Compute
Make intelligence more accessible.
The infrastructure that lowers the cost of intelligence and makes it usable at scale.
- Inference optimization
- GPU marketplaces
- Confidential inference
- Training efficiency
IOSF — THE FRONTIER FUND
RESEARCH. THESIS. CONVICTION.We back the founders turning frontier research into the next generation of AI companies.
Explore our thinkingMake intelligence more accessible. Inference, capacity and the infrastructure that makes more possible.
ExploreTurn capability into agency. The environments, tools and controls that put agents to work.
ExploreBuild for the real world. Products that complete useful work and create value for customers.
Explore01 / Our thesis
The next chapter
of AI is useful work.
Our thesis begins with a shift: from training models to putting intelligence to work. As inference becomes more accessible, new kinds of products and workflows become possible.
Agents connect that capability to productivity. They need efficient compute, reliable infrastructure and real problems to solve. We invest across all three.
THE FOUNDATION
Accessible intelligence.
THE ENABLER
Capable agents.
THE OUTCOME
Measurable productivity.
02 / Investment focus
One stack.
Three connected layers.
Agent infrastructure
Help agents work better.
The environments, tools and controls that let teams build, test and deploy capable agents.
- Sandboxes & runtimes
- Tool governance & MCP
- Evaluation environments
- Efficient world models
Agent applications
Turn capability into useful work.
Products that complete real tasks, fit into existing workflows and create measurable value for customers.
- Enterprise workflows
- Financial services
- Developer tools
- Vertical agents
BEYOND SOFTWARE
AI hardware
Where a physical device unlocks adoption, we also support the hardware and software that bring intelligence into everyday use.
03 / How we think
Research becomes conviction.
Research first.
We go deep into technical shifts and work with researchers to understand what is becoming possible before the market has a name for it.
Efficiency matters.
We look for technology that removes a meaningful cost or constraint for the products built on top of it.
Technical depth.
We value founders who understand the science, contribute to the tools and can turn a difficult problem into a working product.
Real demand.
We ask what work a product completes, who needs it and why they will pay. Customer value is the foundation of our conviction.
04 / For founders
A network
we build
with you.
We share research, make useful connections and help technical founders turn ambitious ideas into enduring companies.
Connect with customers
Warm introductions, business partnerships and practical support for enterprise sales.
Build with better resources
Connections to compute providers, model teams and technical partners across our network.
Learn alongside builders
Focused workshops and direct conversations with researchers and experienced founders.
Bring hardware into the world
Hands-on connections to the Shenzhen supply chain, from engineering validation to production.
05 / Our people
Technical curiosity.
Shared conviction.
Jocy Lin
Backing technical infrastructure founders since 2017. Jocy brings the experience of building IOSG and supporting founders through successive technology cycles.
Yiping Lu
Works at the intersection of enterprise AI adoption, open-source development and venture investing, with experience across AI, crypto and fintech infrastructure.
Research in the room.
Researchers and AI founders help us test products, examine technical assumptions and understand what it takes to bring research into practice.
The G was Genesis.
The F is Frontier.
In 2017, the IOSG team began backing founders building open, programmable systems. IOSF carries that research discipline into the next frontier: intelligence itself.
Discover IOSG06 / Research perspective
THE INFERENCE ECONOMY
From models that answer
to agents that act.
Read our perspectiveClose perspective
THE INFERENCE ECONOMY
From models that answer
to agents that act.
The value of AI becomes tangible when it changes how work gets done. We focus on the systems that connect model capability to that outcome: compute that makes deployment practical, infrastructure that makes agents reliable and applications that solve a customer's problem.
These layers reinforce one another. More efficient inference can support more demanding workflows. Better runtimes and evaluation help teams move from a prototype to a dependable product. Applications reveal which capabilities actually matter in practice.
Our starting question is simple: what useful work does this technology enable? We build conviction by tracing the answer from the underlying research to the people who use it.
FOR FOUNDERS & RESEARCHERS