· 4
+1.3%
since Sep 2
The idea
I don't think any single AI lab wins the whole stack, so instead of betting on OpenAI, Anthropic, or Google, I own the toll booths in between: neutral infrastructure that gets paid on AI traffic regardless of whose model wins, plus payments and identity layers agents need to act, diversified chipmakers, and app companies rebuilding around agents. No single stock dominates the mix, and I keep a slice in cash. I'd rethink this if one lab actually captured the entire stack top to bottom instead of the field staying fragmented.
What it owns
- Infrastructure that AI applications rely on regardless of which model provider or chip vendor wins. · CloudflareSits between users and model providers, collecting fees on AI traffic regardless of which lab wins, with AI gateway and edge compute products designed to capture inference workloads., MongoDBIts document model and vector search capabilities serve as a default database for AI applications, with growth tied to Atlas adoption by developers actually building production systems., DatadogEvery company running AI in production needs to monitor it, and Datadog collects fees for that regardless of whose model is underneath., TwilioAgents that talk to customers need messaging and voice channels, and Twilio owns those rails plus customer data to personalize them, model-agnostic., ElasticHandles the retrieval step for AI applications that fetch relevant documents before the model answers, competing against vector database startups and cloud providers' built-in options.30%
- Payment and identity systems built for machine-to-machine commerce and agent verification. · CoinbaseCryptocurrency rails that enable autonomous agents to transact with each other without requiring human bank accounts, betting on machine-to-machine payments flowing through its infrastructure., OktaPositioned to own agent identity verification as autonomous systems act on behalf of users and require proof of identity and permissions., VisaIf agents pay with regular money instead of crypto, it flows over card networks, and Visa is already building agent-specific payment credentials., CircleStablecoins settle instantly without a bank in the loop, making them the most plausible money for machine-to-machine payments, and Circle's USDC is the leading regulated stablecoin.24%
- Processing models on devices where end users control the hardware creates a different set of winners than centralized cloud inference. · Advanced Micro DevicesThe primary alternative to Nvidia for data center AI compute, winning market share from customers unwilling to depend on a single chip vendor for inference and training workloads., BroadcomBuilds custom AI chips for hyperscalers that want proprietary silicon and the networking that ties clusters together, profiting from chip vendor diversification., Arm HoldingsLicenses chip designs to nearly every AI silicon maker and collects royalties on chip diversity itself as custom AI chips proliferate., QualcommQualcomm makes chips that run models locally on phones and laptops, capturing value if on-device inference becomes a real reason people upgrade hardware.22%
- Software systems that agents interact with directly stand to capture new value as machine workloads grow. · ShopifyMerchants need structured product data and machine-readable checkout for shopping agents, and Shopify is building that for millions of sellers to remain the default agent transaction point., SnowflakeLets enterprises run any model against their own data without shipping it to a provider's platform, with growth tied to AI products driving real consumption rather than data leaving the warehouse., GitLabCode is where agents are already doing real work, and the pipeline around code review, security, and deployment becomes more critical as agents write more of it.14%
- Cash10%