The Spanish startup scene just produced a funding round that tells a much bigger story than one company’s success. Multiverse Computing has secured a Series C worth 570 million dollars at a pre money valuation of about 1.7 billion dollars, a step change that crystallizes two of the strongest forces in artificial intelligence right now: the need to cut the cost of large models and the push in Europe for sovereign AI infrastructure. This is not just another big number. It is a signal that investors now see efficient, locally controlled AI as strategic infrastructure, not a niche optimization project.
From quantum specialist to efficient AI infrastructure player
Multiverse Computing was founded in Spain and initially became known for applying quantum and quantum inspired algorithms to complex optimization and simulation problems, especially in finance and industry. Over time the team pivoted from pure quantum positioning to a broader identity as a leader in efficient AI, using those algorithmic ideas to shrink and accelerate large models while preserving accuracy. As AI becomes critical infrastructure, dedicated governance frameworks have emerged to ensure responsible deployment.
The company’s profile jumped in 2025 with a Series B round of roughly 215 million dollars, about 189 million euros, led by Bullhound Capital alongside investors such as HP Tech Ventures, Forgepoint Capital International, and several European public and quasi public funds. That capital helped Multiverse turn early compression research into a commercial stack and validate the technical foundation that would later be branded as CompactifAI.
In parallel, the firm expanded its sector focus beyond financial services into manufacturing, energy, mobility, and public sector use cases, where running powerful models on constrained infrastructure is a recurring pain point.
By mid 2026, Multiverse had also begun releasing more visible AI assets, including an open source large model called HyperNova 60B that showcased its ability to work with frontier scale architectures rather than only small or toy examples. The new Series C round effectively confirms that the market has accepted Multiverse as a serious infrastructure player rather than a speculative quantum bet.
What this Series C round actually includes
The new financing totals around 570 million dollars or 500 million euros of Series C capital, valuing Multiverse at about 1.7 billion dollars pre money, approximately 1.5 billion euros. That valuation represents roughly a fivefold step up from the Series B, reflecting extremely rapid growth in commercial traction and perceived strategic importance.
Some coverage has described the round as the largest private investment ever raised by a Spanish technology company, and certainly among the largest late stage financings for a Spanish startup.
When this funding is added to earlier rounds, the company’s total capital raised is expected to reach around 800 million dollars. That level of backing leaves Multiverse with approximately 800 million dollars to build out its compression technology and sovereign AI infrastructure. Importantly, Multiverse and some investors have indicated that the round may remain open to select additional strategic backers, suggesting that the final total could end up near the top of the communicated range rather than below it.
The investor syndicate is intentionally broad. Forgepoint Capital International, BNPP Solar Impulse Venture Fund, and Bullhound Capital are co-leading the round, bringing together cybersecurity and infrastructure experience, climate-oriented capital, and growth-stage technology expertise.
They are joined by a long list of financial and strategic investors, including Santander Alternative Investments, Tikehau Capital, HP Inc, Orange Ventures, Scania Invest, NAventures from National Bank of Canada, Qatar Development Bank, and Zouk Capital.
Public and quasi public investors play a noticeable role, consistent with the sovereign AI positioning. The roster includes Spain’s Sociedad Española para la Transformación Tecnológica, the European Innovation Council Fund, and the Basque Government’s Hazten Scale Up Fund, along with local Basque institutions such as Gestión de Capital Riesgo de Euskadi and Kutxa Fundazioa.
JP Morgan and Santander Corporate and Investment Banking advise the transaction, which reinforces just how institutional this round has become compared with earlier venture stages.
CompactifAI and the rise of model compression as a strategic layer
The fresh capital is primarily earmarked for CompactifAI, Multiverse Computing’s platform designed to compress and optimize large AI models so they become cheaper to train, easier to deploy, and less resource-hungry in production.
In public materials, the company has claimed that its techniques can reduce model size on the order of eighty to ninety-five percent in some scenarios while maintaining similar performance, using a mix of advanced quantization, pruning, distillation, and proprietary optimization methods.
