OpenAI, one of many best-funded AI startups in enterprise, is exploring making its personal AI chips.
Discussions of AI chip methods inside the firm have been ongoing since at the least final 12 months, in response to Reuters, because the scarcity of chips to coach AI fashions worsens. OpenAI is reportedly contemplating plenty of methods to advance its chip ambitions, together with buying an AI chip producer or mounting an effort to design chips internally.
OpenAI CEO Sam Altman has made the acquisition of extra AI chips a high precedence for the corporate, Reuters studies.
Presently, OpenAI, like most of its rivals, depends on GPU-based {hardware} to develop fashions reminiscent of ChatGPT, GPT-4 and DALL-E 3. GPUs’ capacity to carry out many computations in parallel make them well-suited to coaching right now’s most succesful AI.
However the generative AI increase — a windfall for GPU makers like Nvidia — has massively strained the GPU provide chain. Microsoft is dealing with a scarcity of the server {hardware} wanted to run AI so extreme that it would result in service disruptions, the corporate warned in a summer time earnings report. And Nvidia’s best-performing AI chips are reportedly bought out till 2024.
GPUs are additionally important for working and serving OpenAI’s fashions; the corporate depends on clusters of GPUs within the cloud to carry out clients’ workloads. However they arrive at a sky-high value.
An evaluation from Bernstein analyst Stacy Rasgon discovered that, if ChatGPT queries grew to a tenth the dimensions of Google Search, it’d require roughly $48.1 billion price of GPUs initially and about $16 billion price of chips a 12 months to maintain operational.
OpenAI wouldn’t be the primary to pursue creating its personal AI chips.
Google has a processor, the TPU (quick for “tensor processing unit”), to coach massive generative AI programs like PaLM-2 and Imagen. Amazon affords proprietary chips to AWS clients each for coaching (Trainium) and inferencing (Inferentia). And Microsoft, reportedly, is working with AMD to develop an in-house AI chip known as Athena, which OpenAI is alleged to be testing.
Actually, OpenAI is in a robust place to speculate closely in R&D. The corporate, which has raised over $11 billion in enterprise capital, is nearing $1 billion in annual income. And it’s contemplating a share sale that would see its secondary-market valuation soar to $90 billion, in response to a current Wall Avenue Journal report.
However {hardware} is an unforgiving enterprise — notably AI chips.
Final 12 months, AI chipmaker Graphcore, which allegedly had its valuation slashed by $1 billion after a cope with Microsoft fell by, mentioned that it was planning job cuts as a result of “extraordinarily difficult” macroeconomic surroundings. (The scenario grew extra dire over the previous few months as Graphcore reported falling income and elevated losses.) In the meantime, Habana Labs, the Intel-owned AI chip firm, laid off an estimated 10% of its workforce. And Meta’s customized AI chip efforts have been beset with points, main the corporate to scrap some its experimental {hardware}.
Even when OpenAI commits to bringing a customized chip to market, such an effort may take years and price tons of of hundreds of thousands of {dollars} yearly. It stays to be seen if the startup’s buyers, certainly one of which is Microsoft, have the urge for food for such a dangerous guess.