Quantum software startup Multiverse Computing has raised €25m as it turns its attention to the AI sector.
The company — which has developed a useful quantum software tool that’s been used by major banks — will use the investment to develop and launch new applications for large language models (LLMs), as well as to expand internationally and double down on its existing products.
Multiverse’s CEO and cofounder Enrique Lizaso tells Sifted that the round, which he says is the biggest raised by a quantum software startup in the EU, triples the company’s valuation to around €100m. The investment has been led by Spanish VC Columbus Venture Partners along with Quantonation Ventures, European Innovation Council and Redstone QAI Quantum Fund.
Quantum for AI
Lizaso says the huge volumes of cash pumped into AI since last autumn are having a spillover effect into the interest in quantum software. That’s because “quantum-inspired” computing techniques — which the startup has been working on since its founding in 2019 — could potentially address some pressing problems the industry faces today with AI models.
One of those is copyright: multiple AI companies are facing lawsuits over the use of copyrighted material to train their LLM models.
Multiverse Computing plans to launch a new product aimed at erasing knowledge contained in LLMs, without having to retrain them completely. It is — slightly terrifyingly — called “Lobotomizer” and Lizaso says it uses technology called quantum tensor networks to allow AI companies remove connections in their models to copyrighted material that might have been accidentally included in the training data.
“This will be the Holy Grail if someone sues you for using content with copyright,” Lizaso says.
Whether or not such a technique will satisfy lawmakers and judges who are clamping down on copyright in AI will need to be tested, but it’s not hard to see why such a product could be very attractive to AI companies that are trying to argue they’re not breaching copyright.
Power hungry
A second problem in AI that Mulitverse is focusing on is the immense power needed to run LLMs.
The costs of training these algorithms is rising exponentially as developers feed them with increasing amounts of data to make them more powerful.
Europe is home to 16% of over 8,000 data centres spread across the world. Ireland’s 82 data centres alone consumed 17% of the country’s electricity in 2022, and the International Energy Agency estimates this percentage will spiral to 32% by 2026.
Lizaso says a share of the new funding will go towards refining CompactifAI, a product launched in November 2023 that attempts to substantially cut the memory and storage space that an LLM requires.
“CompactifAI can compress a model by 85% or more without losing accuracy,” he says. “This halves its costs, which are mostly linked to its energy requirements, both when training and when using it.”
Multiverse says it’s already shown that the product can make Meta’s LLaMA models more efficient with Lizaso adding it can allow an LLM to run on a small device such as a smartwatch, rather than needing cloud access.
Spin-off discussion
If CompactifAI becomes a success, “it can change the company completely,” says Lizaso. The former university professor continued that he and his cofounders have already discussed the possibility of creating a spin-off around it. That’s because the potential customer base is very distinct from its original product Singularity and appeals to a different set of investors.
“This is a very strategic decision,” Lizaso adds. “The spin-off could become even larger than the original company.”
As Multiverse weighs up how to position its new AI-focussed tools, it’s also expanding its existing product into new sectors.
Singularity, which has been used by customers including BBVA, Bankia the European Tax Agency and the Bank of Canada, is also being used in the energy industry where the quantum software can help optimise electric grid supply and demand, says Lizaso.
With the funding from this round, he hopes to increase Singularity’s customer base in energy and other industries including defence, cybersecurity and manufacturing. Multiverse also wants to break into life sciences, where the product can be used to sequence proteins, and spacetech, where it’s adapting its algorithms to run onboard satellites.
How to grow
Lizaso says the company’s performance in the second half of last year “was tremendous” with revenue doubling compared to 2022. As of the end of that year, it was one of Spain’s fastest-growing startups outside the country’s main hubs of Madrid and Barcelona.
Part of the funding from this Series A round will fuel the startup’s expansion to the US, where Lizaso is already assessing sites in San Francisco, Texas and Boston. The company has offices in Munich, Paris, London and Toronto — from where it is already supplying the US market — in addition to its San Sebastián headquarters.
He plans to hire 50 employees this year to add to his 150-strong team and expects the company’s number of patents to grow from 93 to 150 by the end of 2024.