If those numbers generalize beyond hand-picked benchmarks, the implications are significant. Compression at that scale can multiply the effective capacity of existing GPU fleets, reduce inference latency, and shrink the energy footprint of large deployments, particularly in edge and on-premise environments where hardware is constrained.
It also directly attacks one of the core economic bottlenecks of generative AI, where serving costs for large language models and multimodal systems can erode margins even for well-funded companies.
The Series C funding will expand Multiverse’s library of production-ready efficient models, extend its research into new compression algorithms, and strengthen integration with customer infrastructure from edge devices to cloud clusters.
In practice, that means moving beyond reference demos to hardened offerings for sectors such as manufacturing, financial services, energy, aerospace, healthcare, cybersecurity, defense, and the public sector, where reliability, latency, and auditability matter as much as raw accuracy.
There is a broader industry context here. Over the last few years, the AI community has swung from a focus on scaling parameters and training compute toward a more nuanced view that prizes inference efficiency and total system cost of ownership.
Hardware vendors and open-source communities already push quantization and serving optimizations, but many enterprises lack the in-house expertise to compress and validate complex models safely. A specialized platform that treats efficiency as a first-class goal can fill that gap, provided it delivers transparent benchmarks and does not lock customers into opaque pipelines.
Sovereign AI and the European strategic agenda
Multiverse is explicitly framing this round as a way to build sovereign AI infrastructure that aligns with European and national priorities around strategic autonomy.
That language tracks a growing policy trend in Europe, where regulators and governments want to ensure that critical digital infrastructure, including foundational models, can be controlled, audited, and operated within European legal and energy constraints rather than depending solely on non-European hyperscalers.
The company plans to invest part of the capital into what it describes as AI gigafactory infrastructure and the associated software stack, with deployments spanning from edge devices to regional and national data centers.
That concept echoes battery gigafactories and cloud regions: large, standardized facilities that can host and serve efficient models with strong guarantees around data residency, uptime, and cost predictability.
Geographically, Multiverse intends to deepen its base in Spain and the wider European market while expanding to East and Southeast Asia, the Middle East, Canada, and the United States.
That mix of regions reflects both growth potential in emerging AI markets and the reality that sovereign infrastructure is often regional rather than global. A bank or energy company in the Middle East may want similar efficiency and control guarantees as a European public agency, but they will likely insist on deployments in their own jurisdiction.
Public investors in the round reinforce the link between this strategy and European policy goals. Institutions such as the EIC Fund and regional Basque vehicles have mandates that include technological competitiveness, industrial resilience, and local employment, which helps explain their interest in an AI company that both reduces energy consumption and anchors high-value capabilities in the region.
Traction, revenue growth, and what investors are betting on
Several reports indicate that Multiverse’s annualized revenue has grown more than tenfold since mid 2025, a pace that helps justify the fivefold valuation increase between the Series B and Series C.
While the company has not disclosed detailed financial statements, public commentary emphasizes recurring deals in industries with heavy regulatory and operational requirements, which are typically slower to adopt new technology but stickier once deployed.
Investors appear to be betting on three things. First, that the cost and energy pressures around generative AI will only intensify as models become larger and more pervasive, creating sustained demand for deep compression and optimization expertise.
Second, that many enterprises and governments will favor suppliers who can deliver these efficiencies in a way that aligns with local regulatory and sovereignty requirements.
Third, that Multiverse’s mix of quantum inspired algorithms, model compression know-how, and early positioning in Europe can translate into a defensible moat rather than being quickly commoditized.
There are risks embedded in that thesis. Model compression is an active research area with contributions from large cloud providers, chip makers, and academia, and techniques that look proprietary today can become widely known quite quickly.
Claims of extremely high compression ratios without accuracy loss need continuous third-party verification, especially as models are deployed in safety-critical domains like healthcare or industrial control.
Efficient AI infrastructure is also a competitive field, with rivals in both open-source communities and major cloud platforms pushing their own optimized runtimes and specialized hardware.
What this means for enterprises and the AI ecosystem
For enterprises already struggling with AI compute bills, a well-executed CompactifAI style platform could change the calculus of what is viable to deploy in production.
