Inside the NVentures portfolio

After NVIDIA’s latest shopping trip through Europe, I wanted to take a closer look at what it’s backing. Its October announcements include healthcare, materials, robotics and enterprise software, with some investments still at the planning stage. The range made me curious about how these businesses fit together.

So I focused on NVentures, NVIDIA’s venture capital arm, and mapped the 125 companies in its public portfolio directory: what they build, where they are based and how they connect to NVIDIA.

Inside the basket

Portfolio by theme

The biggest group is Physical Systems & Engineering, accounting for just over a quarter of the portfolio. It covers robotics, autonomous vehicles, manufacturing, engineering software, materials discovery and chip design. Healthcare & Digital Biology has 23 companies, or 18.4% of the directory.

Almost half of NVentures’ portfolio is about making things happen in the physical world: designing molecules, controlling machines, inspecting factories, modelling physics and supporting clinical decisions.

The software side is also more varied than a broad “foundation models” label suggests. Lovable helps people build apps. CodeRabbit reviews code. Factory develops software-engineering agents. Their value sits partly in making models useful within a workflow.

Inception Labs, Sarvam, Nous Research and Reflection AI are building language models. World Labs, Odyssey, Runway and others explore visual intelligence, video and simulated environments. Harmonic focuses on mathematical reasoning and formal verification.

Three ways the portfolio connects

New demand for computing

Drug discovery, robotics, engineering and generative media create new markets for accelerated computing. As these products gain customers, they could expand demand for NVIDIA’s technology across more industries. The scale depends on how widely they are adopted and how much computing each application requires.

Making AI practical to deploy

For AI to work inside a business, it needs access to the right data and software.

Unstructured prepares documents for AI systems. Exa gives agents access to web search. n8n connects models to business tools and automates tasks. Coding platforms help teams write, review and maintain software.

These companies make AI easier to adopt in existing processes. Their advantages may come from proprietary data, integration, customer relationships or ownership of an important workflow.

Removing constraints on growth

More computing requires reliable electricity, fast connections between systems and the capacity to manufacture equipment at scale.

Optical networking improves data transfer, energy software and battery storage help manage power availability, and manufacturing tools improve production quality. Advanced nuclear and fusion developers could provide additional electricity over a longer timeframe. All of these pushes the limit of computing infrastructure.

Why take money from NVentures?

For a startup, the appeal can extend beyond the cheque.

NVIDIA describes NVentures’ support through technical expertise, platform integration and commercial relationships.

Skild uses NVIDIA’s Isaac Lab, Omniverse and Cosmos for robot simulation and training data, helping it develop intelligence that can work across different machines and environments.

CodeRabbit receives technical and commercial support. Its CEO says NVIDIA uses the product internally, helped integrate Nemotron models and made introductions across its enterprise network.

Basecamp Research uses BioNeMo to develop biological AI models, bringing NVIDIA’s software into research built around specialist biological data. Turning that research into effective medicines still requires experimental and clinical validation.

These relationships can help startups develop products and reach customers. For NVIDIA, they can expand adoption of its technology while offering potential returns on its equity investments. The support varies by company. An investment does not automatically provide free compute, guarantee sales or require the startup to use NVIDIA products.

Following the money

The potential loop is straightforward: NVIDIA funds a startup, the startup spends on compute, and demand for that compute supports purchases of NVIDIA hardware. The portfolio directory does not quantify these flows, and cloud spending includes much more than GPUs.

Portfolio companies can also help improve the infrastructure itself.

PhysicsX is working with Siemens on AI-assisted design of data-centre power-distribution systems using NVIDIA technology. AI is helping engineers design the equipment needed to run more AI.

Emerald AI works with NVIDIA and energy companies to make computing workloads more responsive to grid conditions, potentially unlocking capacity for further deployment.

Instrumental helps NVIDIA and its partners detect manufacturing failures early and improve production of AI trays and racks, according to its CEO.

The feedback therefore extends beyond compute purchases. Portfolio companies can create demand for NVIDIA’s technology while helping address the power, engineering and manufacturing constraints on its growth.

The future this system suggests

Over the next few years, I expect more AI to move into factories, warehouses and research labs, while businesses adopt it through software they already use. Robots will need to save money and work reliably, and scientific models will need to produce results that hold up in the lab. That could create more demand for NVIDIA’s computing across industries, with opportunities for both model developers and companies building useful applications.

Power supply and network capacity will increasingly shape how quickly that demand can be served. Energy management and storage could help sooner, while new nuclear, fusion and quantum technologies remain longer-term bets. Europe could build stronger AI businesses while becoming more reliant on NVIDIA’s platform. NVIDIA has several ways to benefit from this growth, but lasting demand will depend on whether the companies it backs deliver enough value for customers to keep paying.