Invest in AI Infrastructure Pre-IPO Companies
AI Infrastructure is one of the six America 2030 sectors. We invest in the layer the AI buildout runs on, not the models on top of it: inference silicon and photonics, built for AI as it moves to live, local, and real-time demand. The model labs get the headlines; we back the picks and shovels.
Why now
The next AI bottleneck is inference, not training.
Inference
As AI moves from centralized training runs to live, local, real-time use, demand shifts to silicon built for inference and the photonics that feed it. That is where we position.
Picks & shovels
The model layer is crowded and commoditizing fast. We invest one layer down, in the infrastructure every model has to run on, where the margins are more durable.
The first phase of the AI buildout was about training: vast, centralized data centers and the race to build the largest models. The next phase is about inference, running those models live, locally, and in real time, everywhere the demand actually sits. That shift changes what matters at the infrastructure layer. The winners are less likely to be the clouds built for training runs, and more likely to be the companies building silicon purpose-built for inference and the photonics that move data with far less energy. The model labs, from OpenAI to Mistral, dominate the headlines and the funding, but the application layer is crowding and commoditizing fast. We invest one layer down, where the margins are more durable and the demand is only beginning to inflect: inference chips and photonic infrastructure. We treat that focus as the thesis, not a hedge.
Sub-sector map
AI infrastructure, mapped from the chip to the model.
America 2030 tracks AI infrastructure across four clusters we invest in, inference silicon, photonics, advanced semiconductors, and quantum hardware, plus one we watch but hold back on today: the model labs, where the layer is crowding and commoditizing.
Inference Silicon
Chips built to run AI live, local, and real-time.
Photonics
Moving data with light, at a fraction of the energy.
Advanced Semiconductors
The energy-efficient silicon under the buildout.
Quantum Hardware
Compute for problems classical silicon cannot reach.
Foundation Models & LLMs
The headline cluster. We cover it closely and hold back: the layer is crowding and commoditizing fast.
Company reports
Where we invest, and what we watch.
We back the infrastructure the AI buildout runs on: inference silicon and photonics, built for AI moving to live, local, real-time demand. We cover the model labs closely, and stay one layer down.
Where we invest
Groq
Deterministic inference silicon built for speed and energy efficiency, purpose-made for running models, not training them.
Research inside →d-Matrix
In-memory compute for generative AI inference, attacking the memory-compute bottleneck in the data center.
Research →Lightmatter
Photonic processors moving data with light, for energy-efficient AI computation at scale.
Research →SambaNova
Full-stack AI systems with a dataflow architecture built for enterprise inference workloads.
Research →SiFive
Energy-efficient RISC-V silicon, an open alternative to proprietary architectures across compute and AI.
Research inside →Rebellions
Full-stack, domain-specific AI processors optimized for energy-efficient inference.
Research inside →VSORA
Inference-first AI chips positioned as a lower-energy GPU alternative for data-center and edge AI.
Research inside →PsiQuantum
A scalable, fault-tolerant quantum computer built on photonics, for problems classical silicon cannot reach.
Research →The model race we're watching
OpenAI
The largest still-private model lab. We cover it closely; the model layer is crowding and commoditizing fast.
Research →Perplexity
AI-native search built on top of frontier models, competing in a fast-commoditizing application layer.
Research →Mistral AI
Europe's open-weight model champion. We track it closely and invest in the silicon it runs on instead.
Research inside →Cohere
Enterprise and sovereign LLMs with retrieval-augmented generation, in the crowding model layer.
Research inside →How to get exposure
Two ways in.
America 2030
Get diversified exposure
Gain exposure to AI Infrastructure alongside the other five America 2030 sectors, Defense, Energy, Robotics, Critical Minerals, and Space, through a single allocation to America 2030, a series of IPO CLUB II LLC. This is the route open today.
Explore America 2030 →Single-Name SPVs
Back one company directly
When an AI infrastructure allocation opens, members can take a concentrated position in a single company through a dedicated SPV. No AI infrastructure single-name is open right now; members are notified inside the Data Room when one is.
Enter the Data Room →Research
Go deeper on AI infrastructure.
Member reports · in the Data Room
AI-Infrastructure Report →
The picks-and-shovels of the AI buildout: inference silicon, photonics, and where durable margins sit. September 2025. Read in the Data Room.
Resilience Capital Stack →
How the six America 2030 sectors fit together, from inference silicon to advanced manufacturing. February 2026. Read in the Data Room.
Humanoids Investing →
Our read on the AI-to-physical-AI frontier, and why we stay disciplined on valuation. November 2025. Read in the Data Room.
