Serving technology enthusiasts for over 25 Years.
TechSpot is a source of trusted tech advice and analysis.
Big picture: One of the biggest challenges in analyzing an ever-growing company like Nvidia, is to make sense of the many businesses it is involved in, its numerous products, and the strategy it’s following. The keynote speech of CEO Jensen Huang was followed by the company’s annual This year, the task of presenting at GTC Conference was especially daunting. Huang, as usual, covered a wide range of topics in a long presentation, and left many people scratching their head.
But, in an enlightening session with industry analysts a couple of days later, Huang shared a number of insights that made the reasoning behind all the product and partnership announcements he had covered crystal clear. He said that Nvidia was now an AI infrastructure company, creating a platform that cloud computing providers, technology vendors, and enterprise IT departments could use to create AI-powered applications.
It’s clear that this is a far cry from Nvidia’s role as a supplier of graphics chips for PC games, or its efforts to drive the creation machine learning algorithms. It unifies a number of seemingly disparate announcements and gives a clear indication as to where the company is headed.
Nvidia has moved beyond its semiconductor design house roots and reputation to become a critical infrastructure enabler in the future world of AI powered capabilities. Or, as Huang described it: “intelligence manufacturer.”
Huang discussed Nvidia’s efforts to enable the efficient generation of tokens, which are linked to intelligence, that organizations can leverage to generate future revenue. He described the initiatives as building a factory of AI, which is relevant to a wide range of industries.
Despite being ambitious, the signs are increasingly clear of an emerging information-driven economic system and the benefits AI can bring to traditional manufacturing. We are moving into a new era of economics, from businesses built around AI services like ChatGPT to robotic manufacturing and distribution.
In the context of this, Huang elaborated on how Nvidia’s latest offerings enable faster and more efficient token generation. He began by addressing AI inference. This is often considered simpler than AI training processes, which initially brought Nvidia to prominence. Huang, however, argued that inference will require 100 times more computing power than current methods of inference. This is especially true when using new chain-of thought reasoning models, such as DeepSeek’s R1 or OpenAI’s O1. There’s no reason to worry that more efficient language model will reduce the need for computing infrastructure. We are still in the early stages of AI Factory infrastructure development.
One of Huang’s most important yet least understood announcements was a new software tool called Nvidia Dynamo,is designed to enhance the inference for advanced models. Dynamo is an upgraded version Nvidia’s Triton Inference server software that dynamically allocates GPU resource for various inference phases, such as prefilling and decoding, each with different computing requirements. It also creates dynamic caches to manage data efficiently across memory types. Dynamo intelligently manages data and resources necessary for token creation in AI factory environments. Nvidia has named Dynamo “OS of AI factories.” In practice, Dynamo allows organizations to handle up 30 times more inference request with the same hardware resources.
Ofcourse, it wouldn’t have been GTC without Nvidia’s chip and hardware announcements. And there were many this time around. Huang presented a roadmap of future GPUs. This included an update to the Blackwell series, called Blackwell Ultra series (GB300 series), which offers enhanced onboard HBM Memory for improved performance.
Huang also revealed the new Vera Rubin Architecture, which features a new Arm CPU named Vera and a GPU of the next generation named Rubin. Each GPU has significantly more cores and advanced abilities. Huang hinted at a generation beyond, named after mathematician Richard Feynman. He projected Nvidia’s roadmap to 2028 and beyond.
During a Q&A session that followed, Huang explained how revealing future products in advance was crucial for ecosystem partners to prepare for upcoming technology shifts.
Huang emphasized a number of partnerships announced at the GTC this year. The presence of other tech companies showed their willingness to participate in the growing ecosystem. Huang explained on the compute side that to fully maximize AI infrastructure, advancements were needed in all traditional computing areas, including storage and networking.
Nvidia announced a new silicon photonics-based technology for optical networking among GPU-accelerated server racks, and discussed a partnership between Cisco. Cisco’s partnership with Nvidia allows Cisco silicon to be used in routers and switch designed for integrating GPU accelerated AI factories into enterprise settings, along with a common software management layer.
Nvidia worked with leading hardware companies and data platform providers to ensure their solutions could benefit from GPU acceleration. This increased Nvidia’s influence in the market.
Finally, Huang, building on his diversification strategy, introduced more work the company is doing in autonomous vehicles (notably a GM deal) and robotics. Both of these he described as being part of the next major stage in AI development, physical AI.
Nvidia is aware that as an ecosystem and infrastructure provider, they can both directly and indirectly benefit as the tide of AI computing rises. This will also increase their direct competition
Nvidia provides components to automakers and robotics platforms since several years. What’s new is that these devices are now tied to AI infrastructure, which can be used to train models that will be installed into the devices and provide the real-time data needed to operate them. This tie-back to infrastructure may seem like a modest step forward, but in the context of the company’s overall AI infrastructure, it makes more sense. It also helps tie together the company’s various initiatives into one cohesive whole.
It’s not easy to make sense of the many elements that Huang and Nvidia announced at this year’s GTC, especially because of their firehose-like nature and the much wider reach of the company. When the pieces come together, Nvidia’s strategy becomes clear. The company is taking on an even larger role than before and is well positioned to achieve its ambitious goals.
Nvidia is aware that as an ecosystem and infrastructure provider, they can both directly and indirectly benefit from the rise of AI computing, even though their direct competition will increase. It’s a smart strategy that could lead to greater growth in the future.
Bob O’Donnell, founder and chief analyst at TECHnalysis Research, LLC, a technology consultancy firm, provides strategic consulting services and market research to the technology industry as well as the professional financial community. You can follow him @bobodtech (19659028]
