Don't Just Focus on GPUs! A Look at the Five Key Highlights of NVIDIA GTC 2026

Author | Eric

The recent tensions in geopolitical situations have weighed on US technology stocks. As a result, this year's GTC conference has become particularly crucial. During times of heightened market sentiment, investors tend to look for landmark events that can reshape the future narrative of the industry. GTC 2026 fits perfectly into this expectation. $NVIDIA (NVDA.US)$ Typically, the next-generation platform will be unveiled on this stage, pointing out the next technical bottleneck and indicating the direction of the next wave of corporate capital expenditure.

Five Key Highlights of GTC 2026:

1. The Main Event Feynman Takes the Stage: Targeting Foundry and CPU

The most anticipated highlight of this conference is the unveiling of the Feynman platform. Rumor has it that this will represent a major leap beyond the Rubin architecture, with industry insiders even suggesting that its GPU might use $Taiwan Semiconductor (TSM.US)$ the A16 process technology for manufacturing.

The CPU aspect is equally noteworthy. NVIDIA and $Intel (INTC.US)$ previously announced their collaboration, with Intel set to manufacture customized x86 CPUs for NVIDIA. These chips will connect through NVLink technology, deeply integrating into NVIDIA’s AI infrastructure platform. If the GTC conference reveals more details, it will further boost Intel’s foundry narrative.

2. SRAM architecture inference chips may become an unexpected focus of speculation

NVIDIA poached key executives from Groq, one of the three major players in SRAM inference, at the end of last year and obtained relevant technology licenses. This move has given NVIDIA more confidence in promoting the SRAM inference architecture. Specifically, Groq's LPU integrates 230MB of on-chip SRAM as the primary weight storage, directly replacing the off-chip HBM bandwidth that traditional GPUs heavily rely on.

In terms of the foundry landscape, the situation is very clear. Groq's current generation of chips is manufactured using $GlobalFoundries (GFS.US)$ a 14nm process, while the next-generation LPU will be outsourced to Samsung for production. Meanwhile, Cerebras' WSE-3 is a wafer-scale chip built on $Taiwan Semiconductor (TSM.US)$ a 5nm process, with chiplets enabling an on-chip SRAM total capacity of up to 44GB. Notably, these embedded SRAMs are all manufactured within their corresponding logic processes.

The latest developments from OpenAI further reinforce this trend. OpenAI released a research preview version of GPT-5.3 Codex Spark on the Cerebras platform, specifically designed for low-latency programming. In addition, OpenAI announced a multi-year procurement agreement for wafer-scale systems totaling 750MW. These tangible actions have given 'non-GPU inference' real investment value.

Finally, chip architectures that heavily rely on SRAM significantly increase the complexity of physical design and verification, further solidifying the position of EDA tools. As part of the deepened cooperation between both parties, NVIDIA invested $2 billion in $Synopsys (SNPS.US)$ . If market funds start speculating on the grand narrative of custom inference chips, Synopsys will undoubtedly become a pure-play toolchain target.

Optical communication and CPO have evolved into the main theme of capital allocation

At the beginning of the month, NVIDIA separately injected $2 billion into $Coherent (COHR.US)$and$Lumentum (LITE.US)$ and signed multi-year procurement agreements. This move aims to secure silicon photonics capacity in advance and fully accelerate the iteration of optical communication technologies in AI infrastructure. This action alone sends a strong signal that optical communication will continue to be a key highlight at this year’s GTC conference.

Moreover, the market may currently be significantly underestimating the potential of microLED-based optical communication technology. NVIDIA, which recently released its earnings report, $Credo Technology (CRDO.US)$ regards microLED as an effective path to substantially reduce unit power consumption in near-packaging optical solutions, claiming thatthe Total Addressable Market (TAM) for ALC will be twice the current AEC.According to TrendForce's forecast, in specific data center application scenarios, microLED CPO technology can cut total interconnect power consumption to about 5% of traditional copper cable solutions, potentially emerging as a standout at this year's GTC conference.

4. The memory sector is no longer just about HBM, with the core observation pool expanding significantly.

The Rubin architecture is driving the hardware stack towards more highly customized inference configurations. Meanwhile, NVIDIA has also disclosed detailed technical specifications of Rubin CPX, positioning it as a context-phase accelerator within rack-level systems. This architectural design keeps memory bandwidth and capacity as the crucial determinants of performance.

From the perspective of short-term trading dynamics, the most direct memory-related stocks in the US market remain $Micron Technology (MU.US)$ . Particularly as the entire industry accelerates toward the HBM4 era, major manufacturers are increasingly tightly coupled with NVIDIA through co-design efforts in developing new memory modules.

A new highlight in the sector is HBF technology. Within the architecture of inference infrastructure, it is precisely positioned as a new storage layer between HBM and traditional SSDs. Currently, SK Hynix and $SanDisk (SNDK.US)$ Synopsys have initiated the OCP working group, dedicated to advancing the technical standardization of HBF. This move breaks the market's stereotypical view of the cyclical nature of traditional SSDs and introduces a completely new investment narrative for the memory sector.

5. 800V high-voltage DC: A seemingly ordinary yet pivotal catalyst for unlocking real value

As rack power consumption enters the megawatt era, the power supply system is becoming the biggest bottleneck constraining the development of AI computing power. Last year, NVIDIA unveiled its blueprint for an 800V HVDC high-voltage DC architecture and is actively driving the construction of a vast ecosystem around this architecture. Various research reports focused on the data center supply chain have also detailed lists of relevant partners, covering semiconductors, power components, and infrastructure across three major fields.

In the power and infrastructure sectors, relevant targets span the entire upstream and downstream of the industry chain:

- First in line are power semiconductor manufacturers (e.g., $INFINEON TECHNOLOG (IFNNY.US)$ 、 $STMicroelectronics (STM.US)$ 、 $Texas Instruments (TXN.US)$ 、 $Analog Devices (ADI.US)$ 、$Monolithic Power Systems (MPWR.US)$ 、 $ON Semiconductor (ON.US)$ 、 $Navitas Semiconductor (NVTS.US)$ 、 $Alpha & Omega Semiconductor (AOSL.US)$ 、 $Power Integrations (POWI.US)$ ；

- Next are system and data center power solution providers (e.g., $Flex Ltd (FLEX.US)$ 、 $Eaton (ETN.US)$ 、 $Vertiv Holdings (VRT.US)$ , as well as those with exposure to the grid side $GE Vernova (GEV.US)$ ）；

- Finally, high-density server power supply and PCB manufacturers (e.g., $Vicor (VICR.US)$ 、 $TTM Technologies (TTMI.US)$ ). Once improvements in power density and the acceleration of rack redesign progress, these companies will undoubtedly become explosive second-tier beneficiaries.

Summary

The significance of the GTC conference has long surpassed the scope of a mere product launch event. It is simultaneously a clear directive from NVIDIA to the capital markets, directly identifying the next technological bottleneck in the semiconductor industry.

Looking ahead to 2026, several clear indicators have emerged: The Feynman computing platform, bold attempts at SRAM architecture inference chips, the scaled application of optical communication technology, the emergence of a new memory hierarchy, and the comprehensive transition of power architecture to 800V high-voltage DC. Once market funds choose to focus on this technological event, the spillover effects generated by the aforementioned supply chain segments will present investors with excellent opportunities to capitalize on.

769K Views