SemiAnalysis: The Ice and Fire of Nvidia's Rubin Platform

The semiconductor research institute SemiAnalysis has released two assessments, outlining both the opportunities and challenges in Nvidia's future—the "ice and fire" duality.

The latest forecast posted by SemiAnalysis on the X platform on June 30 shows that Nvidia's data center computing business revenue in the second half of the 2027 fiscal year will be about 20% higher than unanimous Wall Street expectations. The core basis of this optimistic view lies in the fact that the previous HBM4 memory supply issue, which limited large-scale shipments of the Rubin platform, has been resolved, and front-end wafer capacity is now fully reserved, clearing substantial obstacles to a performance spike in the latter half of the year.

However, earlier that same day, SemiAnalysis disclosed another bearish message: Nvidia's original 4-chip Rubin Ultra was canceled approximately three months after its GTC 2026 release, and the revised "Rubin Ultra" has been reduced to half its original size, with actual performance also halved.

On one hand, there is the optimistic upward revision of revenue after supply bottlenecks have been lifted; on the other, there is the pessimistic correction to the technical strategy following the shrinkage of the flagship product. These two diametrically opposed judgments from SemiAnalysis anchor Nvidia with two completely different narrative coordinates, respectively from the perspectives of performance fulfillment and technological moat.

HBM4 Bottleneck Resolved, Rubin Platform Set for Volume Release in Second Half of the Year

SemiAnalysis, through its Accelerator Model, makes its latest prediction: Nvidia will experience a large-scale ramp-up in the second half of this year.

The institute expects that, driven strongly by the Rubin platform, Nvidia's data center computing business revenue in the second half of the 2027 fiscal year will be about 20% higher than consensus market expectations. The previous HBM4 issue that once affected Rubin's progress has now been resolved, and front-end wafer supply has been preemptively stocked, meaning the previously delayed Rubin platform will enter a rapid growth phase.

SemiAnalysis particularly points out that its forecasting logic differs significantly from that of traditional sell-side analysts. Most Wall Street institutions tend to establish relatively conservative earnings forecasts to leave room for companies to "beat expectations" later; however, SemiAnalysis bases its conclusions more on frontline industry chain research, striving for greater accuracy to actual market dynamics.

Its Accelerator Model builds a cross-verification system across the entire chain, with data sources covering material suppliers, wafer manufacturing, key components, server integrators, and other supply chain links, as well as the procurement and deployment practices of hyperscale cloud service providers and cutting-edge AI labs, allowing for multi-dimensional validation of supply-demand relationships.

It is worth noting that this model not only tracks Nvidia, but also covers AI chip makers such as Broadcom, AMD, MediaTek, Marvell, and, along with the HBM Model, continually tracks the evolution of the AI computing power industry chain overall.

CUDA Moat Being Eroded, Rubin Ultra Cutback Reflects Rise of In-House ASICs

However, a previous comment by SemiAnalysis regarding Rubin Ultra has sparked widespread discussion in the market.

The institute stated that Nvidia's original plan for a 4-chip design for Rubin Ultra was adjusted about three months after its GTC release this year, with the new version being significantly reduced in scale compared to the original design due to the difficulty of advanced packaging manufacturing.

SemiAnalysis believes that what is more noteworthy is not simply the cutback of Rubin Ultra itself, but the changes in industry competition revealed by this event. The institute points out that over the past year, Nvidia's biggest competitive pressure is no longer just from traditional GPU manufacturers like AMD, but increasingly from hyperscale cloud providers and AI model companies adopting in-house developed ASICs—dedicated chip systems customized for training or inference scenarios.

For example, Anthropic has now developed a multi-platform computing architecture consisting of Google TPUs, Amazon Trainium, and Nvidia GPUs. Large volumes of Claude model training are run on the TPU platform, Claude Code inference is increasingly deployed on Trainium, and Nvidia GPUs are more often used for frontier research and general computing tasks. SemiAnalysis notes that a year ago, it was still unthinkable for TPUs and Trainium to grow to today's scale, but now the CUDA moat is slowly being eroded.

Risk Warning and DisclaimerThe market carries risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the particular investment goals, financial status, or needs of any individual user. Users should consider whether any opinions, perspectives, or conclusions in this article fit their specific circumstances. Investment decisions are at one's own risk.