OpenAI is reportedly developing its first custom AI chip with Broadcom. The chip could be manufactured as early as 2026. The move could help reduce the cost of running OpenAI-powered apps. be
OpenAI is one step closer to developing its first AI chip, as the cost of cloud computing and the number of developers creating apps on its platform soar, according to a new report.
It was first reported in July that the ChatGPT maker was in talks with multiple chip designers, including Broadcom. Now, Reuters claims that a new hardware strategy could see OpenAI settle on Broadcom as its custom silicon partner, with its chips potentially launching in 2026.
By then, it looks like OpenAI will be adding AMD chips to Microsoft Azure systems alongside Nvidia’s existing chips. The AI giant’s plans to create a “foundry” (a network of chip factories) have been scaled back, Reuters reported.
The reported reason for this move is to reduce the ballooning costs of AI-powered applications. OpenAI’s new chips will apparently not be used to train generative AI models (the domain of Nvidia chips), but will instead run AI software and respond to user requests.
At today’s DevDay London event (following the San Francisco version on October 1st), OpenAI announced some improved tools it’s using to attract developers. Our largest API, the Real-Time API, is essentially an advanced voice mode for app developers, and we’ve added five new voices with increased range and expressiveness.
Three million developers around the world are currently using OpenAI’s API (application programming interface), but the problem is that many of its features are still too expensive to run at scale.
OpenAI says it has reduced the price of API tokens (i.e. how much it costs developers to use its models) by 99% since the release of GPT-3 in June 2020, but there is still a long way to go. is. AI chips could be an important step toward making AI-powered apps cost-effective and truly mainstream.
Apps powered by OpenAI are now available
(Image source: Getty Images)
The prohibitive costs of cloud AI processing remain a deterrent to apps incorporating OpenAI’s tools into their products, but some startups are already taking the plunge.
Veed, a popular online video editor, is connected to several OpenAI models and offers features such as automatic transcription and the ability to select the best soundbites from long-form videos. An AI-powered notepad called Granola also leverages GPT-4 and GPT-4o to transcribe meetings and send follow-up tasks without having a meeting bot join the call.
Apart from consumer apps, a startup called Tortus is using GPT-4o and OpenAI’s voice models to assist doctors. The tool can listen to conversations between doctors and patients and automate many administrative tasks, such as updating health records, while also clearly improving diagnostic accuracy.
Potential privacy and hallucination concerns with AI models aside, developers are clearly keen to harness the power of OpenAI’s tools, and its low-latency conversational voice mode has great potential for customer service. There is no doubt that it is.
Still, while it is likely that calls to stores and customer service lines will soon be made using one of OpenAI’s voice models, the running costs of these AIs may slow adoption rates. there is. As such, OpenAI appears to be keen on developing its own AI chip. Sooner or later.
