Robinhood Ventures Fund II is not just a bet on AI applications; it is a bet on the machinery underneath them. The fund holds roughly 80 positions financed with $250K SAFEs, concentrates on Y Combinator alumni, and lists on the NYSE on August 13 under the ticker RVII with shares priced at $25. A large share of the portfolio is devoted to AI infrastructure and model startups, the companies providing compute, runtimes, and specialized models to everyone else in the ecosystem.
The compute layer leads the list. Cumulus Compute Labs builds a serverless GPU cloud, letting teams rent capacity for individual workloads instead of reserving machines. The serverless model fits the way modern AI teams actually consume compute: bursty, workload-driven, and hard to predict weeks in advance. Paying for what a job actually uses, rather than what a cluster might need, is the same shift that reshaped application hosting a decade ago.
Expanse Compute approaches the same scarcity from the efficiency side, building an intelligence layer that unlocks wasted GPU capacity. Its pitch is that much of the world’s expensive accelerator fleet sits idle or underutilized, and the company’s layer routes work into that spare capacity. If the thesis holds, Expanse is a way to add supply to the market without building a single data center, which makes it a structurally interesting bet for a fund that wants compute exposure without heavy capital costs.
The runtime layer is another cluster. ReasonBlocks builds a runtime layer for AI agents, the execution environment where agents run, keep state, and coordinate their tools. Agent runtimes are shaping up as the operating system of the agent era, and the fund has placed a position in one of the early builders. The analogy to cloud platforms is direct: whoever controls the runtime can shape how the entire application ecosystem above it develops.
Speedtrain is working on AI systems that run AI infrastructure, automating the operation of the compute itself. The thesis is that infrastructure will increasingly be managed by AI rather than by human operators, and the company builds the systems that make that possible. As GPU fleets grow too large for manual administration, the teams that run them will depend on software that monitors, tunes, and repairs itself.
Virtualization and portability also appear in the portfolio. Smol Machines builds portable subsecond VMs, environments that boot fast enough for agent tasks that need to spin up and tear down in moments. Limrun takes cloud sandboxes to mobile, letting agents run on phones with the heavy lifting done in the cloud. Both companies are attacking the same constraint from different angles: agents need lightweight, fast, and accessible execution environments.
Terminal Use hosts background agents, giving long-running AI workers a place to live while they process work asynchronously. The service is aimed at the growing class of agents that do not finish in a single request but keep working in the background over minutes or hours. Persistent hosting for background agents is a category that barely existed a year ago and is now a requirement for serious agent deployments.
On the model side, JigsawStack builds models for deterministic tasks, the ones where an answer must be exactly right rather than creative. Its models target extraction, routing, and other structured work where reliability matters more than fluency. KelAI builds autonomous AI quants for funds and traders, applying agentic systems to quantitative analysis and trading. The company sits at the intersection of two of the fund’s themes: capable models and software that acts on its own.
What the fund’s infrastructure positions have in common is a bet on the inputs of the AI economy rather than its outputs. Compute clouds, runtimes, background hosting, and portable VMs are all things that get used no matter which model or application wins, and that diversified exposure is the point. For a fund investing small checks across many companies, infrastructure is a way to participate broadly in an industry where the winners are hard to predict.
Robinhood set the offering at $25 a share with an August 13 NYSE debut under RVII. With $250K SAFEs and a Y Combinator lean, the fund’s infrastructure and model holdings give retail investors a portfolio-level view of the plumbing that powers the AI boom. The category mix suggests the fund’s managers expect the value of the AI stack to keep compounding from the ground up.
