Corporations require powerful graphics processing units (GPUs) to fuel their artificial intelligence systems. However, obtaining these chips has proven to be a significant challenge, given that only a limited number of semiconductor companies manufacture them.
Among the prominent players heavily involved in the field of AI is OpenAI, the creator of ChatGPT. According to a report from Reuters, OpenAI appears to be exploring the possibility of developing its own AI chip internally. Additionally, there is speculation that the company may be considering acquiring a chip manufacturer to ensure a consistent supply of customized chips.
Nvidia is the Big Player in the Chip Market
NVIDIA currently holds a dominant position in the chip market. OpenAI, along with many of its competitors, relies on AI chips developed by NVIDIA, specifically the A100 and H100, which are among the most sought-after chips in the industry. OpenAI boasts a wealth of GPUs, a status explained by Dyan Patel in his blog as being a company with substantial access to computing power.
One of OpenAI’s standout AI products is ChatGPT, which operates on a whopping 10,000 of NVIDIA’s top-tier GPUs. These computing chips come at a significant cost. In fact, both NVIDIA and AMD have raised the prices of certain chips and graphics cards over the past year.
OpenAI Bleeding Money Powering AI Software
This substantial financial outflow has been detailed in a Reuters report, which indicates that each query processed by ChatGPT costs OpenAI approximately 4 cents, as analyzed by Bernstein analyst Stacy Rasgon.
If ChatGPT were to scale its queries to even a fraction of Google’s search volume, OpenAI would need to allocate approximately $48.1 billion for the acquisition of required GPUs and an additional annual expenditure of approximately $16 billion for chips to sustain operations. Therefore, it becomes logical for the company to consider in-house chip production.
OpenAI’s CEO, Sam Altman, has repeatedly addressed the GPU shortage issue in recent times. In a now-archived blog post by Raza Habib, CEO of the London-based AI firm Humanloop, Sam Altman acknowledged concerns about the API’s slow speed and reliability problems faced by OpenAI’s customers. He attributed a significant portion of these challenges to the scarcity of GPUs.
Habib also mentioned that Altman’s vision for the company includes the development of a more affordable and faster GPT-4. He emphasized that OpenAI’s overarching goal is to drive down the cost of intelligence as much as possible, and they are committed to continually reducing the cost of their APIs over time.
OpenAI’s Big Moves in AI Hardware
According to sources knowledgeable about OpenAI’s intentions, Reuters also disclosed that OpenAI is contemplating the possibility of acquiring a chip manufacturing company, similar to Amazon’s acquisition of Annapurna Labs in 2015.
Moreover, as per a report from The Information, OpenAI might be considering a more ambitious approach beyond merely exploring AI chip development. There are indications that OpenAI could venture into the realm of hardware products. The report suggests that Sam Altman, OpenAI’s CEO, has engaged in discussions with Jony Ive, the former Chief Design Officer at Apple, with the goal of creating a device akin to an iPhone.
Notably, there are reports indicating that Masayoshi Son, the CEO and investor of SoftBank, is considering an investment of $1 billion in OpenAI to facilitate the development of a product referred to as the ‘iPhone of AI.’
OpenAI faces challenges in obtaining powerful GPUs for AI systems. OpenAI, a significant player in AI, is exploring in-house chip development and potential acquisitions to secure chip supplies. NVIDIA dominates the chip market, but the cost of GPUs poses financial challenges for OpenAI.
To address these issues, OpenAI’s CEO, Sam Altman, envisions reducing costs and exploring hardware products. Additionally, there is potential investment from SoftBank to create a groundbreaking AI device. These developments reflect OpenAI’s determination to shape the future of AI hardware.