In what appears to clarify weeks of speculation across the semiconductor industry, Charlie Kawwas, president of Broadcom’s Semiconductor Solutions Group, has confirmed that OpenAI is not the mystery $10 billion customer that Broadcom referenced during its September earnings call.
Kawwas made the clarification on Monday during a joint appearance with OpenAI President Greg Brockman on CNBC’s “Squawk on the Street.” The two executives discussed their newly announced partnership to jointly build and deploy 10 gigawatts of custom artificial intelligence accelerators — a project that analysts had widely assumed made OpenAI the unidentified mega-client in Broadcom’s order books.
“I would love to take a $10 billion [purchase order] from my good friend Greg,” Kawwas said on air. “He has not given me that PO yet.”
Register for Tekedia Mini-MBA edition 19 (Feb 9 – May 2, 2026): big discounts for early bird.
Tekedia AI in Business Masterclass opens registrations.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab: From Technical Design to Deployment (next edition begins Jan 24 2026).
The comment came just hours after the companies officially announced a long-term plan to develop and deploy a new class of AI infrastructure designed specifically for OpenAI’s frontier models, confirming months of collaboration rumors between the two firms.
Broadcom did not immediately respond to CNBC’s follow-up request for comment on Kawwas’s statement.
A Partnership Built for AI Scale
OpenAI’s collaboration with Broadcom represents one of the most ambitious infrastructure initiatives in the AI sector, aiming to create a custom line of chips optimized for training and inference at massive scale. The two companies said they will begin deploying racks of these accelerators by late 2026, with full completion of the 10-gigawatt deployment expected by 2029.
“By building our own chip, we can embed what we’ve learned from creating frontier models and products directly into the hardware, unlocking new levels of capability and intelligence,” Brockman said in a statement.
The move is part of OpenAI’s rapid expansion of compute infrastructure to keep up with demand from its ChatGPT platform and enterprise products, which now serve hundreds of millions of users globally.
Broadcom’s Mystery Customer Still Unknown
During Broadcom’s September earnings call, CEO Hock Tan disclosed that a fourth large customer — alongside Google, Meta, and ByteDance — had placed a $10 billion order for custom AI chips. The revelation immediately sparked speculation among analysts and investors, with many pointing to OpenAI as the likely client due to its aggressive data center buildout.
However, Kawwas’s confirmation on Monday effectively rules out OpenAI, leaving the identity of the mystery customer unknown.
Broadcom has not disclosed the client’s name, citing confidentiality agreements, but analysts have suggested potential candidates, including Amazon Web Services, Oracle, or Tesla’s xAI project — all of which are investing heavily in AI compute infrastructure.
The order, according to Tan, will significantly boost Broadcom’s AI revenue forecast for next year, with shipments of the new custom chips expected to begin in 2026.
OpenAI’s Infrastructure Blitz
OpenAI’s infrastructure expansion has been nothing short of aggressive. Over the past few months, the company has struck multi-billion-dollar agreements with major chipmakers, including Advanced Micro Devices (AMD), Nvidia, and CoreWeave — all part of an effort to diversify and secure long-term compute supply.
The startup, now valued at $500 billion, is working to scale up capacity in anticipation of its next-generation AI models. These deals have made OpenAI one of the largest buyers of GPUs and AI accelerators globally, rivalling hyperscalers such as Google and Amazon in chip procurement.
Its partnership with AMD, for instance, involves deploying six gigawatts of Instinct GPUs over multiple hardware generations, while the deal with CoreWeave focuses on providing cloud infrastructure optimized for OpenAI’s workloads.
Broadcom’s custom chip collaboration, therefore, fits squarely into OpenAI’s broader strategy of building proprietary hardware systems that could eventually reduce its dependence on traditional suppliers like Nvidia.
Broadcom’s Expanding AI Footprint
For Broadcom, the deal with OpenAI cements its growing role in the AI semiconductor ecosystem — an area historically dominated by Nvidia. The company designs and manufactures high-performance custom chips used by some of the largest tech firms, including Google’s Tensor Processing Units (TPUs) and Meta’s AI servers.
Broadcom’s AI-related revenue has grown sharply, and CEO Hock Tan recently forecast that AI chip sales could exceed $10 billion in 2025, accounting for nearly a quarter of the company’s semiconductor revenue.
Still, the firm’s decision not to name its fourth major customer has kept analysts guessing. Some speculate the client could be a nontraditional tech player developing in-house AI capabilities, or even a government-backed initiative seeking long-term chip supply security.
A Race to Control AI Hardware
The partnership between OpenAI and Broadcom underscores a broader industry trend, where AI companies are increasingly moving toward designing their own chips to optimize performance, efficiency, and cost.
Rival firms have followed similar paths. Google developed its own Tensor chips for AI and cloud computing, Amazon built its Trainium and Inferentia chips for AWS, and Meta announced plans earlier this year to deploy its custom Meta Training and Inference Accelerator (MTIA) chips across data centers.
OpenAI has joined that elite circle — a move analysts believe will give it a competitive advantage as AI workloads grow exponentially.
As Kawwas and Brockman appeared side-by-side on CNBC, their comments made clear that while OpenAI may not be Broadcom’s secret $10 billion customer, the two companies are forging one of the most significant partnerships in the AI chip race.