FAQ
The best-known model labs, including OpenAI, Perplexity, Mistral, and Cohere, are still private, and IPO CLUB covers them. We hold that access to our members, but our capital goes to the infrastructure layer underneath, inference silicon and photonics, because we think the model layer is crowding and commoditizing fast.
- OpenAI and Perplexity have dedicated research pages; Mistral, Cohere, and others are covered inside the Data Room.
- We cover the model labs closely but invest one layer down, in the infrastructure they run on.
- Membership is free; live allocations require accredited investor or qualified purchaser status.
We research the model labs in depth, but we are not deploying capital into the application layer at today's prices. Our AI infrastructure investments focus on the layer every model has to run on: inference silicon and photonics, built for AI as it moves to live, local, real-time demand.
- Model labs: covered and tracked, approached with valuation discipline.
- Where we invest: inference chips and photonic infrastructure.
- Full model-layer research is available to members in the Data Room.
We believe the model layer is crowding and commoditizing fast, while the infrastructure it runs on is undercapitalized relative to the demand it has to meet. As AI shifts from centralized training to live, local inference, the durable margins move to purpose-built inference silicon and to photonics. That is where we position, one layer down from the headlines.
- The application and model layer is crowded and commoditizing.
- Inference silicon and photonics are undercapitalized relative to demand.
- We back the picks and shovels, not the models on top of them.
Our AI infrastructure coverage centers on inference silicon and photonics: Groq (deterministic inference chips), d-Matrix (in-memory compute for inference), SambaNova (dataflow AI systems), Lightmatter (photonic processors), SiFive (RISC-V silicon), Rebellions and VSORA (energy-efficient inference chips), and PsiQuantum (photonic quantum computing).
- Focus areas: inference silicon, photonics, advanced semiconductors, and quantum hardware.
- Company theses, pricing, and any live allocations are inside the Data Room.
- Access is available through America 2030 or single-name SPVs when open.
IPO CLUB gives accredited investors and qualified purchasers access to private AI infrastructure companies through America 2030, a diversified fund spanning six sectors, and through single-name SPVs when an allocation is open. Becoming a member is free; the next step is registering interest inside the Data Room.
- America 2030 offers diversified exposure across six sectors, including AI Infrastructure.
- Single-name SPVs open periodically; members are notified in the Data Room.
- Membership is free; live deal access requires accredited investor or qualified purchaser status.
There is no single answer, because the right name depends on where in the stack you want exposure. Across the AI infrastructure companies we cover, the inference and silicon names include Groq, d-Matrix, SambaNova, Lightmatter, SiFive, Rebellions, and VSORA, spanning deterministic inference, in-memory compute, dataflow systems, photonic processors, and energy-efficient chips.
- We prioritise inference-era silicon and photonics over training-era GPU clouds.
- Coverage spans inference accelerators, photonics, RISC-V, and quantum hardware.
- Company-by-company detail is inside the Data Room.
A single-name SPV gives concentrated exposure to one company, so the outcome depends entirely on that company. America 2030 spreads exposure across many companies in AI Infrastructure and five other sectors, Defense, Energy, Robotics, Critical Minerals, and Space, trading concentration for diversification. No AI infrastructure single-name is open right now; America 2030 is the route open today.
- Single-name SPVs concentrate risk and return in one company.
- America 2030 spreads exposure across AI Infrastructure and five other sectors.
- Neither approach removes the underlying illiquidity and loss risk of private investing.
Pre-IPO investments are speculative, illiquid, and involve the risk of total loss. AI infrastructure companies carry added risk: hardware and chip programs are capital-intensive and can slip on tape-out, yield, and customer adoption, and the pace of AI change can strand a technical approach. These investments are suitable only for accredited investors and qualified purchasers who can bear a complete loss of capital.
- Pre-IPO investments are speculative, illiquid, and can result in total loss.
- Chip and infrastructure programs add capital-intensity and technology-obsolescence risk.
- Suitable only for accredited investors and qualified purchasers who can bear a complete loss.
America 2030 has a minimum investment of $50,000. Minimums for single-name SPVs vary by deal, depending on availability and secondary supply at the time, and are confirmed inside the Data Room for each live allocation.
- America 2030 minimum investment: $50,000.
- Single-name SPV minimums vary by deal and are confirmed inside the Data Room.
- Availability is subject to secondary supply at the time of investment.
Next step
You've read the thesis. See the deals.
Membership is free and takes two minutes. Inside the Data Room: full company research, current pricing, and live allocations for accredited investors.