Origen: Inside Multiverse Computing, the Spanish startup vying to be Europe’s next AI champion | Sifted

 

Inside Multiverse Computing, the Spanish startup vying to be Europe’s next AI champion

HQ’d in Donostia in northern Spain, Multiverse Computing has grown into one of Spain’s most high-profile tech companies. How did it get here?

Last month, Spanish scaleup Multiverse Computing, which has built a technology to reduce the compute requirements of large AI models, announced it is nearing the close of a €500m Series C.

When completed, the round will be one of the largest fundraises ever secured by a Spanish tech startup, bringing total funding to more than €700m.

With the Series C valuing Multiverse at €2bn post-money, it also means the scaleup is joining a cohort of another dozen Spanish unicorns — and is the country’s only billion-dollar tech business operating in the field of frontier AI.

Unlike AI labs like French darling Mistral and US giants Anthropic and OpenAI, Multiverse does not develop its own models; the company has built a technology inspired by quantum physics to “compress” existing large-language models (LLMs). This, it says, can reduce model size by up to 95% while causing little accuracy loss.

Compressed models require fewer compute resources and infrastructure, meaning significant cost savings for users. The thesis has attracted interest from high-profile customers ranging from the Bank of Canada and the European Space Agency to Bosch and Mercedes-Benz.

With more than 100 customers globally, this year Multiverse expects to achieve €200m in annual recurring revenues (ARR), a significant jump from the €13m it recorded in 2024 (when the company started commercialising its model compression technology).

A lot of hope is riding on the scaleup, which from the city of Donostia on the northern coast of Spain has grown into one of the country’s leading tech businesses, counting 500 employees across Europe and North America.

And despite some of the constraints that come with operating from a smaller tech hub, Multiverse has no plans to slow down. Equipped with a unique technology and a highly commercially-driven mindset, the company is determined to become a Europe-born AI champion playing in the big leagues.

From quantum to AI

Multiverse was not always focused on AI.

Founded in 2019 by Enrique Lizaso, who previously worked in banking and finance, together with quantum physicists Román Orús and Samuel Mugel, and Alfonso Rubio, who had spent the previous decade working in the tech sector, the company originally focused on quantum computing.

At the time, Multiverse offered pure quantum algorithms intended to run on quantum computers, as well as “quantum-inspired” software, which refers to algorithms leveraging mathematical concepts from quantum physics but run on classical hardware.

In 2022 Multiverse was approached by Bosch with a request to help improve the efficiency of an AI system deployed in the German company’s factories. The existing system was too large to be deployed on-premise in the factories; it had to run through the cloud, creating latencies.

Multiverse developed quantum-inspired algorithms based on mathematical tools called “tensor networks” to shrink the size of the system without losing accuracy.

With OpenAI’s first public release of ChatGPT in late 2022, LLMs were soon the talk of the town. This, Lizaso tells Sifted, is when the company decided to start using tensor networks to compress large foundational AI models, focusing on open-source systems, which are available to freely download and modify.

Multiverse published a research paper in 2024 showing the success of the method, and pitching the technology’s potential to cut the energy consumption of AI systems.

“We realised that the topic of energy consumption would be particularly relevant, because energy is related to computation, and computation is related to the price you will be charged,” says Lizaso. “This is when we thought: ‘Oh, that’s it. This is going to be a hit.’”

Olivier Tonneau, partner at French-Canadian quantum VC firm Quantonation, which in 2019 invested pre-seed capital in the company, says this was a pivotal time for the company. “The research paper went around and they started receiving inbound calls, because this was a huge topic for everyone,” he says.

A few months later, in June 2025, Multiverse secured a $215m Series B led by growth investor Bullhound Capital.

“Everyone we spoke to said these guys were at least two-to-three years ahead of everyone else,” says Callum Stewart, investment principal at Bullhound Capital.

While the VC firm was originally interested in Multiverse’s  quantum computing pitch, Stewart says Bullhound Capital was willing to “underwrite what was basically a brand new product” when it became obvious that the model compression technology would be “transformational”.

Lizaso says the Spanish scaleup is still carrying out research and development focused on quantum computing, with more than 300 patents registered in the field. It plans to leverage this expertise once quantum computers are mature enough to enable practical use cases.

“We know that some AI applications are going to be accelerated by quantum computing,” says Tonneau. “So Multiverse will be ideally positioned to implement quantum within its own AI projects.”

“The company’s valuation includes what they are doing in AI but also the possibility of developing quantum software when the hardware is ready.”

AI at the edge

Just over a year after the Series B, Multiverse is raising an additional €500m in Series C funding, which Tonneau says is near closing.

The round, which is co-led by Bullhound Capital, Forgepoint Capital International and BNP Paribas’s climate-focused Solar Impulse Venture Fund (SIVF), has attracted commitments from strategic investors including US electronics giant HP Inc.

“Because they are from a smaller hub they may have been overlooked by other headline VC names, but strategic investors are piling in and saying this is really useful,” says Stewart.