Instead of restricting advanced models to a handful of high-value use cases, companies could push them into more routine workflows in operations, risk management, customer support, and internal knowledge management, especially where on-premise deployment is a regulatory or security requirement.
This funding round also illustrates a broader shift in where value is perceived in the AI stack. A few years ago, most excitement centered on foundational model training and application layer startups.
Now there is increasing recognition that the middle of the stack, where infrastructure, optimization, security, and governance live, is just as critical. The fact that institutions with backgrounds in cybersecurity, climate finance, and industrial technology are co-leading the round underlines that perception.
For the open-source ecosystem, Multiverse’s work is a double-edged development. On one hand, better compression and serving techniques can make open models more competitive by allowing them to run efficiently on commodity hardware.
On the other hand, if critical compression know-how is locked into proprietary platforms, it could widen the gap between organizations that can afford commercial tooling and those that rely solely on community resources.
The company’s history of publishing research and releasing at least one open model suggests it understands the value of openness, but the balance between open and closed components will be an important signal to watch.
Regulators and policymakers may see this as a proof point that European AI can attract late-stage capital without abandoning commitments to safety and sovereignty.
However, the test will be less about funding headlines and more about whether real-world deployments deliver on promised energy savings, robustness, and transparency.
For example, if Multiverse can demonstrate that compressed models materially reduce data center energy use while meeting the technical requirements of the EU AI Act and sectoral regulators, that would strongly validate the thesis behind this round.
Key takeaways and what to watch next
Multiverse Computing’s 570 million dollar Series C at a 1.7 billion dollar valuation marks a pivotal moment for both Spanish tech and the global efficient AI ecosystem.
It confirms that model compression and optimization have moved from a niche research topic to a core pillar of AI infrastructure strategy.
It also shows that sovereign, regionally aligned AI platforms can attract serious institutional capital when they tie efficiency to policy goals.
The most important questions from here are practical. Can Multiverse translate its research claims into consistent, independently verifiable efficiency gains across many model families and use cases, not just carefully selected benchmarks?
Can it build out AI gigafactory infrastructure that is flexible enough for very different regulatory regimes in Europe, Asia, the Middle East, and North America?
And can it sustain its reported revenue growth in the face of intensifying competition from cloud providers, semiconductor firms, and other efficiency-focused startups?
If the company executes, this round could be remembered as a turning point in how the industry thinks about scaling AI: not just bigger, but leaner, cheaper, and more sovereign.
If it falls short, the underlying trend will continue regardless, because the economic and political forces pushing toward efficient, locally controlled AI are only getting stronger.
For now, Multiverse Computing has both the capital and the mandate to try to define that future.
Conclusion
The latest funding round for Multiverse Computing is a clear signal that efficient artificial intelligence is moving from supporting role to core infrastructure. A seven hundred seventy million dollar valuation jump and a huge new capital pool tell a simple story. The future is not just about bigger models. It is about making them smaller, cheaper and responsible enough to run everywhere.
Why this Series C round matters now
Multiverse Computing has secured a five hundred seventy million dollar Series C round at a pre money valuation of one point seven billion dollars, a fivefold increase over its previous Series B valuation. This is one of the largest private investment rounds ever raised by a Spanish technology company and lifts total funding to around eight hundred million dollars when earlier rounds are included.
The investor syndicate is unusually broad. Forgepoint Capital International, BNPP Solar Impulse Venture Fund and Bullhound Capital are leading the round, alongside strategic commitments from HP, Orange Ventures, Scania Invest, Santander Alternative Investments, Qatar Development Bank and several European public funds, as well as regional funds tied to the Basque government and European Innovation Council. That mix of financial and strategic capital suggests investors view efficiency in AI as a long term infrastructure play rather than a short term feature bet.
The raise comes after a period of rapid growth. Since its June twenty twenty five Series B, Multiverse reports that annualized revenue has increased more than ten times, with first quarter twenty twenty six sales growing ninety six times year over year and its compressed models already running on millions of devices. Those numbers are aggressive and will need continued performance to be fully validated, but they explain why investors are willing to pay a steep premium.
