−5.2%
since Sep 30
The idea
I don't think anyone can call which AI lab wins, so I'd rather own what gets paid regardless of the outcome. This basket leans into the toll booths of the agentic economy: neutral rails that carry and log the traffic, the payments and identity layer agents need to transact, a spread of chipmakers instead of one compute monopoly, and software rebuilding itself around agents as users. No single stock dominates, cash sits on the side, and the whole thesis breaks if one lab ends up controlling the entire stack top to bottom.
What it owns
- These companies provide essential infrastructure that AI agents will rely on regardless of which labs or models dominate, as long as that traffic stays on open networks and platforms rather than moving into closed proprietary systems. · CloudflareIt sits between users and the internet, routing and securing traffic, so more AI agents making requests means more paid traffic flowing through its network., MongoDBIt is a default database for new applications, so agents generating and querying data at machine speed would create more database workloads., DatadogIt monitors AI systems running in production and charges by the volume it watches, so more agents doing more things would increase monitoring costs., ElasticIt provides the search layer that lets AI systems pull relevant answers from a company's own documents, which most AI applications need., TwilioIt carries text, calls, and emails between agents and humans, charging per message as more agents communicate with more people.30%
- These companies enable agents to transact on behalf of humans, whether through crypto payments, identity verification, or traditional card networks, assuming machine commerce uses these existing channels rather than routing around them. · CoinbaseIt is the regulated US venue where crypto transactions settle and assets get custodied, so agent-driven cryptocurrency payments would route through it., OktaIt verifies who agents are and what they are allowed to do, building the identity layer that agents need before they can act on anyone's behalf., CircleIt issues USDC, the stablecoin positioned as money for machines to move, and earns yield on the reserves backing each dollar in circulation., VisaIt processes card transactions regardless of who built the agent, taking a cut each time an agent buys something with traditional money.24%
- These companies benefit when multiple vendors build AI chips and silicon instead of a single architecture dominating, since buyers want alternatives to any one supplier and these companies get paid across that fragmented landscape. · Advanced Micro DevicesIt is the main alternative to Nvidia for AI compute, and chip buyers want a second source to avoid dependence on a single vendor., BroadcomIt builds custom AI chips for large labs and the networking gear connecting them, getting paid by buyers trying to avoid dependence on a single chip supplier., Arm HoldingsIt licenses chip designs to nearly every company building AI silicon, collecting royalties as long as chipmaking stays fragmented across many vendors., QualcommIt powers on-device AI inference on phones and laptops, collecting fees from every device that runs AI at the edge instead of in data centers.22%
- These companies power the platforms where agents will operate and where enterprises will run their data and software, assuming companies keep their AI workloads on neutral platforms rather than consolidating into one lab's ecosystem. · ShopifyIt powers the storefronts agents will shop from and is building interfaces that let agents browse and buy directly from merchants., SnowflakeIt organizes enterprise data where agents can run AI against it without moving everything into one lab's closed ecosystem., GitLabIt provides the pipeline for code review, testing, and deployment that agent-written code still needs to go through before it runs in production.14%
- Cash10%