OpenAI Is Making Its Own AI Chips to Challenge NVIDIA

The Financial Times reports that OpenAI is preparing to enter mass production of its in-house AI chips starting in 2026, in partnership with Broadcom, with the aim of meeting soaring computational demands while reducing its heavy reliance on NVIDIA GPUs.

Broadcom CEO Hock Tan recently revealed that the company had secured a chip design order worth as much as $10 billion. Although he did not disclose the client, industry insiders confirmed that the order came from OpenAI, noting that the customized chips are intended solely for internal use and are not planned for external sale.

As early as 2023, OpenAI CEO Sam Altman publicly lamented the shortage of GPUs, which he said constrained the speed and reliability of the company’s API services. Reports at the time suggested that OpenAI was already exploring the possibility of building its own chips, engaging in talks with Broadcom and TSMC.

Following the launch of GPT-5, which drove a surge in computing demand, OpenAI not only set out plans to double its server capacity within five months but also began pursuing in-house chip development to offset potential GPU shortages and cut long-term hardware costs.

It remains uncertain whether TSMC will continue to play a role in the project. However, Broadcom’s expertise in networking chip design, combined with TSMC’s leadership in advanced process manufacturing, positions both companies as crucial partners in OpenAI’s bid for hardware autonomy.

OpenAI’s chip initiative is not merely a short-term response to supply constraints—it has the potential to reshape the broader AI semiconductor landscape.

NVIDIA continues to dominate the AI market, reporting a 56% year-over-year revenue increase in its latest quarterly results, even as U.S. export restrictions to China delay shipments of its H20 chips.

With Google, Amazon, and Microsoft already developing their own AI processors, OpenAI’s entry into custom chipmaking underscores the industry’s rapid shift toward “XPU” heterogeneous computing architectures, designed to optimize AI workloads while reducing dependence on any single supplier.

For OpenAI, the move is less about challenging NVIDIA’s market leadership than about securing the stability of its own models and services, ensuring they are not constrained by supply bottlenecks or rising costs. Proprietary chips would also allow tighter integration of model-specific requirements with hardware design, enabling greater efficiency and lower latency.

As the 2026 production timeline draws near, the success of OpenAI’s chip program will be pivotal not only to the company’s own growth trajectory but also to the broader wave of AI infrastructure self-sufficiency—marking a shift in the global AI race from a contest of algorithms to a new era of software–hardware co-design.

Related Posts:

• Apple’s “Made in India” Revolution: iPhone 17 Production Ramps Up Ahead of Launch
• Facebook plans to develop its own chip
• The EU unease about Broadcom attempts to buy Qualcomm: Privacy data may be leaked
• Samsung Eyes Perplexity AI for Galaxy S26, Less Google Reliance
• Xbox AI Gaffe: Layoffs, Ill-Timed Recruitment, and a Questionable Image