GTC 2026 Day 2 – Cementing Dominance

Telecom and enterprise.

• Following on from Mobile World Congress, GTC is continuing to push the idea of AI moving to both the edge and the near-edge, which in many cases could be the mobile basestation.
• This is both through running the mobile basestation on an Nvidia Computer (like Nokia) or simply using Nvidia GPUs for AI workloads at the basestation (like Ericsson).
• This is being taken a step further at GTC with AI Grids which take existing infrastructure and add AI capabilities so that workloads can be run where it makes most sense.
• While MWC saw the Europeans sign up, at GTC, Comcast, Spectrum, Akamai, T-Mobile and ATT also signed up to put Nvidia AI in their infrastructure.
• Also of interest was Nvidia’s support of visionOS for the Apple Vision Pro, which takes a failed consumer device and pushes it into the enterprise, where I suspect it might see more success.
• Despite this, I still see the Apple Vision Pro as an insurance policy against the obsolescence of the iPhone, but if it can wash its face in the enterprise at the same time, then this will help pay back some of the massive cost that went into developing it.

China – Stop, start, stop start

• Production of the cancelled, then restarted, then cancelled again H200 has been restarted with Jensen stating at the press conference that he has both the licenses from the US Department of Commerce and orders from Chinese customers to begin production again.
• However, what Jensen does not have is permission from the Chinese state to sell his products to customers in mainland China.
• I have long been of the opinion that China’s refusal to grant import permits is not a negotiating tactic but a deliberate policy to wean China off its dependence on foreign technology.
• Furthermore, the only statement it has made recently is that it might allow some H200s for special purposes such as research and development.
• This sounds like imports for the purposes of learning how to make high-quality AI semiconductors to replace imports, making these sorts of imports self-defeating.
• Consequently, I continue to think that the Chinese market is a lost cause for Nvidia and all foreign AI semiconductor vendors.
• However, Chinese alternatives are so uncompetitive that whatever share they may lose in China, they will more than make up elsewhere.  

Retail – OpenAI and Google onboarded.

• This GTC has witnessed many launches of reference platforms, which are blueprints that Nvidia customers can use and then customise to create their own version of the product without having to start from scratch.
• Nvidia’s DSX AI Factory is a reference design for building an Nvidia data centre, while the Nvidia Blueprint for retail agentic commerce does the same to help retailers connect their ecommerce platforms to AI agents.
• OpenAI and Gemini protocols are both supported, meaning that a user can use ChatGPT or Gemini to research, decide and execute a purchase.
• The reference platform also allows retailers to control and run their own payment security, promotions and recommendations.
• This is significant because the two leading consumer AI agents are already on board, which gives this platform a pretty good chance of being selected by retailers looking for an easy way to enable AI agents to use their services.
• This, in turn, will reinforce Nvidia’s position as the main supplier of AI systems, which, at the end of the day, is what GTC is all about.

Take Home Message

• Nvidia’s approach to GTC 2026 is all about making the case for how deploying Nvidia provides the user with better inference and, critically, with the best economics.
• The keynote claim that Vera Rubin with Groq will allow revenues per GW of AUI capacity to rise 5x to $50bn will be a game changer if it proves to be real, but history is not on Nvidia’s side here.
• Likewise, day 2 of GTC is all about demonstrating how Nvidia makes it easy to deploy AI if one uses its products and platforms.
• It’s one-generation-or-better lead in AI silicon is a key factor in making the economic argument for using Nvidia, even though one has to pay a significantly higher price to do so.
• It all looks great on stage and on PowerPoint, but it is how much of these benefits can be delivered that will determine whether competitors can find a weakness in Nvidia’s armour.
• The economics of AI compute both in the data centre and elsewhere is the key factor to examine to determine how well Nvidia and its partners manage to turn these promises into reality.