In context: OpenAI wants to free itself from its dependence on Nvidia chips, and CEO Sam Altman believes that the best way to achieve this is by developing its own semiconductor business. However, it’s an expensive proposition, so Altman is traveling the globe in search of backers. His latest stop is the United Arab Emirates.
OpenAI may be getting closer to securing the funds it needs to launch its own semiconductor business, aiming to reduce its dependence on Nvidia for AI processors. According to a report in the Financial Times, the company is in talks with the state-funded Abu Dhabi investment company MGX for backing in this venture. The report cites two individuals familiar with the discussions.
MGX is a newly launched, AI-focused fund chaired by the UAE’s national security adviser, Sheikh Tahnoon bin Zayed al-Nahyan. One of its objectives is to establish Abu Dhabi as a hub for AI development, utilizing the country’s financial resources for this purpose.
The report highlights OpenAI’s ongoing efforts to attract investors as it seeks financing for its ambitious, albeit costly, enterprise. Estimates suggest the project could require as much as $7 trillion.
Since at least 2022, OpenAI has been exploring various options for chip development, considering possibilities such as creating its own AI chip, acquiring a chip manufacturer, collaborating more closely with other providers like Nvidia, and diversifying its supplier base beyond Nvidia.
Earlier this year, reports surfaced that Altman was in discussions with investors, including the UAE government, to raise funds ranging from $5 trillion to $7 trillion for a tech initiative aimed at addressing constraints to OpenAI’s growth, including the shortage of AI chips. Altman denied seeking $7 trillion specifically for chip development but acknowledged that the investment required would be substantial.
Altman has emphasized the need for increased industry capacity, stating on X in February, “Building massive-scale AI infrastructure and a resilient supply chain is crucial to economic competitiveness. OpenAI will try to help.”
we believe the world needs more ai infrastructure–fab capacity, energy, datacenters, etc–than people are currently planning to build.
building massive-scale ai infrastructure, and a resilient supply chain, is crucial to economic competitiveness.
openai will try to help!
– Sam Altman (@sama) February 7, 2024
The reasons behind Altman’s ambitions are straightforward: there is a scarcity of the GPUs necessary to power OpenAI’s software, and the costs associated with running hardware are exceedingly high.
Altman has frequently complained about the shortage of chips, noting that OpenAI is severely limited by GPU availability, which has caused delays in many of their short-term plans. He also highlighted that the primary customer complaint relates to the reliability and speed of the API, with much of the issue stemming from GPU shortages.
The availability of GPUs also presents challenges for fine-tuning the API, as it requires significant computing resources to manage. Moreover, the GPU availability issue limits the capacity of its dedicated offering, which provides customers with a private copy of the model.
Then there are the costs to consider. OpenAI relies on a supercomputer developed by Microsoft, which utilizes 10,000 of Nvidia’s GPUs, making ChatGPT an expensive system to operate. Bernstein analyst Stacy Rasgon estimates that each query costs roughly four cents. Scaling up, if ChatGPT queries were to reach a tenth of the scale of Google search, it would necessitate approximately $48.1 billion worth of GPUs initially and about $16 billion worth of chips annually to sustain its operations.