AI Writes the Code
Fredrik Björk@fbjork
−8.5%
since Oct 7
The idea
AI is already writing real production code, and I think that shift is early and lasts for years. Rather than guess which app or startup wins, I'm betting on the companies that get paid regardless: the coding tool makers, the cloud and model platforms underneath them, and the chip and data center hardware powering all of it. The basket is broad and anchored in large caps, with smaller higher-upside slices in pure-play tool sellers and infrastructure suppliers, no single name dominating, and a modest cash reserve set aside to add on dips.
What it owns
- Every AI model that writes code trains and runs on these processors, the scarcest layer of the whole stack. · NVIDIAIt manufactures the processors that every AI coding model trains and runs on, holding its position at the center of compute spending as AI adoption grows., Taiwan SemiconductorIt manufactures advanced chips for all the major AI chip designers, so it gets paid regardless of which processor company wins., BroadcomIt builds custom AI chips for cloud companies that want an alternative to buying from a single supplier, capturing a share of the broader chip buildout., Advanced Micro DevicesIt manufactures AI chips as a credible alternative to the dominant supplier, allowing buyers to avoid single-source dependence as compute demand scales., Micron TechnologyIt manufactures high-bandwidth memory that AI models require in enormous quantities, capturing revenue as data center memory demand grows.33%
- AI coding runs on rented cloud computing and models, so these platforms get paid on every line generated. · AlphabetIt operates both a cloud platform and an AI coding model, collecting revenue from two layers as AI development tools run on its infrastructure., AmazonIt operates AWS, the platform hosting a large share of AI tools and startups, getting paid for compute regardless of which coding tool wins., OracleIt sells cloud capacity directly to the model companies powering coding tools, collecting revenue concentrated on the compute demand driving the buildout.19%
- The companies selling the AI tools that let teams ship the same software with fewer engineers. · MicrosoftIt owns GitHub Copilot, the AI coding tool closest to real production use, getting paid per seat as long as companies keep adopting AI-assisted development., GitLabIt sells an AI-powered development platform directly to engineering teams, capturing higher per-seat value as teams do more with fewer people., AtlassianIt sells workflow and project management tools to engineering teams, collecting revenue as leaner teams still need planning software with AI features.18%
- AI compute needs buildings, power systems and networking, and these companies supply that physical layer. · Vertiv HoldingsIt supplies the cooling and power infrastructure that AI data centers require, supporting the physical layer underneath all compute buildout., EatonIt supplies the electrical power systems that AI data centers depend on, providing a large-cap way to own the infrastructure buildout., Arista NetworksIt builds the networking hardware that connects AI chips inside data centers, supporting the infrastructure layer as clusters expand.13%
- More chips means more of the machines that make chips, a quieter bet one step behind the headlines. · ASML HoldingIt manufactures the lithography machines required to make advanced AI chips, capturing revenue on every new fab that opens as the buildout continues., Applied MaterialsIt sells the manufacturing equipment that chip factories use to produce processors, capturing revenue on semiconductor capital spending across the buildout.9%
- Cash8%