Deeptech bucked the overall downward funding trend in tech last year. European deeptech startups raised more than $20bn in 2023, surpassing 2022’s $18bn figure and making for a record 44% of total capital invested in Europe, according to UK VC Atomico.
AI companies like Paris-based Mistral and Germany’s Aleph Alpha raised monster rounds in the last six months, and have largely contributed to the deeptech craze. Many VCs are paying closer attention to pitch decks selling what may be the next scientific or engineering breakthrough.
But it’s not all about AI. Deeptech ranges from quantum technologies to space, through energy and robotics. So what ideas are likely to get founders a call with an investor? Sifted asked deeptech VCs for their take.
Emily Meads, deeptech associate at Speedinvest
Skilled, adaptable and intelligent robots
Robotic hardware has advanced by leaps and bounds over the past few years — systems that used to cost tens of thousands of dollars can now cost as little as a few thousand dollars. But these systems often perform best in a few select use cases in constrained environments, and today it remains a challenge to upgrade them for more complex and varied tasks.
AI and LLMs are posing as a fantastic way to better interact with, command and teach our robots about the world. But how can we best integrate all these moving pieces to help clients use skilled, adaptable and intelligent robots for their varied use cases? We are excited about anyone helping solve this challenge.
Sustainable semiconductor production
The explosion of generative AI and LLMs has made our reliance on semiconductors and chips clearer than ever. And it’s not just about making more chips, but also producing them more efficiently. Improving production would enable a better use of our fabrication facilities, lower energy costs and allow us to get more ‘bang for our buck’.
It’s still a challenge to change the way chips are produced due to fixed supply and manufacturing chains, as well as established fabrication techniques which often need fundamental engineering, or even physics to improve. I’d love to see ideas around this, as well as more companies supporting, more broadly, sustainable chip production — whether through green energy sources, tackling Scope 3 emissions or materials recycling and waste production.
Space infrastructure
The cost of sending systems into space has plummeted, widening access to space, especially low earth orbit. We’re getting good at gathering data, researching microgravity and opening access to space for both space and non-space companies to develop their applications. Now we need to make sure we have the underlying infrastructure to support an increasingly dense space ecosystem. I’d love to see ideas modernising our space communication stack, optimising usable surfaces and rides and improving our overall use of data for earth applications.
Clément Vanden Driessche, investment director at Elaia
Unleashing more compute power
As the need for more compute power for cutting-edge AI technologies increases, innovations in quantum and photonics that can improve performance while minimising energy consumption are becoming front-of-mind. Development and integration of such deep technologies are challenging, but major players such as Taiwan’s TSMC are already designing manufacturing lines to offer these technologies at scale. As enterprises continue to integrate AI into their workflows, infrastructure that can set new standards for compute will be top of mind for investors.
Estelle Godard, spacetech and deeptech investor at Promus Ventures
Hybrid AI
As they continue to grow in complexity, large language models (LLMs) will demand significant computational and energy resources. Hybrid AI could become an exciting evolution, combining the strengths of on-device processing and cloud computing. Put simply, hybrid AI is like having a smart helper that knows when to use the power of the internet and when to rely on the power of the device on which it is running. It splits the tasks between your device and the cloud to give you better experiences while saving energy, money and time.
Models with more than 1bn parameters are now operational on mobile devices with cloud-comparable performance, and devices are expected to handle models with 10bn parameters or more soon. In space operations, hybrid AI could show promising results to manage space traffic. With more than 2,000 active satellites and approximately 600k pieces of space debris, efficient and real-time processing capabilities are essential. Hybrid AI could enable on-device processing on satellites, enhancing decision-making for navigation and debris avoidance, while cloud analytics provide deep insights to support these operations.
Wireless power transmission for energy supply
I am particularly interested in technologies that address today’s energy challenges, such as wireless power beaming, which transmits electrical energy without wires and addresses the global need for constant and accessible energy supply. Investment into the tech was up 20% in 2023. From charging smartphones without the need for cables to powering isolated research stations in Antarctica, this technology has versatile applications. Part of its appeal is its ability to deliver energy seamlessly across various contexts, drastically reducing reliance on traditional infrastructure.
Augmented GNSS for high-precision location services
Global navigation satellite systems (GNSS) are networks of satellites that help devices determine their position, speed and accurate time. They are like a GPS, but include multiple satellite systems from different countries, such as GPS (USA), GLONASS (Russia), Galileo (EU) and others. Augmented GNSS enhances accuracy, especially for tasks like guiding autonomous vehicles or precision farming. The industry is growing fast, with revenues expected to nearly double from €260bn in 2023 to approximately €580bn by 2033. It’ll benefit from broader integration with IoT and smart devices.
Beyond its use for giving directions, this technology enhances safety and boosts operational efficiency in sectors like transportation and logistics. In space applications, it improves satellite coordination and navigation, essential for ambitious missions such as asteroid mining and Moon exploration.
Antoine Moyrood, partner at Lightspeed Venture Partners
Vision-language models for robots
Integrating vision-language models into robotics takes us closer to a future where robots benefit from enhanced perception and an ability to plan or make decisions in environments they haven't previously seen. These new models allow robots to interpret visual and textual data, enabling deployments in complex environments such as industrial settings.
By enabling robots to understand and react to their surroundings using advanced visual and language cues, we reduce the need for extensive customisation and manual programming. This results in faster implementation, lower operational costs and improved safety. Companies such as RobCo are already working to reduce the deployment time of these robots. In the foreseeable future, we anticipate robots widely deployed on factory floors, learning through imitation or from video feeds that demonstrate tasks to be automated.
Resolving AI’s memory bottlenecks
AI model sizes and compute consumption have grown at a faster rate than hardware capabilities, especially on the memory front. Model sizes […] rapidly create memory bottlenecks that increase latency. Some startups are working on hardware solutions (hardware accelerators, high-capacity RAM) while others are looking at software optimisation. Whoever succeeds in offering a material breakthrough stands the chance of offering the highest price/quality ratio, offering both cheap serving of models as well as high performance.
Zoe Mohl, investor at Balderton Capital
Cloud cost management and optimisation
Enterprises think 30% of their cloud spend is wasted and 80% want better visibility into their cloud spend, according to McKinsey. Once a company reaches a certain scale, cloud costs become not only very apparent but also out of control. It is hard to project cloud costs properly and to track what causes these costs to go up or down. We are excited by all companies building in the space that make cloud costs more understandable by breaking these costs down into categories.
The other area is of course optimisation. Once a company understands its cloud costs, what can they do to optimise those costs? Some new providers are technology specific and solely optimise one workload type, product or cloud provider; others offer application code change suggestions; and others are holistic in their approach and analyse all cost types and recommend changes.
