Computex Day 1 – Nvidia Opens.

Nvidia is not standing still.

• Nvidia Founder Jensen Huang opened Computex 2025 with a confident session that launched new products aimed at the enterprise, but also admitted that in the AI factories, not everything is going to come from Nvidia opening the door for other silicon offerings.
• I see this as a wise move as the AI data centre market is still very concentrated among a few players, all of whom have or are designing their own AI chips and wish to customise the infrastructure for their own requirements.
• Nvidia made a number of announcements, many of which we saw at CES or GTC, but two new things stood out in particular.

• • First, NVIDIA NVLink Fusion: which is by far the most significant announcement that was made.
  • NVLink Fusion allows a data centre builder to switch out the Nvidia in-house CPU and replace it with whatever they wish.
  • NVlink Fusion is a piece of silicon that sits next to a custom CPU that allows it to manage the Nvidia GPUs, the memory and the inter-chip communications just as the Nvidia Grace CPU does today.
  • As far as the official documentation is concerned, NVLink Fusion only allows the CPU to be switched out, although Jensen seemed to imply in his commentary that more would be possible.
  • I think this was an incorrect implication by Jensen as it makes zero sense from Nvidia’s business model perspective to allow its own GPUs to be switched out, as that is where it makes its real money.
  • Interestingly, both MediaTek and Qualcomm were called out in the keynote and have quotes in the press release, although neither of them has announced a datacentre CPU product.
  • Combine this with the announcements made with Humain last week (see here), and one can only conclude that both MediaTek and Qualcomm are working on a data centre CPU specifically for AI training and inference.
  • This is obviously the result of customer feedback, and seeing as this flexibility is unlikely to hurt Nvidia’s GPU market share or its margins, it looks like a good move to make on its part.
  • Second, NVIDIA RTX PRO: which is a new series of enterprise servers that allows an enterprise to run all of its legacy systems as well as add AI and specifically agentic AI to its systems.
  • The RTX PRO is built with Blackwell server edition GPUs as well as the right software that allows an enterprise to shift from CPU-based servers to GPU.
  • This is yet another threat for Intel, whose data centre business is already under siege and where it will now have to fight yet another defensive battle.
  • Nvidia has been quick to rope in the data centre ecosystem and the press release is replete with the who’s who of IT builders who are jumping on Nvidia’s latest bandwagon.
  • How much success Nvidia will see with this is unclear at this stage, but if it can do a good job of migrating the legacy to accelerated GPUs cost-effectively, then this is a good proposition for any enterprise looking to start using AI in its systems.
  • For Nvidia, this opens up another large revenue opportunity and with faith in Intel and x86 seriously flagging, I suspect that there will be a lot of interest in this new product line.
• This was a more nuanced keynote than usual, but moving to accommodate the big customers without ceding market share and going for Intel’s business while Intel is in poor form to defend it, look like good moves to me.
• Neither of these is going to have immediate financial benefit, but in the long run, they will serve to support Nvidia’s market share and augment revenue and profit growth.
• Nvidia is doing a good job of not resting on its laurels but using its current dominant position to create support and engagement from the ecosystem with new product categories.
• Nvidia’s shares are trading on 30.7x 2025 and 24x 2026 PER, which is not unreasonable as long as Nvidia can keep better than 25% growth going.
• These product launches and this Computex are clearly aimed at doing precisely that.