Multiverse is betting on enabling AI on the “edge” — when computations happen on local devices, such as laptops and phones, rather than through the cloud in centralised data centres.

Edge computing, which reduces latency and energy consumption while keeping data safely on-premises, is becoming critical for a wide range of use cases including autonomous vehicles, smart factories and manufacturing, but also banking, healthcare and defence.

“It was the thesis behind HP’s investment,” says Tonneau. “If they want to put AI models in their PCs, they need systems that can work with no wifi, without killing the battery, without taking too much compute bandwidth.”

Lizaso says Multiverse is already collaborating with HP to deploy the technology, as well as with South Korean electronics giant Samsung. “You can think of a future where we use LLMs nearly everywhere — mobile phones, laptops, TVs, your Amazon Alexa,” says Lizaso.

The company is also partnering with chip manufacturers, including Dutch startup Axelera, which builds low-cost chips optimised for inference on edge devices. Through the collaboration, customers can run Multiverse’s compressed LLMs on Axelera’s chips.

Axelera cofounder Fabrizio Del Maffeo says the partnership will enable cost savings of up to 10x compared to using traditional LLMs running on hardware like chip titan Nvidia’s products.

“It enables a double optimisation of both hardware and software for our customers,” says Del Maffeo. “This is critical to reduce costs when you’re serving thousands of customers.”

Although the collaboration is still in early stages, Del Maffeo says it is being tested with a handful of clients, with plenty of interest from players in telecoms and manufacturing.

Unique IP

Compressing AI models to improve efficiencies is a thesis that is gaining momentum.

US startup PrismML recently announced a compressed version of Chinese open-source model Qwen, which it said retains up to 95% of accuracy. The company is reportedly in talks to be acquired by Apple.

Several other competitors operate in the space, with different approaches to compressing and optimising algorithms. They include Pruna AI in Paris and Latent AI in the US, as well as Ora Computing, based in Vienna, which similarly to Multiverse employs quantum-inspired methods.

But Del Maffeo says no company matches the scale that Multiverse has achieved yet. “We spoke to other companies with similar solutions, but I think Multiverse is leading the market, there is no company matching their size,” he says. “They can really become a global leader.”

Stewart says the Spanish company’s technology, which is based on highly technical quantum computing intellectual property (IP) developed by Orús at the Donostia International Physics Centre (DIPC), is a key differentiator for the business.

“It’s one of the really interesting things about Europe: there is this unique specialist research group within the university, with very specific technical knowledge, and the ecosystem around it, which if given access to capital can perform great things,” says Stewart. “This is very hard to replicate with capital alone.”

Multiverse is now building a suite of products to help customers optimise their AI workloads. The company offers a platform matching models with tasks, and bringing together model compression, GPU management, AI services and control systems.

In addition to large corporates, the scaleup is partnering with data centre providers to improve efficiencies directly on premises. The thesis, says Del Maffeo, is particularly relevant in Europe.

“If you have compute infrastructure like in the US, using large LLMs is not an issue, but Europe’s problem is we don’t have this infrastructure,” he says. “Multiverse is a necessary answer to democratise AI rapidly in countries where there is not enough capital to build large-scale infrastructure.”

Multiverse is part of public-private consortiums in France and Spain which are competing for the EU’s flagship gigafactory scheme, an initiative backed by €10bn in public funding aimed at building a new generation of data centres across the bloc.

If successful, Multiverse will deploy its software in the gigafactories, as well as invest in and become a joint shareholder for the projects.

Looking for deeper pockets

Some say the Multiverse’s rapid growth is uncommon.

“Even when they were a quantum company, they were very focused on commercial development,” says Tonneau. “They were very quickly working with dozens of customers to test use cases.”

“In quantum computing there were no other businesses with that same desire to go after commercial growth.”

For Stewart, this is what enabled Multiverse to become a brand name, even for a company born outside of traditional VC circles.

“The Series C hopefully helps catalyse that: the unicorn status, the big presence, the bold ambitions,” he says. “They identified an opportunity and figured this amount of capital will help them go after it. They are almost speaking the language of US tech.”

 “It will help them to not only be part of the European story but also get in with global names.”

Lizaso is keen to expand the scaleup’s footprint on the other side of the Atlantic. While Multiverse is firmly a Spanish company, he says, it also requires access to deep pockets of capital.

“We lack the financial fabric that the US has,” says Lizaso. “VCs in the US are in a different kind of game.”

He points to the EU’s Scaleup Europe Fund, a €5bn pot of money launched at the end of last year to back late-stage deeptech companies with tickets ranging from €100m-500m.

“We think of it as a tremendously large fund […] but this is peanuts in some fields,” says Lizaso, pointing to Anthropic’s latest round of funding, a $65bn Series H closed in May.

And while Multiverse’s Series C includes a mix of US and European investors, Lizaso is already thinking about the next steps — a potential IPO, which he says the company could consider in the coming years.

“It would be in the US,” says Lizaso. “If you’re doing an IPO in Europe only, you’re letting go of 80% of the value of the company.”