London-based AI hardware startup OLIX has raised $312 million (£231.8 million) in a Series B funding round, securing a valuation of $3.3 billion (£2.45 billion). The two-year-old company is working to scale its specialised silicon platform for AI inference, a segment that has become increasingly critical as demand for generative AI and large language models continues to grow.

OLIX is aiming to tackle the compute efficiency bottleneck that has become one of the most pressing challenges in the AI industry. As models grow larger and more complex, the cost and energy required to run them has skyrocketed. While much of the industry has focused on training ever-bigger models, the inference phase—where a trained model is actually used to generate answers, translate text, or create images—has emerged as a major constraint. OLIX believes that the answer lies not in general-purpose GPUs, which are designed to handle a wide variety of tasks, but in specialised silicon tailored to the specific stages of token generation.

The company's X-1 platform takes a distinctive approach. Instead of relying on a single monolithic processor, the platform unrolls models across a network of custom chips that are connected via a high-speed optical interconnect. This interconnect, which the company describes as “slow and wide,” uses light rather than copper to transfer data between hardware nodes. The design is intended to reduce the energy and latency penalties associated with traditional electrical signaling, allowing the chips to work together far more efficiently when running inference workloads.

OLIX's initial chip, the DX-1 decode accelerator, is built around a deliberately different memory strategy. Rather than relying on High-Bandwidth Memory (HBM) or advanced packaging technologies, the DX-1 uses fast on-chip SRAM. This choice allows OLIX to sidestep the supply chain shortages that have become a major bottleneck across the wider semiconductor industry. HBM has been in particularly tight supply due to surging demand from AI accelerators, and advanced packaging capacity has also been strained. By using SRAM, OLIX believes it can deliver a high-performance inference chip without competing for the same scarce manufacturing resources.

Funding details and investors

The Series B round attracted a mix of strategic and financial investors. Fundomo, Arm, and Hudson River Trading participated, alongside high-profile angel investors including Netflix co-founder Reed Hastings. Existing backers Hummingbird Ventures, Crane, Plural, Creandum, Phoenix Court, and Transition all increased their commitments in the two-year-old firm. The UK government's Sovereign AI venture fund has also backed the group, underscoring the strategic importance the government places on domestic AI hardware development.

The new capital injection will fund the path to deliver the DX-1 chip to launch customers by the second half of 2027. It will also support supply chain commitments and enable the company to expand its engineering teams across London, Bristol, Austin, Toronto, and San Francisco. OLIX has assembled a global team of engineers and researchers, and the funding will allow it to accelerate its development timeline while building out the infrastructure needed to bring a custom chip to market.

Government backing and UK AI strategy

The involvement of the UK government's Sovereign AI venture fund reflects a broader policy push to position Britain as a leader in the foundational technologies that power AI. In a statement, AI Minister Kanishka Narayan highlighted the importance of supporting companies like OLIX. “The future of AI will be built on chips that power models. Countries that build chips will build leverage,” he said. “OLIX is exactly the kind of ambitious company we want to back through Sovereign AI. In just two years, it has established itself as one of the UK's most exciting AI startups, developing breakthrough chip technology with the potential to help shape the future of AI.”

Narayan added: “If we want Britain to lead in AI, we need to back the technologies that sit underneath it. That's how we'll attract investment, create high-skilled jobs and ensure the UK remains a country that builds the future of AI, not just uses it.” The quote underscores a growing recognition among policymakers that AI's long-term economic impact will depend on owning the hardware layer, not just the applications.

The inference bottleneck and OLIX's approach

The challenge of AI inference is fundamentally different from training. During training, massive amounts of data flow through a model in parallel, and GPUs are well-suited for that workload. Inference, on the other hand, is often sequential in nature. When generating text, a model produces one token at a time, and each token depends on the previous one. This sequential dependency means that a large portion of the chip's compute capacity remains idle during inference, as it waits for the previous token to finish. OLIX's DX-1 is designed specifically for this decode phase, where the model generates output tokens one by one.

By tailoring the architecture to the demands of token generation, OLIX aims to achieve much higher utilization and energy efficiency than a general-purpose GPU. The use of SRAM, while more expensive per bit than DRAM, offers extremely low latency, which is critical when each token generation step depends on fast memory access. The optical interconnect further reduces latency by enabling efficient communication between multiple chips, allowing the model to be scaled horizontally across a cluster without being slowed down by copper-based data transfer.

The company's strategy of unrolling models across custom chips is reminiscent of approaches taken by other AI hardware startups, but OLIX differs in its focus on inference and its deliberate avoidance of HBM and advanced packaging. This could give it a significant advantage in a market where supply constraints are expected to persist for years. The semiconductor industry has seen a surge in demand for AI chips, driven by companies like Nvidia and AMD, but also by a growing number of cloud providers and enterprises building their own silicon. In this landscape, any startup that can offer a viable alternative to HBM dependency could capture meaningful market share.

Scaling plans and global expansion

OLIX has offices and engineering teams in London, Bristol, Austin, Toronto, and San Francisco. The new funding will allow the company to deepen its presence in all of these locations, hiring additional talent in chip design, software, and systems engineering. The company's leadership has emphasized that silicon development is a long-term endeavor, and the 2027 launch timeline reflects the complexity of bringing a new chip to production. However, the backing of strategic investors like Arm, which license intellectual property to nearly every major semiconductor company, gives OLIX access to critical expertise and ecosystem connections.

The decision to target launch customers in the second half of 2027 suggests that OLIX is not looking to rush an unproven product to market. Instead, the company appears to be focused on building a stable, reliable platform that can be deployed at scale. The delay also gives the company time to work on the software stack required to program and deploy its chips, which is often the most challenging part of making custom silicon viable.

Broader implications for the AI hardware market

The valuation of $3.3 billion is striking for a company that has yet to ship a commercial product. It reflects the intense investor appetite for companies that can challenge the dominant position of Nvidia in AI compute. Nvidia's GPUs have become the de facto standard for both training and inference, but their dominance has also created vulnerabilities. The high cost of GPUs, combined with supply constraints and the growing energy demands of AI workloads, has opened the door for alternative architectures.

OLIX is not alone in pursuing this opportunity. A wave of startups, including Cerebras, Groq, and d-Matrix, are all developing specialized AI hardware with different trade-offs. What sets OLIX apart is its focus on the decode phase and its use of optical interconnects, which are still relatively uncommon in chip design. The company's ability to attract investment from Arm and the UK government's Sovereign AI fund suggests that its approach is seen as credible and strategically important.

The UK has been working to establish itself as a hub for AI hardware innovation, and OLIX is one of the most prominent examples of this push. The government's investment in the company is part of a broader strategy to support sovereign AI capabilities, ensuring that the UK has its own supply of critical technologies rather than relying entirely on imports. With the Series B round now complete, OLIX will have the resources to move from research and development to productization.

The next few years will be critical for the company. Delivering a chip that performs well in real-world deployments is fraught with technical and logistical hurdles. However, the funding round gives OLIX the runway it needs to navigate those challenges and, if successful, to become a significant player in the global AI hardware market. As the demand for efficient inference continues to rise, companies like OLIX could play a key role in shaping how AI is deployed across industries.

Source: UKTN News