The AI & Quantum Spiral
Data, models, chips and sensors now reinforce each other. More data builds better models, better models demand faster chips, cheaper chips put intelligence on every asset, and every asset produces more data. This flywheel is why AI is starting to change the economics of energy and materials. This note explains what drives it, what it is worth, and where we think the value sits.
Four drivers that feed each other
Any one of these trends would matter on its own. What makes this moment different is that each one speeds up the next, so the loop keeps turning faster. It is the general enabler behind our two verticals: it is what makes the Dry Lab Revolution and Copper vs. Code possible now rather than in ten years.
More data
Sensors, satellites, simulations and self-driving labs produce more, and more granular, data about the physical world every year.
Better models
More data trains better models, and new algorithms, including multi-agent AI and quantum methods, get more out of every data point.
Faster, cheaper chips
Better models create demand for faster chips. Each euro buys more compute every year, in the data centre and at the edge.
Intelligence at the edge
Cheap compute at the edge makes it worth putting intelligence on every asset, which brings in more data and closes the loop.
Every turn of the loop is getting faster and cheaper
Four data points, one for each driver, show how far and how fast the loop has moved.
Growth in compute used to train frontier AI models
The compute behind the largest models has grown four- to five-fold a year since 2010. Models trained at that scale are now used to predict materials properties, weather and grid behaviour.
Cheaper to run a GPT-3.5-level model
The price of that level of performance fell from about $20 to $0.07 per million tokens between November 2022 and October 2024. Running AI on every asset or every experiment is becoming a rounding error.
For compute per dollar on AI chips to double
Hardware keeps getting cheaper per calculation, which is what makes it possible to move intelligence out of the data centre and onto chargers, batteries, turbines and lab robots.
Connected IoT devices at the end of 2025
Up 14% in a year and forecast to reach 39 billion by 2030. Around 13,000 active satellites add an orbital layer of data on top.5
Quantum is the long-dated accelerator, not the base case
Chemistry and materials are quantum systems, which is why classical computers struggle to simulate them exactly. In December 2024 Google showed for the first time that adding qubits to its Willow chip lowered the error rate instead of raising it, roughly halving errors at each step up in size.6 That is a milestone on the road to useful quantum simulation, not the destination.
We treat quantum as upside on top of the spiral. Quantum-inspired algorithms and hybrid workflows already help today; full fault-tolerant machines will widen the gap further when they arrive. We do not back ventures whose business case depends on that date.
The physical world used to be data-poor. The spiral is turning it into one of the richest data sources there is.
Models commoditise. Closed loops on physical assets do not.
As general-purpose models get cheaper and better, they become something every company can buy. The lasting value sits with those who own the data the models need and the loop that turns predictions into action.
General-purpose models and wrappers
Improving fast, priced lower every quarter, and available to every competitor. A thin layer on top of them is hard to defend.
Proprietary physical data
Sensor networks, lab results and asset histories that nobody else has. Each new deployment adds data a competitor cannot download.
Closed loops that act on the physical world
Systems that measure, predict and then steer a grid, a plant or a self-driving lab, and learn from every result. Value compounds with every turn of the loop.
The risks a careful investor should price in
Hype outruns the physics
Not every “AI for X” claim survives contact with real assets and real labs. We look for ventures with measured results in the field.
Quantum timelines slip
Useful fault-tolerant quantum computing may take longer than roadmaps suggest. We treat it as upside, never as the base case.
Data access is not a given
Asset owners and labs guard their data. Ventures need partnerships or their own sensors to feed the loop.
AI’s own footprint
Compute uses energy and water. Before we invest, we estimate a venture’s own AI footprint against its expected climate benefit, and only proceed when the benefit is many times larger. We then track both every year.
We don’t back the spiral itself. We back where it lands.
The spiral is the general enabler. We invest where it changes the economics of the physical world fastest: advanced materials and the energy transition, with initial tickets of €0.5–3 million from pre-seed to Series A.
- AI at the coreVentures with AI and proprietary data at the heart of the product, not bolted on.
- Physical, not only digitalClosed loops on real assets and real labs, where data is scarce and hard to copy.
- Quantum as upsideQuantum-ready where it helps, never dependent on a breakthrough arriving on schedule.
- Net positiveAn expected climate benefit many times larger than the footprint of the venture’s own AI, measured after we invest.
- Epoch AI, Training compute of frontier AI models grows by 4–5× per year, May 2024.
- Stanford HAI, The 2025 AI Index Report, April 2025.
- Epoch AI, AI and compute, updated February 2026.
- IoT Analytics, State of IoT 2025: number of connected IoT devices, October 2025.
- Look Up and Le Point, Space barometer: 13,026 active satellites as of 1 October 2025.
- Google, Meet Willow, our state-of-the-art quantum chip, December 2024 (results published in Nature).
Important information. This note is issued by AIXTerra Ventures for information and discussion with qualified investors only. It is thought leadership, not an offer or solicitation to buy or sell any security or fund interest, and not investment, legal or tax advice. Illustrative figures and models are simplified assumptions, not forecasts or indications of future returns. Early-stage investing carries a high risk of losing all capital invested and is illiquid. Past performance of the team is no guarantee of future results. Any investment is made solely on the basis of the relevant offering documents.