AI Becomes Work Infrastructure
Tom Creighton@tomcreighton
−5.1%
since Oct 1
The idea
I believe AI assistants are turning into everyday work infrastructure, the way email once did, taking over routine tasks while judgment becomes more valuable. Since the model makers are private, I built exposure to what gets paid regardless of who wins: chips and compute, the clouds hosting these models, the power and networking behind them, and work platforms with real distribution. Large-cap names carry most of the weight, smaller positions cover faster-moving infrastructure, and I keep a small cash reserve to add on weak days. I see this as a multi-year shift, not a quick trade.
What it owns
- Every AI assistant deployed at work runs on chips that someone has to design and manufacture, and spending on those chips grows as assistants become standard across workplaces. · NVIDIA CorporationIt manufactures the chips that train and run most frontier AI models, giving it a default position that persists as long as training and inference remain concentrated on its architecture., Taiwan Semiconductor ManufacturingIt manufactures nearly every advanced AI chip, including those from competitors, so it collects revenue across the entire buildout regardless of which chip designer wins, provided it maintains its lead in advanced production., BroadcomIt manufactures custom AI chips for major clouds seeking to reduce their dependence on external vendors, gaining a second revenue stream alongside NVIDIA as long as clouds continue investing in proprietary silicon., Advanced Micro DevicesIt serves as an alternative chip supplier to cloud providers and AI labs seeking to reduce vendor concentration, gaining a slice of the buildout as long as it continues winning meaningful orders., ASML HoldingIt is the sole manufacturer of machines needed to produce the most advanced chips, so demand for AI chip capacity translates into orders for its equipment, provided its monopoly position holds., Micron TechnologyIt manufactures advanced memory chips that must be installed alongside AI processors to enable high performance, benefiting from AI demand as long as memory remains a capacity constraint.36%
- The private labs building frontier models need public cloud providers to host and run them, allowing those providers to collect revenue on every workplace assistant deployed through their infrastructure. · MicrosoftIt hosts OpenAI's models on its cloud and embeds assistants directly into Office, where workplace workflows already live, positioning it to collect rent on adoption within established software., AlphabetIt builds its own frontier models and operates its own cloud and chips, capturing value across the stack even if it is not the assistant on every desk, as long as its models remain competitive., AmazonIt backs Anthropic and hosts Claude on its cloud for commercial use, positioning it to capture AI workload growth through a major frontier model without needing to own the model itself., OracleIt has secured major contracts to rent massive computing power to AI labs and model developers, capturing value through a backlog of deals as long as those contracts are honored and utilization grows.28%
- A decade of AI growth requires massive buildouts in data center power, cooling, and networking infrastructure, a capital-intensive bottleneck that receives less attention than the headline story of model capabilities. · Eaton CorporationIt manufactures electrical distribution and power management systems that connect data centers to the grid, benefiting from growing electrical demand as AI infrastructure expands., Arista NetworksIt supplies the networking equipment that connects thousands of AI chips within data centers at high speed, a requirement that persists as long as cloud providers continue building out AI infrastructure., Vertiv HoldingsIt provides cooling and power distribution equipment essential to AI data centers, capturing recurring revenue from every facility expansion as long as data center construction continues.15%
- Software platforms that own the workflow and control access to company data can charge for built-in AI assistants, whereas apps that bolt features on top face faster displacement. · ServiceNowIt operates the workflow platform where handoff work lives inside large enterprises and can embed AI assistants directly into those workflows, capturing value as long as customers pay for its AI features., Palantir TechnologiesIt specializes in deploying AI models within real company and government operations to automate grunt work and support judgment, capturing value as long as it wins large contracts for that specific use case., SnowflakeIt stores and manages enterprise data that AI assistants need to access in order to be useful at work, benefiting as long as companies consolidate data within it and use that data to power AI., SAP SEIt operates the finance and operations systems where most large enterprises run, giving its AI assistants distribution that startups cannot match, provided customers continue adopting its cloud and AI tools.14%
- Cash7%