OLIX

In the AI hardware race, the conventional wisdom has been to build a bigger, faster chip. OLIX, a London-based startup, is betting that the future lies not in a single chip, but in a factory—a production line of specialized machines connected by light. The company's thesis, laid out in its manifestos and recent Series B announcement, is that the current approach to AI inference is fundamentally compromised, and that the datacenter must be rebuilt as a 'token factory' to make tokens cheap and abundant.

The Token Bottleneck: Why Inference Is the Next Frontier

OLIX's argument starts with a simple observation: producing a token is not one operation but many, each with different hardware demands. Yet every AI system in production runs all these operations on the same general-purpose chip. This compromise, OLIX argues, is why tokens remain scarce and expensive. The company's language is pointed: 'The constraint sits in how tokens are made.'

This is a shift from the training-centric focus of most AI hardware startups. OLIX is explicitly focused on inference—the process of generating answers, not learning from data. The company's tagline, 'Building the infrastructure for frontier AI inference,' signals that it sees inference as the next bottleneck, as models like GPT-5 and beyond spend vastly more tokens to solve complex problems. The demand for tokens is 'compounding faster than the infrastructure beneath it can scale,' according to OLIX's website.

The 'Slow and Wide' Bet: Optical Interconnects Over Brute-Force Chips

OLIX's core technical bet is on a 'slow and wide' optical interconnect. Instead of pushing a single chip to ever-higher clock speeds, OLIX plans to use many specialized chips working in parallel, connected by optical links that are slower per lane but wider in aggregate bandwidth. This is a deliberate counter to the industry's obsession with shrinking transistors and boosting clock rates.

The company argues that moving work between chips has always cost too much energy and time for anything else to be viable. By using optical interconnects, OLIX aims to make inter-chip communication cheap and fast, enabling a more granular division of labor. This is not just a chip design; it's a network design. The 'slow and wide' approach is reminiscent of the shift from wide and slow to narrow and fast in some computing architectures, but applied to the interconnect layer.

Rack-Scale Codesign: Rebuilding the Datacenter as a Factory

OLIX's vision extends beyond the chip to the rack and the datacenter. The company emphasizes 'rack-scale codesign,' meaning that the chip, the optical interconnect, and the surrounding systems are designed together as a unified whole. This is a departure from the modular approach where chips are designed in isolation and then plugged into standard servers.

The factory analogy is central to OLIX's branding. The company describes its mission as building 'the token factory for frontier AI.' Just as a car factory has specialized machines for stamping, welding, and painting, OLIX envisions a datacenter with specialized chips for different parts of the inference process—some for attention, some for feed-forward layers, some for memory access. The optical interconnect is the conveyor belt that moves tokens between these stations.

This is a bold vision, but it's also a complex one. It requires not just chip design but also system integration, software, and networking expertise. The company's website mentions that it is growing teams across the US, Canada, and the UK, suggesting a broad talent acquisition strategy.

The DX-1 Chip: A 2027 Promise in a Fast-Moving Market

OLIX's first product, the DX-1 chip, is scheduled for delivery in H2 2027. That's a long lead time in a market where Nvidia releases new architectures every year or two. The company's Series B announcement does not provide detailed specs for the DX-1, but the focus on inference and optical interconnects suggests it will be a specialized chip, not a general-purpose GPU.

The 2027 timeline is a risk. By the time DX-1 ships, the market may have moved on, and competitors may have adopted similar approaches. However, the long timeline also reflects the complexity of the task. Building a new chip architecture from scratch, with a new interconnect, is a multi-year endeavor. The company's investors seem willing to wait, given the size of the round.

What the Series B Says About the Market's Appetite for Risk

OLIX's €270.5 million Series B at a €2.8 billion valuation, just two years after founding, is a significant bet on the company's vision. The round included Fundomo, Arm, and Hudson River Trading, along with angel investors like Reed Hastings, co-founder of Netflix. The participation of Arm, a major player in chip IP, suggests that OLIX's approach has credibility within the semiconductor industry.

The valuation is notable for a company that has not yet shipped a product. It reflects the market's hunger for alternatives to Nvidia, and the belief that inference will be a major growth area. But it also raises the bar: OLIX will need to deliver on its promises to justify that valuation.

Naming the Factory: OLIX and the Brand of Infrastructure

The name 'OLIX' is short, punchy, and vaguely technical. It doesn't explicitly reference optics or inference, but it has a clean, almost chemical quality—like a compound or a material. The domain, olix.com, is a premium asset, adding to the brand's credibility.

The company's branding leans heavily on the 'token factory' metaphor, which is memorable and evocative. It positions OLIX not just as a chip maker, but as a builder of infrastructure. The word 'factory' suggests efficiency, scale, and industrial production—exactly the values OLIX wants to project.

The risk is that 'OLIX' is generic and could be mistaken for a software company or a chemical firm. But the brand is still young, and the 'token factory' narrative is strong enough to carry it. As the company ships its first products, the name will become more associated with the specific technology it brings to market.