The idea
I don't think any single AI lab wins the whole stack, so instead of betting on OpenAI versus Anthropic versus Google, I own the layers that get paid no matter who is ahead: neutral rails like Cloudflare, MongoDB, and Datadog, the payments and identity plumbing agents need to transact, a spread of chipmakers rather than one architecture, and application companies rebuilding around agents instead of getting replaced by them. No single stock dominates, and I keep a cash buffer. I'd rethink this if one lab actually captured the entire stack top to bottom.
What it owns
- These companies sit between users and AI models, collecting fees on traffic regardless of which model provider wins or how the technology evolves. · CloudflareIt sits between users and every model provider, so it gets paid on AI traffic no matter which lab wins, as long as its AI gateway and Workers products keep growing with inference at the edge., MongoDBAI applications need a database regardless of which model they call, and its document model plus vector search make it a default choice for developers building them., DatadogEvery company running AI in production needs to monitor it, and the platform collects fees for that monitoring regardless of whose model is underneath., ElasticMost useful AI apps retrieve relevant documents before the model answers, and its search and vector tools serve that retrieval step for any model., TwilioAgents that communicate with customers need messaging and voice channels, and the platform owns those rails plus customer data to personalize them across any model.30%
- These companies provide the payment and identity infrastructure that autonomous agents need to transact with each other and on behalf of people. · CoinbaseAutonomous agents that transact with each other naturally fit crypto rails because they don't require a human with a bank account, and this bet on agent payments depends on them actually materializing rather than staying in demos., OktaWhen agents act independently, someone must verify the agent's identity and permissions, and the company is positioned to own agent identity as a product category., CircleIts USDC is the leading regulated stablecoin, and stablecoins are the most plausible money for machine-to-machine payments because they settle instantly without a bank in the loop., VisaIf agents pay with regular money instead of crypto, it flows over card networks, and it is already building agent-specific payment credentials to win if agent commerce grows on existing rails.24%
- These companies benefit from competition in AI chips and silicon, winning as buyers refuse to depend on a single vendor. · Advanced Micro DevicesIt is the most direct bet that AI compute buyers will refuse to depend on a single chip vendor, holding up as long as big inference buyers commit real money to MI-series chips as a Nvidia alternative., BroadcomIt builds custom AI chips for companies that want their own silicon instead of buying whatever one vendor sells, plus the networking that ties clusters together, holding up as long as hyperscalers keep expanding custom chip programs., Arm HoldingsIt licenses chip designs to nearly everyone building AI silicon, so it collects royalties on chip diversity itself as long as custom AI chips keep getting built on its designs and royalty rates step up with newer architectures., QualcommIt is the bet that a lot of inference moves onto phones and laptops where no lab controls the chip, holding if on-device AI becomes a real reason people upgrade hardware and it keeps its lead in chips that run models locally.22%
- These companies build tools that agents use to accomplish real work on behalf of people or businesses. · ShopifyIf shopping agents buy on people's behalf, they need structured product data and checkout that machines can use, and it is building exactly that for millions of merchants, holding up if it stays the default place agents transact with independent sellers., SnowflakeIt lets companies run AI against their own data by bringing any model to the data instead of shipping data to external providers., GitLabCode is where agents are already doing real work, and it is the bet that the pipeline around code, review, security, and deployment matters more when agents write more of it, so watch whether revenue per customer grows as agent coding spreads.14%
- Cash10%