In the wider market, the timing aligns with growing pressure on model providers and enterprises to reduce energy consumption, manage hardware costs and support sovereign AI strategies. Governments and large companies are no longer comfortable with critical workloads depending entirely on a few hyperscale clouds. That is why investors are backing companies that can make powerful models practical on local servers, sovereign data centers and consumer devices.
From the race to scale to the race to shrink
The first wave of modern generative AI was dominated by a simple metric. More parameters and more compute generally meant better performance. That led to extremely large foundation models that demanded vast amounts of specialized hardware and energy.
Over the last two years, the narrative has begun to change. Providers started aggressively quantizing weights, pruning redundant components and distilling knowledge into smaller student models to make deployment practical. Google has introduced techniques such as TurboQuant to cut memory footprints and accelerate inference of large models. Academic groups and industrial labs, including teams at MIT and Max Planck, have proposed training time compression methods like CompreSSM that reduce state dimensions significantly while maintaining accuracy and improving training speed.
Multiverse Computing sits squarely in this second wave. Instead of competing on sheer size, it focuses on compression and efficiency, originally with quantum inspired techniques and now through a broader portfolio of algorithms. This shift from maximum scale to optimal efficiency echoes earlier transitions in computing, where raw clock speeds eventually gave way to multicore architectures and energy aware design.
What the Series C capital is funding
Multiverse has been explicit about how it intends to use the new money. The round is meant to expand what the company describes as a library of the most efficient models available, covering a spectrum of tasks from language to multimodal workloads. The emphasis is on pre compressed models that enterprises and governments can deploy without needing deep in house expertise in low level optimization.
A significant portion of the capital is earmarked for continued research and development into proprietary algorithms for shrinking models while preserving or even improving performance. Given the pace of innovation in compression techniques, sustained investment in original algorithmic work is necessary if Multiverse wants to remain ahead of generic quantization or pruning approaches that are increasingly commoditized.
Another key use of funds is investment in sovereign AI gigafactory infrastructure and the accompanying software stack. The term gigafactory in this context refers to large facilities dedicated to training, fine tuning and deploying models under the control of specific countries or regions, rather than relying entirely on global cloud providers. Multiverse aims to supply both the efficient models and the orchestration software that sit on top of such infrastructure.
The company also plans to build out its presence in East Asia, Southeast Asia, the Middle East, Canada and the United States, reflecting demand from governments and regulated industries in those regions for sovereign and efficient AI solutions. That geographical spread positions Multiverse as a potential bridge between European regulatory priorities and global deployment needs.
Inside the compression technology and how it fits a broader trend
The technical anchor for this raise is Multiverse’s CompactifAI technology. The company claims it can shrink large language models by eighty to ninety five percent while preserving accuracy closely enough for real world deployment. The compressed models reportedly run faster, consume less energy and can be served on consumer devices, on premises servers and sovereign data centers rather than only in hyperscale environments.
Although Multiverse does not publicly disclose all implementation details, the approach appears to combine several strands of modern compression. These likely include lower precision representations of weights, structured pruning of attention heads or layers and forms of distillation that transfer outputs from a large teacher model to a smaller student. The company also draws on earlier work in quantum inspired optimization, which can help identify and remove redundant information in high dimensional spaces more efficiently than naive methods.
The broader ecosystem reinforces the significance of Multiverse’s claims. Academic work like CompreSSM has demonstrated that compressing models during training rather than as a separate post process can yield models that are both smaller and faster while retaining high accuracy on benchmarks such as CIFAR ten and architectures like Mamba. Consumer oriented efforts, including startups reportedly in talks with a major smartphone manufacturer to run advanced models directly on phones, show there is real commercial appetite for compression that makes powerful AI practical on everyday devices.
In that context, CompactifAI is not an isolated breakthrough but part of a larger movement to treat efficiency as a first class design goal rather than a late stage optimization.
Strategic implications for businesses and sovereign AI
For enterprises, the practical implications are direct. If a company can deploy a model that has been reduced by up to ninety five percent without a meaningful loss in performance, it can cut spending on specialized hardware, reduce dependence on a single cloud provider and bring more workloads closer to where data is generated. That matters especially in finance, health care and industrial operations where latency, privacy and regulatory compliance are critical.
