The Vibecoding Trade
dragos@dragos
−4.3%
since Sep 30
The idea
I believe software's old subscription moat is cracking as AI lets anyone build their own tools in an afternoon, so this basket skips classic software entirely and owns the side doing the disrupting. Most of the weight sits in chips and data-center buildout plus the model builders and platforms powering it all, since that is where spend lands first, with a smaller sleeve in companies whose proprietary data is too hard to rebuild around as the slower, structural piece. Cash stays at 10% to use on dips in what is a genuinely volatile theme.
What it owns
- Every vibecoded app runs on AI chips, so this is where the infrastructure spending lands first. · NVIDIAThe default chip for training and serving AI models across the industry remains the foundation of AI infrastructure spending., Taiwan SemiconductorAlmost every advanced AI chip gets manufactured here, making it the production bottleneck regardless of which chip maker wins., BroadcomBig clouds pay it to design custom AI chips to diversify away from single-supplier chip risk, so orders rise as total AI spending climbs., Micron TechnologyAI chips need vast amounts of fast memory strapped directly to them, and supply from this producer stays tight as demand grows., Advanced Micro DevicesIt is the only viable second source of AI chips for clouds that need to escape single-supplier pricing, so wins matter as much as NVIDIA's.31%
- These companies build and sell the AI models that replaced traditional software subscriptions. · MicrosoftIt owns the most widely used AI copilots and the Azure cloud where models run, positioning it to capture infrastructure spending if AI drives real cloud growth., AlphabetIt builds its own models, its own chips, and operates its own cloud, capturing value across the entire AI stack if those businesses execute., Meta PlatformsIt licenses its free open models to fuel the vibecoding wave while its ad business keeps user data that can't be rebuilt overnight., AmazonMost vibecoded tools run on AWS after creation, so it collects rent from the trend regardless of which model or chip wins the race., OracleIt converted itself into a landlord for AI training with massive multi-year cloud contracts that convert into actual revenue.30%
- Data built over decades keeps its pricing power even as the tools for building new data become free. · S&P GlobalIts ratings and index data took decades to build and can't be vibed together in an afternoon, keeping pricing power across market cycles., Intercontinental ExchangeIts market data is generated by its own exchanges and can't be copied by any tool, keeping subscriptions resilient through tech change., Verisk AnalyticsIts insurance data comes from decades of claims that can't be rebuilt, keeping subscriptions resilient as AI tools become cheaper., RedditIt owns twenty years of human conversation that model builders license for training, capturing value on both sides of the AI wave.15%
- AI compute needs data centers, networking and power gear, a bottleneck the headline story skips. · Arista NetworksIt wires thousands of AI chips together into single computers for cloud builders, so orders rise as clusters grow larger., Vertiv HoldingsAI chips run hot and data centers need its cooling and power infrastructure, so orders flow with construction growth., CoreWeaveIt rents GPU and AI computing infrastructure to model builders who face chip shortages, so its growth depends on compute remaining scarce and its largest contracts staying intact., EatonEvery new AI data center needs serious electrical gear, making this the slowest-moving but steadiest piece of infrastructure buildout.14%
- Cash10%