching both the AI demand and consumer adoption sides of the squeeze., Seagate TechnologyIts hard drives provide the cheap bulk storage layer that AI data centers need in massive quantities., RambusIt makes the interface chips and licenses the technology connecting fast memory to processors, so it earns from the memory squeeze regardless of who wins.23%
- General-purpose processors run every server and device, and AI workloads are forcing an upgrade cycle across all computing platforms. · Advanced Micro DevicesIt is gaining server CPU share from Intel and pushing into AI accelerators, providing a second source for processors in data centers., Arm HoldingsIt collects royalties on nearly every phone chip and a growing share of server CPUs, so it earns from both consumer and data center AI adoption., QualcommIts chips are in most phones, so if on-device AI drives phone upgrades, it benefits from mass consumer adoption of the technology., IntelIt is the lower-cost option for processors, and a successful turnaround could give it room to capture share in a tight chip market.16%
- Mass consumer adoption means AI output travels over edge networks and data center switches to reach users, pulling that infrastructure harder. · CloudflareIt runs the edge network that delivers AI to phones and browsers, the infrastructure that gets pulled when usage grows., Arista NetworksIt sells the switches that move data inside AI data centers, so every new cluster built requires more of its networking gear., Akamai TechnologiesIt is the incumbent content delivery network that grows with traffic usage, plus a growing security business that benefits from the squeeze., CienaIt builds the optical networking gear that carries traffic between data centers and to consumers, the literal pipes of the plumbing.16%
- More computing and more users means more to protect, and security spending grows with usage regardless of which chip wins. · CrowdStrikeIt is the leader in protecting devices and workloads, and security spending grows with usage regardless of which chipmaker wins., Palo Alto NetworksIt consolidates security spending onto one platform, so companies deploying more AI route more of their security budgets to it., ZscalerIt secures traffic between users and the cloud, the path that grows when AI reaches mass consumer adoption.13%
- Cash8%