On the sovereign AI front, Multiverse’s positioning is particularly notable. The company highlights sovereign and efficient AI as its focus, aligning with European and global concerns about data control, strategic autonomy and resilience. Compressed models that run effectively in national or regional gigafactories allow governments to train and serve systems using local data and infrastructure, without sending everything to a handful of foreign hyperscalers.
This funding round also signals a shift in European technology ambition. Historically, Europe has lagged US and Chinese firms in headline grabbing AI funding and scale. A five hundred seventy million dollar round for a European compression specialist, backed by both private capital and public funds tied to national and regional initiatives, suggests that efficiency and sovereignty may be the angles through which Europe seeks to carve out a durable role.
For cloud providers and chip designers, the rise of companies like Multiverse creates both opportunity and pressure. Efficient models can expand the overall market for AI by enabling more use cases on less expensive hardware, which benefits vendors who can sell into a wider customer base. At the same time, compression reduces demand for the most resource intensive configurations, which may force some players to recalibrate assumptions about perpetual growth in high end compute consumption.
Risks, trade offs and unanswered questions
The enthusiasm around this round should not obscure the risks and uncertainties. Compression always involves trade offs. While Multiverse and others report limited accuracy loss, the details matter. Some tasks, especially those involving long context reasoning or edge case safety behavior, may be more sensitive to the removed parameters than headline benchmarks suggest. Enterprises will need independent evaluations, robust monitoring and clear documentation before relying on heavily compressed models in high stakes workflows.
There is also the question of durability of competitive advantage. Techniques like quantization and pruning are increasingly available in open source tooling and major frameworks. If CompactifAI’s core methods are too easy to replicate, Multiverse could face pressure on pricing and margins as similar capabilities spread. Continued investment in proprietary algorithms and close integration with sovereign infrastructure projects will be crucial to defend its position.
Another risk lies in complexity. Deploying compressed models at scale across millions of devices and heterogeneous data centers requires sophisticated orchestration, update management and security discipline. Efficient code paths can introduce new failure modes and make debugging more difficult. If efficiencies are achieved at the cost of greater system fragility, enterprises may hesitate despite the theoretical savings.
Finally, the geopolitical dimension of sovereign gigafactories should not be underestimated. As countries invest in their own AI infrastructure, questions about interoperability, standard setting and cross border governance will intensify. Companies like Multiverse that sit at the intersection of technical innovation and national strategies will need to navigate policy landscapes carefully to maintain trust across regions.
Key takeaways and what to watch next
Multiverse Computing’s Series C round confirms that investors now see AI efficiency as a foundational layer of the technology stack, not just a niche speciality. By raising five hundred seventy million dollars at a one point seven billion dollar valuation, the company has secured both validation and a substantial runway to expand its library of compressed models, deepen algorithmic research and help build sovereign AI infrastructure across multiple continents.
The central bet is that shrinking models by eighty to ninety five percent while preserving useful performance will unlock AI deployments on devices, in enterprise data centers and in national gigafactories that would otherwise be uneconomical or politically uncomfortable. If that bet pays off, the next phase of AI adoption may be defined less by spectacular frontier models and more by quietly efficient systems that are small enough and trusted enough to become ubiquitous.
Over the next few years, several indicators will reveal whether this vision is becoming reality. The depth and transparency of independent benchmarks for compressed models, the pace of enterprise migration from cloud only deployments to mixed or on premises architectures, and the concrete progress of sovereign AI initiatives will all matter. The way chip makers and cloud platforms respond, whether by embracing compression or resisting it, will shape the economics of the entire sector.
If Multiverse delivers on its promises, this round may be remembered as one of the moments when the industry shifted from sheer scale to responsible efficiency as the default expectation for advanced AI. If it stumbles, the concerns around replicability, trade offs and operational complexity outlined above will have been underestimated. Either way, efficient AI is now central to the strategic conversation, and that alone marks a step change in how the field is evolving. reddit
