AI Eats Billable Hours
hugo alves@huguini
−5.2%
since Sep 30
The idea
I believe AI is eating research and consulting work, and the budgets it frees up land at the model and cloud layer rather than staying with the firms doing the shrinking. So this basket owns the clouds that rent out the models, the chipmakers those models depend on, a slice of pure-play compute rentals, and agent software that competes directly with billable hours. It skips single-company bets on proprietary data moats, since I think public data plus a good model wins out. Cash sits at 10% to add on dips, held with a multi-year view rather than reacting to any single quarter.
What it owns
- Cloud providers that rent AI models to businesses shifting budget from traditional consulting and research services. · MicrosoftIt rents the most widely used AI models through its OpenAI partnership and serves them on Azure, capturing budget shifts from consulting to cloud inference., AmazonIt hosts Anthropic's models on AWS, positioning itself as the cloud destination for inference workloads moving away from consulting budgets., AlphabetIt owns both the Gemini models and Google Cloud, capturing the budget reallocation at both the model and infrastructure layers., OracleIt positioned itself as a GPU landlord for model builders, competing directly for inference workloads and custom AI infrastructure contracts.34%
- Chip designers and manufacturers whose products run the models and inference workloads powering the shift away from billable consulting hours. · NvidiaIt makes the inference chips that every major cloud and model builder currently relies on to run deployed models at scale., BroadcomIt designs custom AI chips for hyperscalers seeking alternatives to standard inference processors, growing as cloud providers build their own silicon., Taiwan SemiconductorIt manufactures the inference chips used across this take, collecting revenue on every major AI processor from every supplier., Micron TechnologyIt makes the high-bandwidth memory that inference chips require to run models, with demand tied to the growth of deployed AI workloads., Advanced Micro DevicesIt supplies inference chips as the credible alternative to dominant suppliers, growing as clouds seek a second source for AI processing power.30%
- Software companies selling products that automate work previously billed by consultants and research firms by the hour. · PalantirIt sells software that automates workflow and decision-making tasks that consulting firms previously billed hours to complete., ServiceNowIt converts workflow automation into AI agents that perform tasks a services team would have billed for, replacing billable hours with software revenue., SalesforceIt prices Agentforce per conversation, selling automated agent labor directly against the billable hours of human consultants., SnowflakeIt serves as the data layer where companies store proprietary information for AI agents to access, replacing traditional consulting workflows.17%
- Pure-play GPU cloud providers capturing the growing demand for raw computing power behind model inference and AI workloads. · CoreWeaveIt rents GPUs purely to meet inference demand from model builders, with revenues directly tied to the growth of deployed AI workloads., Nebius GroupIt rents compute capacity to model builders at faster growth rates than larger competitors, riding the same inference demand wave.9%
- Cash10%