Dropped out of school at 16, and by the age of 25 had built an AI chip unicorn valued at 22.3 billion yuan, which has now raised another 2.1 billion yuan in financing.

A 16-year-old British boy who dropped out of high school spent 9 years growing his third company to a valuation of 3.3 billion US dollars (about 22.3 billion yuan). 

His newly founded AI chip company OLIX has just closed a $312 million Series B financing round, with both Arm and Netflix co-founder participating in the investment. Only half a year after the previous round of financing, its valuation has surged from $1 billion to $3.3 billion. 

OLIX's chips have not yet been manufactured, and the first batch of products is scheduled to be delivered in the second half of 2027. 

Why are investors scrambling to pay for it right now? 

The answer lies in his re-interpretation of "AI inference". 

Targeting the Trillion-Dollar Inference Market

James Dacombe, founder of OLIX, dropped out of high school at the age of 16. At 18, he founded CoMind, a brain monitoring startup that develops medical devices for monitoring brain activity. The company has raised $100 million (about 680 million yuan) to date. 

In 2024, 23-year-old Dacombe founded OLIX in London to design AI inference chips. 

Dacombe judges that AI inference is becoming the next trillion-dollar market, and the existing hardware solution, NVIDIA's GPU, is not designed for inference. He used a metaphor: a data center is like a factory that produces text, and generating a piece of text requires many processes, but the current factory runs all processes on the same machine. 

His solution is to equip each process with a dedicated machine. 

This is the core idea of OLIX's X-1 platform — break down tasks and distribute them to a large number of chips to run simultaneously, with each chip focusing only on processing a specific part of the entire task. 

Traditional chips use copper wires to transmit data, which leads to data loss, while OLIX uses optical signals instead. Light travels in glass fibers just like driving on an unobstructed highway: no interference, no loss, and lower power consumption.

More uniquely, OLIX adopts an optical interconnection technology called "slow and wide". The traditional approach is "fast and narrow", which builds several highways with extremely high speed but is prone to accidents. Extra money has to be spent on guardrails — error correction circuits.

"Slow and wide" works the opposite way. It builds a large number of ordinary roads, each with a low speed limit but no accidents. No guardrails are needed. The cost is lower and the system is more reliable.

OLIX's optical interconnection technology architecture Source: ConvergeDigest 

Targeting large models with hundreds of billions of parameters, DX-1 can generate more than 10,000 tokens per second for a single user, and is more power-efficient than general-purpose chips. When multiple machines are connected together, it can handle super-large-scale models with 10 trillion parameters. 

Customers of DX-1 include data centers, cloud service providers, and large model companies. 

OLIX does not sell chips separately, but delivers full racks — the chips, supporting optical components, network equipment and software are all packaged into a complete cabinet that customers can plug in and use right after purchase. Electricity cost is the second largest expense of data centers after equipment, and the electricity cost saved by optical interconnection directly translates to profits. 

DX-1 is scheduled to deliver the first batch to customers in the second half of 2027. 

Valuation Triples in Half a Year

The lead investor of this round is Fundomo, an investment institution based in New York. The three most notable names on the co-investor list are: 

Arm — the world's largest chip design company whose technology is used in almost all mobile phone chips. Arm's investment in OLIX indicates that this chip giant has also spotted the opportunities in next-generation AI computing. 

Reed Hastings — co-founder of Netflix, who invested in his personal capacity. 

Hudson River Trading — a Wall Street quantitative trading giant. Quantitative firms invest in chip companies because they themselves are major buyers of computing power: the cheaper and faster the computing power is, the more profitable their trading strategies will be. 

All existing shareholders also participated in the follow-on investment, including Hummingbird Ventures, Crane, Plural, Creandum, etc. 

In February this year, OLIX just completed a $220 million Series A financing round, with a valuation of just over $1 billion. In half a year, the valuation jumped from $1 billion to $3.3 billion, tripling. According to media reports, OLIX's total financing has exceeded $562 million (about 3.8 billion yuan) so far. 

The company currently has offices in London, Bristol, Toronto, Austin and San Francisco. It has also recruited Matt Beales, former CFO of British fintech company Wise, as its chief financial officer, and Nick McKeown, professor at Stanford University and co-inventor of software-defined networking technology, to join its board of directors. 

The Track Is Red Hot, But Can OLIX Win?

The AI inference chip track is heating up rapidly. 

In July 2026, AI inference chip company Etched closed a $300 million Series C financing round with a valuation of $10.3 billion — this company designs dedicated inference chips for Transformer models. SambaNova completed a $1 billion financing round in July at a valuation of $11 billion, focusing on enterprise-level AI inference. Fractile, another British inference chip company, closed a $220 million financing round in May. 

Capital is voting with real money: inference is the next main battlefield for AI. 

OLIX's advantages are obvious: its technical route is different from others (optical interconnection + no HBM required + full rack delivery), the founder is young and daring to take risks, and its investor roster is extremely strong. 

But the challenges are equally significant: 

The chips have not been manufactured yet. DX-1 will not be delivered until the second half of 2027. In the chip industry, every step from design to mass production may go wrong, and it is normal that plans often fail to keep up with changes. 

There are too many competitors. NVIDIA will not sit back and watch others take its business. Groq, Cerebras, Etched, SambaNova, and Fractile are all competing for the same market share. Whoever achieves large-scale mass production first and secures big customers first will take the lead. 

It is not easy for customers to switch. Data centers are accustomed to NVIDIA's entire system. Switching to a new platform requires time, confidence, and proof that "you are truly better than NVIDIA". 

Before a single chip is even produced, a group of investors have already valued the company at $3.3 billion. But $3.3 billion is only the starting point, and the tapeout result in 2027 is the real test — whether the chips can be manufactured on schedule, whether the performance can meet the standards, and whether the first batch of customers are willing to place repeat orders. What determines whether OLIX can truly challenge NVIDIA is not its current valuation, but the results two years from now. 

The content of this article is for reference only and does not constitute investment advice. 

This article is from the WeChat Official Account "Pencil News" (ID: pencilnews), written by Song Ge, and published with authorization from 36Kr.