Broadcom Jumps as OpenAI Bets Everything on Custom Chips

 bullish.beats
 
  bullish.beats
 
Follow  Message
 
 
 
Streaking.Winners tracks stocks achieving consecutive gains, focusing on consistent performers and market momentum to keep investors informed.
 

•  13 Oct 2025
 

•   AMD  
   
  -1.29%
•   AVGO  
   
  -0.25%
•   NVDA  
   
  +0.02%

OpenAI, the company behind ChatGPT, has signed a multi-year agreement with $AVGO Broadcom to jointly develop custom chips and networking equipment. This represents another step in the startup's ambitious strategy to expand its computational infrastructure, this time with an emphasis on direct control over core technology. Broadcom shares are jumping 10% following the announcement as the market recognizes the company is becoming a central player in the artificial intelligence world.

Under the agreement, OpenAI will design the hardware and collaborate with $AVGO Broadcom on practical development. The goal is to add 10 gigawatts of capacity to artificial intelligence data centers, with deployment of the first servers expected to begin in the second half of 2026. The timeline is ambitious but not impossible. The two companies intend to complete the full installation by the end of 2029, meaning about four years of intensive work on design, development, testing, and deployment at a scale unprecedented in the industry.

The custom processor adaptation will allow OpenAI to embed the knowledge it has accumulated from developing models and artificial intelligence services directly into hardware. This is supposed to unlock new levels of capability and intelligence, according to the company's statement. But beyond marketing slogans, this move indicates something more fundamental: a desire to become less dependent on external suppliers and build a technological advantage that is difficult to replicate. When a company controls the chips, it can set the pace and direction of innovation rather than adapting to what the market offers.

For $AVGO Broadcom, a manufacturer of components for products ranging from the iPhone to optical networks, this represents a leap into the boiling artificial intelligence market. The agreement confirms an arrangement that company CEO Hock Tan hinted at in an investor call last month, though he didn't detail the client's name at the time. The stock had risen 40% this year through last weekend, outperforming the 29% gain in the Philadelphia Semiconductor Index. The stock continues climbing as the market understood the full potential of the partnership.

OpenAI has been relentlessly active this year, signing several giant deals aimed at easing constraints on computing power, which has become the industry's real bottleneck. $NVDA Nvidia, whose chips handle most of today's artificial intelligence work, announced last month it would invest up to $100 billion in the company to support new infrastructure with a target of at least 10 gigawatts capacity. Last week, OpenAI published another agreement to deploy 6 gigawatts of $AMD AMD processors over several years.

But alongside purchasing chips from others, OpenAI is also working on designing its own semiconductors. These are primarily intended to handle the inference stage of running artificial intelligence models, the phase that comes after the technology has been trained, when the model needs to answer questions from real users. This is where custom chips can truly make a difference, because the requirements in the inference stage are very different from training stage requirements. Here you can build a real competitive advantage with hardware that is precisely tailored to the specific needs of the models being run.

Sam Altman, OpenAI's CEO, said the company has been working with $AVGO Broadcom for 18 months, long before the agreement became official. In a podcast the company released, he explained they are rethinking the technology starting from transistors, the most basic component in a chip, all the way to what happens when someone asks ChatGPT a question and receives an answer. Optimization across all these technology components can bring enormous improvements in efficiency: models that work faster, cost less to operate, and deliver better performance. This is a holistic approach that few companies can afford.

Broadcom CEO Hock Tan recently said: "If you manufacture your own chips, you control your own destiny." In a world where dependence on one or two suppliers can paralyze a company's growth and dictate schedules and prices, OpenAI has decided to spread risks and build real technological independence.

$AVGO Broadcom is increasingly seen as one of the main beneficiaries of artificial intelligence spending, which is driving the stock price this year. The 40% increase through last weekend is significant, especially as it outperforms the 29% rise in the Philadelphia Semiconductor Index that includes all the sector's giants. OpenAI, meanwhile, has achieved a valuation of $500 billion, making it the world's largest startup by any measure. The combination of these two companies creates an interesting dynamic, with Broadcom bringing knowledge in manufacturing and hardware development while OpenAI brings understanding of what's needed to run artificial intelligence models optimally.

By using $AVGO Broadcom's networking technology, OpenAI is also reducing its dependence on $NVDA Nvidia. Broadcom's Ethernet-based options offer a real alternative to Nvidia's proprietary technology, which until now has dominated the market for communication between artificial intelligence servers almost without challenge. OpenAI will also design its own networking equipment as part of working on custom hardware, adding another layer of control and flexibility.

Broadcom won't build the data centers itself, as that's not its domain. Instead, it will manufacture and deploy servers with the custom hardware in facilities managed by OpenAI or its cloud computing partners, like $MSFT Microsoft, which is also the company's largest investor. This is a different arrangement from deals with $NVDA Nvidia and $AMD AMD. Here there's no component of mutual investment or stock exchange, just a pure development and supply deal.

One gigawatt is roughly the capacity of a regular nuclear power plant, representing an enormous amount of energy. But 10 gigawatts of computing power alone isn't enough to support OpenAI's vision of achieving artificial general intelligence, systems that can perform any intellectual task that humans are capable of performing. Greg Brockman, co-founder and president of the company, said after the deal that this is a drop in the ocean compared to what needs to be achieved to realize the vision.

View All Headlines