Preview of Nvidia GTC Conference: Integrating Groq technology for a major push into inference chips, Samsung manufacturing for the first time, OpenAI may be among the first customers

```

On Monday local time, the highly anticipated annual NVIDIA GTC Developer Conference is about to open in San Jose, California, USA. CEO Jensen Huang’s keynote speech has always been viewed as a key indicator for the AI industry.

For investors, the most noteworthy aspects of this year’s conference are NVIDIA’s strategic shift of focus from training to inference, as well as adjustments to its supply chain layout.

Based on information from The Information, The Deep View, and other media, this GTC may reveal three key signals:

First, NVIDIA may leverage the integration of Groq technology to make a major entry into the AI inference market;

Second, in the short term, chip manufacturing may shift partially from TSMC to Samsung;

Third, the ecosystem for robotics physical AI and open-source models is expected to further expand.

Targeting the inference market, Groq chips as key leverage

The AI industry is currently gradually shifting from being “training-first” towards “inference-driven.” In the training field, NVIDIA has established a solid technology and ecosystem advantage with its GPUs; but in the inference market, competitors such as Cerebras are gaining market share with faster and lower-cost solutions.

Against this backdrop, the market is highly focused on NVIDIA’s counterstrategy. According to The Information, Jensen Huang is expected to announce at the conference a new chip system that integrates NVIDIA and Groq technologies. Behind this lies NVIDIA’s acquisition of Groq technology licensing for about $20 billion at the end of last year.

Chips developed by Groq are called LPUs (Language Processing Units), specifically optimized for inference workloads. This will also mark the first time that NVIDIA directly integrates another company’s AI processor into its server rack systems.

Supply chain restructuring and first major clients confirmed

The manufacturing and commercialization of the new system is also a focus for the capital markets.

According to The Information, the Groq LPU is expected to be manufactured by Samsung in the second half of this year. This arrangement is significant, as it could mark the first time NVIDIA’s server chips are made by a foundry other than TSMC, breaking its long-term reliance on a single supplier.

However, according to the aforementioned media citing insiders, this change may mainly be stage-based. Since the next-generation LPU will require closer integration with NVIDIA’s coming AI chips, subsequent production might still return to TSMC.

On the demand side, NVIDIA is expected to announce that OpenAI will be one of the system’s first major customers. This chip system may be used to power AI agents executing tasks such as coding.

Underlying architectural changes and future technology roadmap

For investors focused on chip technology, the architectural design of the NVIDIA-Groq system also reveals potential integration challenges and opportunities.

According to The Information, the new rack structure will be distinctly different from the current system: each rack will feature 256 Groq chips. At the same time, Intel processors will be responsible for communications management in the system. This design also suggests that NVIDIA’s existing architecture has yet to fully integrate with the LPU.

However, NVIDIA clearly has longer-term plans. According to The Information, two insiders involved in development revealed that the company is internally exploring deeper integration of the LPU into its future product roadmap. One proposal is to merge the Groq processor and the next-generation Feynman GPU (the successor to the Rubin architecture) into a single chip to enhance performance and lower overall costs.

Physical AI and the expansion of the open-source model ecosystem

Beyond computational infrastructure, the AI application ecosystem is also a major highlight of the GTC conference.

As the “AI Moore’s Law” continues to advance—with computational efficiency roughly doubling every four months—NVIDIA’s plans in physical AI and robotics are receiving keen attention. Especially amid the accelerated development of China’s humanoid robotics industry, whether NVIDIA and its partners can provide more cost-effective solutions in scenarios such as autonomous driving will be in the spotlight.

At the same time, NVIDIA is rapidly advancing in the open-source model field. Previously, the company released the 120-billion-parameter Nemotron 3 Super model, and announced plans to launch the Nemotron 4 Ultra with four times the parameter scale. The improvement in model capability is expected to further reduce enterprise AI inference costs and improve overall ROI.

The signals sent by this year’s GTC may largely influence the AI industry landscape for 2026.

Risk Warning and DisclaimerThe market involves risk; investment needs caution. This article does not constitute individual investment advice, nor does it factor in the unique investment objectives, financial situations, or needs of any single user. Users should consider whether any opinions, viewpoints, or conclusions in this article suit their specific situations. Invest accordingly at your own risk. ```