The AI boom has led to the creation of some of the fastest-growing companies ever in European tech, given the continent a genuine challenger to US giants such as OpenAI and sent VCs into a tailspin.
But it’s also created an insatiable demand for computing power, which is going nowhere but up as the tech is used more and more.
That’s given a leg up to startups with a focus on the future of compute. A cohort of startups has spied an opportunity to tap into the growing market for new modes of efficient computing, chip design and computing security.
To find out which companies could make it big, Sifted asked VCs for the future of compute startups they’re watching. With one catch: they couldn’t nominate portfolio companies.
Future of compute startups to watch
Amelia Armour, partner at Amadeus Capital Partners
Optalysys — UK
Optalysys develops optical AI systems that enhance computing efficiency and security. Its photonic processing technology enables Fully Homomorphic Encryption (FHE), allowing encrypted data to be processed without decryption, ensuring quantum-secure confidentiality.
The technology’s potential to push the boundaries of secure cloud computing, financial fraud detection and medical data processing also makes it worth following.
Blueshift Memory — UK
Blueshift Memory develops high-performance memory architectures designed to eliminate bottlenecks in data-intensive applications like AI, big data analytics and high-speed computing. This allows for 1,000x faster memory access and more efficient handling of large databases. Its memory module is independent of traditional memory cell technology, offering faster data access and reduced development time for software engineers.
The company says this will make it possible to enhance processing efficiency, reduce latency and optimise computing performance.
Heronic — UK
Heronic Technologies specialises in AI accelerator architectures, or more specifically the optimisation of chip efficiency for machine learning and robotics. Its hardware enhances object detection and human action recognition, outperforming many traditional architectures.
Max Ochs, investor at First Momentum Ventures
Belfort — Belgium
Belfort is developing hardware acceleration for fully homomorphic encryption (FHE). FHE is considered to be the holy grail of confidential computing as it allows data processing without ever decrypting the data. The field has seen quite some hype in the past, but the adoption is very slow because it suffers from extensive computational overhead and thus very slow data processing (even more demanding than typical AI calculations). Belfort is designing dedicated chips to speed up FHE calculations by multiple orders of magnitude (similar to AI accelerators). And it come from one of the best cryptography labs globally.
Chipletti — UK
Chipletti is a young UK startup working on 3D stacked SRAM chiplets. Modern AI inference chips require a lot of high speed memory on the chip. SRAM is the state of the art as it is mature and offers very low latencies. Unfortunately its density does not scale well, which is why these AI chips have very little memory limiting their performance. By stacking multiple SRAM chiplets on top of each other, you can increase memory without sacrificing speed (it’s all about keeping distances between SRAM chips small). So instead of reinventing memory, Chipletti is building on mature technologies.
Ion Hauer — principal at Apex Ventures
Skycore Semiconductors — Denmark
Skycore Semiconductors is a fabless company developing advanced power conversion microchips. Its technology improves the efficiency and performance of power modules, AI servers and battery energy storage systems (BESS).
This work is important because global electrification and AI advancements demand more efficient power solutions. Skycore addresses these challenges by creating integrated circuits that reduce energy waste and enhance performance, supporting the transition to sustainable energy and meeting the power needs of modern technology like AI compute workloads.
Luxtelligence — Switzerland
Luxtelligence pioneers next-generation connectivity for the quantum and AI era, specialising in thin-film lithium materials for optoelectronics. It offers services from prototyping photonic integrated circuits (PICs) to volume manufacturing.
As AI and quantum computing advance, the need for faster, more efficient optical components grows. Luxtelligence’s work in advanced photonics helps provide the foundational technology for these demanding applications, enabling progress in high-speed communication and computation.
Hauer also recommended Belfort.
David Grimm, partner at AlbionVC
Fractile — UK
Faster, cheaper inference is a huge unlock for modern AI, and Fractile is taking a radical approach. It’s building analog compute-in-memory chips purpose-built for transformers that eliminate the memory bottlenecks throttling GPUs. By running computations directly where weights are stored, it’s projecting three orders of magnitude increase in speed for large language model (LLM) inference.
Walter Goodwin, Fractile’s founder, is scarily bright and very driven. He’s building a world-class team and I’m excited to see them reshape AI infrastructure from the silicon up.
Nu Quantum — UK
Placing a bet on any single type of quantum hardware is difficult, as we’re still far from knowing which architecture will ultimately win. That’s why a company like Nu Quantum is so compelling.
It’s building quantum networking technology that enables distributed quantum computing, designed to work across all hardware types. This positions it to be at the centre of future quantum stacks, bridging systems and unlocking scale regardless of which platforms prevail.
Carmen Palacios-Berraquero, the company’s founder, blends academic depth with entrepreneurial drive, a rare and powerful combination that’s key to navigating such a frontier field.
Moray Wright, CEO at Parkwalk Advisors
Intrinsic — UK
Intrinsic’s primary market is the replacement of Flash memory in microcontrollers (MCUs), a type of processor on a single integrated circuit containing memory, processor and input /output peripherals.
Intrinsic has developed a disruptive non-volatile memory technology, Resistive RAM (ReRam), which promises CMOS compatibility, 100x faster read and 1,000x faster write times, lower energy input (by a factor of up to 100x/bit) and better endurance (by a factor of 1m) when compared to Flash.
Its tech will allow data hungry applications to overcome the memory bottleneck caused by current external flash memory, delivering higher performance with lower energy consumption.
Lumai — UK
Lumai is developing a next-generation optical computing architecture purpose-built for AI, offering orders-of-magnitude improvements in speed and energy efficiency. It has the potential to reshape how AI workloads are deployed at scale — offering 50x increase in AI performance, 10% of the cost and a 90% reduction in data centre power consumption.
Wright also recommended Fractile.
#future #compute #startups #watch #VCs