Nvidia, a global leader in technology and graphics processor development, finds itself forced to radically rethink the design of its next-generation artificial intelligence chip platform, named ‘Feynman’. This striking development comes after the company failed to secure sufficient production capacity using TSMC’s latest and most advanced manufacturing processes, according to recent Taiwanese financial press reports.
Alternative plan and complex engineering compromises
Under the revised plan that Nvidia was forced to adopt, only the most critical and sensitive components of the Feynman AI chip will be manufactured using TSMC’s ultra-advanced 1.6nm ‘A16’ manufacturing node process. In contrast, the production of less critical parts of the processor will be shifted to the relatively older 3nm-based ‘N3B’ node.
This compromise solution resorted to by the company may lead to a series of engineering compromises in the chip’s architectural design, as well as causing a noticeable increase in production costs. More importantly, this move could create future supply constraints for a chip platform that is pivotal and essential to Nvidia’s expansion ambitions in generative AI and deep learning.
The bottleneck in production capacities
TSMC’s advanced manufacturing operations, specifically 2nm and below, are facing massive and unprecedented demand from AI and high-performance computing customers. This rush for microchips has fully booked the company’s production capacity through 2028, and this deficit may extend beyond that year.
In a related context, financial institution JPMorgan Chase predicted that production capacity for the 3nm node in TSMC’s factories will reach maximum capacity before 2026. To counter this crisis, the Taiwanese company is aggressively expanding its operations, having submitted environmental review documents to establish a massive new manufacturing facility spanning 15.46 hectares in the Tainan Science Park in southern Taiwan. Construction work on this factory is scheduled to begin this year, with the project slated for completion and stable operations by 2028.
Due to tight supplies and scarce production capacity, TSMC is expected to raise its service prices, which will strengthen the company’s pricing power and dominance over the entire semiconductor industry. In this regard, Jensen Huang, CEO of Nvidia, stated earlier this year that his company is facing exceptionally strong demand for its products, and predicted that TSMC’s total production capacity will more than double over the next decade to meet these massive needs.
Position of the ‘Feynman’ architecture on the roadmap
Nvidia first showcased details of the Feynman architecture during the GTC 2026 conference held in San Jose on March 15, where Huang explained that this platform will be the direct successor to the ‘Vera Rubin’ chip family. The Feynman AI chip, built on TSMC’s A16 process, features ‘backside power delivery’ technology, a revolutionary technique that increases processor efficiency. These chips are targeted for official launch in 2028, with customer deliveries expected to extend into 2029 or 2030.
Supply chain diversification and potential partnerships
Industry analysts at specialized institution TrendForce pointed out that Nvidia may not rely entirely on Taiwan, but rather seek to diversify its Feynman chip supply chain. Circulating reports indicate that the company is studying the possibility of cooperating with American company Intel to produce less complex components, such as the chip’s input/output dies.
- Strategic impact: This crisis highlights the strategic leverage enjoyed by TSMC as the globally dominant manufacturer of advanced chips.
- Physical constraints: These developments highlight the escalating tension between exploding demand for AI technologies and the physical and economic limits of available manufacturing capacity.
- Rapid adaptation: Nvidia’s decision demonstrates significant flexibility in re-engineering processors to ensure their continued presence in the markets without prolonged downtime.
Ultimately, these architectural and engineering shifts confirm that the future of the AI race depends not only on software innovation, but is critically subject to the physical and manufacturing capabilities of semiconductor foundries, placing major tech companies before strategic challenges that require innovative solutions and extreme flexibility.
Frequently Asked Questions
Why did Nvidia decide to redesign its upcoming AI chip?
The company made this decision as a result of its failure to secure sufficient production capacity relying on the latest advanced manufacturing processes (1.6nm node) at TSMC’s factories, forcing it to shift some components to older manufacturing nodes.
When are these new chips expected to be launched in the markets?
The roadmap reviewed by the company indicates that the new platform is targeted for an official launch in 2028, while shipments to customers may continue through 2029 and 2030.
Is there an intention to use other companies in the manufacturing process to reduce reliance on Taiwan?
Yes, analysis and reports in the tech sector indicate that the company may resort to diversifying its supply chains, and is seriously studying the possibility of utilizing Intel to manufacture less complex components like input/output dies.
