Graphcore Scores $200M to Scale Up A.I. Chip Production

Xconomy San Francisco — 

One of the leading contenders in the A.I. chip race, Graphcore, announced Tuesday it pulled down $200 million in a Series D fundraising round that set its valuation at $1.7 billion.

The Bristol, U.K.-based startup, founded in 2016, has now secured a total of more than $300 million in financing from venture capital firms and other investors including Microsoft and the corporate venture arms of BMW and Dell Technologies.

Graphcore’s big haul is a notable milestone in the recent revival of investor interest in computer hardware, due to the demand for ever-greater processing capacity that can maximize the potential of artificial intelligence software. Venture firms are being joined by businesses eager to fund the development of chips that can help them realize their visions for smarter products such as self-driving cars and virtual assistants that understand speech in context.

Established chipmakers such as Nvidia (NASDAQ: NVDA) and Qualcomm (NASDAQ: QCOM), determined not to be left behind, are pursuing their own hardware innovations, while A.I. chip startups keep springing up. Tech giants such as Facebook and Alphabet unit Google also have in-house chip projects in the works.

Graphcore earned its first revenue this year with an early-access launch of its novel Intelligence Processor Unit (IPU), which was designed to provide the immense increase in computing power and speed required for machine learning operations. With its new capital, Graphcore says, it is scaling up production. (The IPU name is a play on the monickers for preceding generations of semiconductor units—the central processing unit or CPU, and the graphics processing unit or GPU. Nvidia’s GPUs have been adapted for A.I. operations, but they were not originally designed with artificial intelligence in mind.)

Graphcore also makes a companion software, Poplar, which prepares data to be processed by its IPUs. The technology duo boosts the speed of training intelligent machines by a factor of 10 to 100, Graphcore says.

“Our mission at Graphcore is to help innovators create new breakthroughs in machine intelligence,” co-founder and CEO Nigel Toon said in a statement.

The Series D funding round was co-led by venture capital firm Atomico, a returning investor, and by European investment firm Sofina, a new investor. Other first-time investors participating in the fundraising round were BMW i Ventures, Microsoft, Merian Chrysalis Investment Company Limited, and other funds managed by Merian Global Investors.

Also piling in on the funding round were previous investors including Sequoia Capital, Amadeus Capital Partners, Robert Bosch Venture Capital, C4 Ventures, Dell Technologies Capital, Draper Esprit, Foundation Capital, and Pitango.

In a drive to scale up the company quickly, Graphcore plans to hire engineers at its Bristol headquarters and at other offices in London, Palo Alto, CA, Oslo, Norway, and Beijing. Other new offices will open in Beijing and in Hsinchu, Taiwan.

The big Graphcore fundraising caps a busy year of innovations and investments in A.I. chips, including:

Two rival MIT spinouts that use light instead of electrical signals to adapt computers and chips for A.I. operations completed early funding rounds in February. Lightelligence, led by CEO Yichen Shen, raised a seed round of about $10 million, while Lightmatter, led by CEO Nick Harris, raised $11 million in a Series A round led by Boston venture firms Matrix Partners and Spark Capital.

Mythic, based in both Redwood City, CA, and Austin, TX, announced in March it had raised $40 million in a Series B fundraising round led by the venture capital arm of Japanese telecom and tech giant SoftBank.

UC Berkeley’s growing startup accelerator, SkyDeck, in October announced the launch of a special track for next-generation chip companies, offering upfront capital and services from partners to help with chip design and prototyping. The outside partners are San Jose, CA-based Cadence Design Systems, TSMC (Taiwan Semiconductor Manufacturing), and San Mateo, CA-based SiFive.

Mountain View, CA-based Esperanto Technologies, a startup A.I. chip developer, announced in November it had raised $58 million in a Series B fundraising round. It didn’t name its new investors, but the company’s previous backers include Western Digital.

Photo credit: Depositphotos