OpenAI is moving forward with plans to reduce its reliance on Nvidia by developing its first generation of in-house artificial intelligence silicon to meet its chip supply.

ChatGPT maker will complete the design of its first in-house chip in the coming months and plans to send it to Taiwan Semiconductor Manufacturing Co. for manufacturing, sources told Reuters. The process of getting the first design into a chip factory is called "tapeout." OpenAI and TSMC declined to comment.

The latest news shows that OpenAI is on track to achieve its ambitious goal of mass production at TSMC in 2026. A typical tapeout costs tens of millions of dollars and takes about six months to produce a finished chip, unless OpenAI pays more to speed up manufacturing. No one can guarantee that the silicon will work properly the first time it is taped out, and if it fails, the company will need to diagnose the problem and repeat the tapeout steps.

Sources said that within OpenAI, this training-focused chip is seen as a strategic tool to enhance OpenAI's bargaining chips with other chip suppliers. After launching the first chip, OpenAI's engineers plan to develop more powerful and advanced processors with each iteration.

If the initial tapeout goes smoothly, the ChatGPT maker will be able to mass produce its first in-house AI chip and may test alternatives to Nvidia chips later this year. OpenAI plans to submit its design to TSMC this year, indicating that the startup has made rapid progress on its first design, a process that may take years for other chip designers.

Big tech companies like Microsoft and META have struggled to produce satisfactory chips despite years of efforts. The recent market crash caused by Chinese AI startup DeepSeek has also raised questions about whether fewer chips will be needed to develop powerful models in the future.

The chip was designed by an internal team at OpenAI led by Richard Ho, who doubled his team to 40 in the past few months and worked with Broadcom. Richard Ho joined OpenAI more than a year ago from Alphabet's AI company. He helped lead the search giant's custom AI chip project at Google.

OpenAI partners with Broadcom and TSMC to develop first chip

OpenAI is reportedly working with Broadcom and TSMC to build its first in-house chips to power artificial intelligence systems, while increasing demand for AMD chips to meet its growing infrastructure needs.

OpenAI, the fast-growing company behind ChatGPT, has explored multiple options to diversify its chip supply and reduce costs. OpenAI considered making all its products in-house and raising funds for a costly plan to build a network of chip manufacturing factories, known as "foundries."

The company has now abandoned its ambitious foundry plans because of the cost and time required to build a network, and plans to focus on internal chip design work instead, according to the sources, who requested anonymity because they are not authorized to discuss private matters.

The company's strategy, detailed here for the first time, highlights how the Silicon Valley startup is using industry partnerships and a combination of internal and external approaches to secure chip supply and manage costs, just like larger rivals Amazon, Meta, Google and Microsoft. As one of the largest buyers of chips, OpenAI's decision to source from a variety of chip manufacturers while developing its custom chips could have broader implications for the tech industry.

Broadcom shares surged after the report, closing up more than 4.5% on Tuesday. AMD shares also extended early gains, closing up 3.7%.

OpenAI, AMD and TSMC declined to comment. Broadcom did not immediately respond to a request for comment.

OpenAI, which has helped commercialize generative AI, which can respond to queries in a human-like manner, relies on massive computing power to train and run its systems. One of the largest buyers of Nvidia graphics processing units (GPUs), OpenAI uses AI chips to train models, which allow AI to learn from data, and to perform inference, which applies AI to make predictions or decisions based on new information.

Reuters previously reported on OpenAI's chip design efforts. The Information reported on talks with Broadcom and others.

OpenAI has been working with Broadcom for months to build its first AI chip focused on inference, according to sources. Training chips are in greater demand right now, but analysts predict that demand for inference chips could exceed that of training chips as more AI applications are deployed.

Broadcom helps Alphabet and others

Broadcom helps companies like Alphabet fine-tune chip designs for manufacturing and provides parts of the designs that help quickly transfer information to and from the chips. This is important for artificial intelligence systems because thousands of chips are linked together to work together.

OpenAI is still determining whether to develop or acquire other elements of its chip design and may bring in more partners, two sources said.

The company has assembled a chip team of about 20 people led by top engineers who built the tensor processing unit (TPU) at Google, including Thomas Norrie and Richard Ho.

Through Broadcom, OpenAI has reached a manufacturing capacity with Taiwan Semiconductor Manufacturing Co. and plans to build its first custom chip in 2026, the sources said. They said this timeline could change.

Currently, Nvidia's GPUs have a market share of more than 80%. But due to shortages and rising costs, major customers such as Microsoft, Meta and now OpenAI are beginning to explore internal or external alternatives.

OpenAI plans to use AMD chips through Microsoft's Azure, first reported here, suggesting that AMD's new MI300X chip is trying to take a bite out of a market dominated by Nvidia. AMD expects to generate $4.5 billion in AI chip sales in 2024, with the chip set to launch in the fourth quarter of 2023.

Training AI models and operating services like ChatGPT are expensive. OpenAI expects to post a $5 billion loss this year on $3.7 billion in revenue, according to the sources. Computing costs — the hardware, electricity and cloud services needed to process large data sets and develop models - are the company's biggest expense, prompting efforts to optimize utilization and diversify suppliers.

OpenAI has been cautious about poaching talent from Nvidia, the sources added, because it wants to maintain a good relationship with the chipmaker it remains committed to, especially to get its new generation of Blackwell chips.

Nvidia declined to comment.

Source: Content compiled from Reuters

View more at EASELINK