Nvidia Corp (NASDAQ: NVDA) is accelerating development of its next-generation Feynman AI platform, pushing Taiwan Semiconductor Manufacturing Co. (NYSE: TSM) to fast-track advanced chipmaking and packaging capacity ahead of an expected 2028 launch. The shift in resources toward Feynman occurs as Nvidia’s Vera Rubin platform enters mass production.

Architectural Evolution

Feynman is expected to feature 3D chiplets, TSMC’s System on Integrated Chips (SoIC) technology, high-bandwidth memory (HBM), and co-packaged optics (CPO). This represents a major architectural evolution from the Rubin platform. Industry sources expect Feynman to use TSMC’s upgraded 2nm-class A16 process.

The platform could push NVLink bandwidth beyond 1,000 terabytes per second, or 1 petabyte per second, as Nvidia builds increasingly large AI clusters.

TSMC Capacity Expansion

The accelerated roadmap is putting additional pressure on TSMC to expand advanced packaging capacity. TSMC is accelerating construction at its AP7 facility in Chiayi and at AP8 in the Southern Taiwan Science Park, while preparing capacity for SoIC, CoWoS-L, and next-generation CoPoS technologies.

TSMC’s SoIC technology enables chips to be stacked vertically, potentially improving bandwidth while reducing latency and power consumption. The report said TSMC could increase monthly SoIC capacity to approximately 50,000 wafers by the end of 2027, up significantly from earlier plans.

Revenue Forecasts

Nvidia is scheduled to report its second-quarter results on Aug. 26. Bank of America analyst Vivek Arya maintained Nvidia as a "top pick" with a $350 price target, representing 56.3% upside from the stock’s $223.96 price at the time.

BofA forecasts Nvidia’s quarterly revenue at $94 billion to $95 billion, about $3 billion to $4 billion above the company’s $91 billion guidance. The bank also expects third-quarter guidance of $107 billion to $108 billion, above the consensus estimate of roughly $104 billion.

What the Numbers Show

Bank of America’s revenue forecast implies a significant upside to Nvidia’s own guidance. The $3 billion to $4 billion gap between BofA’s Q2 forecast ($94 billion to $95 billion) and Nvidia’s guidance ($91 billion) suggests analysts anticipate stronger-than-expected demand or pricing power in the current quarter. Additionally, the Q3 forecast ($107 billion to $108 billion) exceeds consensus estimates ($104 billion) by approximately $3 billion to $4 billion, indicating sustained confidence in growth momentum beyond the immediate quarter.

Market Context

Nvidia has also begun working with major financial institutions on an AI infrastructure financing platform that targets more than $500 billion in third-party capital.

Nvidia closed at $225.30 on Thursday, up 0.54%, while Friday’s premarket shares were down 0.02% at $225.25. According to Benzinga Edge Rankings, Nvidia ranks in the 99th percentile for growth and maintains positive short-, medium- and long-term price trend ratings.

TSMC and Nvidia did not immediately respond to Benzinga’s request for comment.

Nvidia Corp (NASDAQ: NVDA) CEO Jensen Huang stated on Wednesday that the company's older A100 GPUs can remain economically useful for nearly a decade, pushing back against concerns that rapid AI advances could quickly render the hardware obsolete.

Huang said on X that the A100 fleet is "mission-capable from 2020 through 2029." He highlighted the longevity of Nvidia's data center GPUs, noting that the A100 debuted in 2020 as part of the Ampere generation. Huang attributed this durability to the company's CUDA software platform, which he said makes computing versatile and fungible, thereby driving utilization.

CoreWeave Deal Challenges Obsolescence Fears

Huang's comments followed a disclosure from AI cloud provider CoreWeave Inc (NASDAQ: CRWV). The company signed a contract to rent Nvidia A100 GPUs through 2029. CoreWeave CFO Nitin Agrawal described the deal price as attractive. This arrangement means the 2020-era chips could generate revenue nearly nine years after launch, despite several newer GPU generations entering the market.

This development challenges a key concern among AI infrastructure bears that accelerating chip improvements could cause older GPUs to lose economic value within two or three years. Older GPUs can remain useful by shifting from demanding AI workloads to inference and smaller models, according to a Business Insider report cited by Huang.

What the Numbers Show

Rental-market data supports the assessment of GPU longevity. Silicon Data, which tracks GPU rental prices, reported that A100 rates have remained resilient following a strong rebound in 2026. Silicon Data noted that based on residual fair value estimates, A100 stopped depreciating since late 2025 as rising rental rates stabilized cash flows. This suggests GPUs are increasingly viewed as financeable capital assets with stable income streams rather than rapidly depreciating hardware.

Capital Mobilization Partnerships

The implication of longer GPU utility aligns with Nvidia's strategy to promote its chips as revenue-generating assets. On Monday, Nvidia announced partnerships with Apollo Global Management (NYSE: APO), BlackRock Inc (NYSE: BLK), Blackstone Inc (NYSE: BX), Brookfield Asset Management (NYSE: BAM), Goldman Sachs (NYSE: GS) and KKR & Co Inc (NYSE: KKR). These partnerships aim to help mobilize more than $500 billion in capital over time.

Nvidia closed at $224.09, up 3.03% on Wednesday. Shares were down 0.09% to $223.90 in Thursday's premarket trading.