AI Chip Biz Graphcore Seeks Capital To Remain Going Concern

AI chip startup Graphcore must raise new funds from investors within the next few months in order to offset mounting losses incurred over the prior financial year, and remain a "going concern."

The company, based in Bristol in the southwest of England, was touted by some as a potential rival for GPU-giant Nvidia in the AI accelerator arena with its Intelligence Processor Unit (IPU) hardware, associated Poplar software stack, and IPU cloud services.

However, Graphcore Limited Group reported a loss before tax for the year ended 31 December 2022 of $204.6 million, compared to a pre-tax loss of $184.5 million a year earlier. The accounts were filed on October 4.

The losses were largely due to increased operating costs, primarily attributable to investment in research and development, and Graphcore expects further losses for 2023 as it continues to invest in product development.

Revenue for 2022 was $2.7 million, down 46 percent year-on-year. Graphcore said it was "hit by headwinds in the wider macroeconomic environment which resulted in softer demand for hardware sales. As a result of a strategic review, in September the group made the decision to close its operations in Norway, Japan and South Korea and make smaller headcount reductions in the US and UK. These exercises were completed in Q4 of 2022."

Headcount shrank fron 631 in 2021 to 494 last year, the accounts filed with UK company repository, Companies House, confirm.

The business exited 2022 with $157 million in cash and short-term investments.

In the latest annual report, Graphcore confirmed it has prepared a detailed cashflow forecast through to December 2027 which indicates the Group needs further funding before a planned cashflow break-even point is reached.

The forecast indicates that Graphcore will require more money to finance existing requirements.

The 2022 Profit and Loss accounts state: “The Group and Company are in discussions with potential investors to secure the additional funding required, but has not yet reached agreement. However, the Directors expect that appropriate funding can be secured before it is required.

“On this basis, the validity of the going concern assumption depends principally on the ability of the directors to secure additional funding and the Group to achieve its forecast level of revenue and cost bases for 2023. These conditions constitute a material uncertainty in relation to going concern.“

Graphcore head of communications Iain Mackenzie told us: “Obviously, it’s a little while since we completed and filed the accounts, but we don’t have any update to share publicly today. We’re also not going to share the amount of funding sought at the moment."

Graphcore has so far raised more than $600 million from investors, according to the Financial Times, and the company hit a market valuation of $2.8 billion back in 2020.

Its technology has received favorable appraisals, including from our sister publication, The Next Platform, and was mentioned earlier this year in connection with a potential UK supercomputer proposed by Prime Minister Rishi Sunak.

It appears, however, the popularity of Nvidia GPUs for accelerating some AI workloads is capturing much of the capital funding, with the US biz recently given a stock market valuation in excess of $1 trillion.

Customers are taking longer with purchasing decisions for hardware, Graphcore says in the P&L accounts. "In contrast, the cloud computing market for AI workloads has shown strong growth and is expected to overtake on-premises hardware sales in the coming years," the accounts state.

Graphcore’s annual report says that its near-term goal is to increase IPU adoption with a focus on delivering this through the cloud.

The company is going to keep investing in research and development to expand the use cases for its tech, and expand its eco system. The group will also continue to work on designing its next generation technology.

In other related news, it has emerged that OpenAI, the Microsoft-backed company behind the ChatGPT large language model, is considering making its own AI chips and may be eyeing a potential acquisition to achieve this.

• When is a PC an AI PC? Nobody seems to know or wants to tell
• China suggests America 'carefully consider' those chip investment bans
• TSMC warns AI chip crunch will last another 18 months
• Intel slaps forehead, says I got it: AI PCs. Sell them AI PCs

According to Reuters sources, the company has considered various options over the past year to solve a worldwide shortage of AI accelerators that OpenAI relies on.

OpenAI was not immediately available to comment, but SambaNova, another company that makes its own AI silicon, said the move comes as no surprise.

In a statement sent to The Register, SambaNova EMEA GM Alex White said: “It’s driven by a shortage of AI chips in the market which is hurting OpenAI’s ability to scale to service enterprise customers. These bigger businesses want models which are more bespoke than the consumer-oriented GPT-4, but OpenAI can’t deliver these yet.” ®