GTC Paris: Sovereign factories vs. distributed visions vs. tomorrow’s RAN – What's missing?

Author: Peter Jarich, Head of GSMA Intelligence

NVIDIA ensured that its inaugural GTC Paris was accompanied by a metric ton of news. Product launches. Partner solutions. Customer success stories. If you pay attention to the AI and/or broader tech spaces even marginally, I probably don’t need to provide a highlights reel. Of course, the vibe on the ground is always different from what makes it into the news. And, since I was lucky enough to be there in person, I wanted to put my thoughts together – including the good and the bad. 

Having spent some time with the GSMA Intelligence team (in partnership with NVIDIA and Dell Technologies) digging into the technology and market dynamics of distributed inferencing, I was especially interested in the Edge AI story at GTC Paris. I covered my thoughts in a series of videos from the show floor. If you’ve not seen them yet, go check them out at the GSMA Intelligence Linkedin page. 

Of course, this was really just one small part of the GTC Paris story. 

• Sovereign AI: With fresh commitments in the Italy, UK, France, Germany and beyond, Sovereign AI was the news that stole the show. The message here – both implicit and explicit – was that countries need to own their AI destinies. But where we’ve seen this framed in terms of ensuring that sensitive data does not leave national borders, the vision of Sovereign AI as a way to drive citizen services, foster AI ecosystems and support innovation was refreshing. So too was the tacit acknowledgement that this is essentially telco’s opportunity to lose given their role in connectivity, regulatory fluency, and the successes so far.
• Scale Up vs. Scale Out: While continuing to push the bounds of AI processing (including compute and networking innovations), NVIDIA was quick to flag support for scaled down form factors in line with diverse customer needs: PCs, workstations, 1U servers…and more. The show floor was full of these, focused on real world enterprise and niche vertical applications. But the core message from NVIDIA and its CEO was generally that “bigger is better”; an overarching focus on AI Factories in the opening keynote set a tone that was later confirmed in a press and analyst session where Huang extolled the virtue of ”big brains.” To be sure, for most use cases this may well make sense. Too much bias in promoting the high-end at the expense of the low-end could put NVIDIA at odds with partners who have a clear interest in driving AI distribution.
• NVIDIA as AI AND Compute Champion: Any trade show will be largely about self-promo, and that was no different for GTC Paris. But, beyond specific product launches and new market successes, there was a clear message from NVIDIA. AI is just part of a broader “accelerated compute” vision that has always guided the company. This is key to explaining how it used to be a graphics, then a crypto mining darling and now the belle of the AI ball. More than just a company, though, they took pains to remind everyone that they represent an ecosystem, building products from hundreds of suppliers and integrated into the solutions of hundreds of partners. In the interest of growing its competitive moat, this ecosystem role is key. So, too, is making sure the market understands that. 

With so many supporters and throngs of fans clamouring for the CEO's picture, it was hard not to come away in awe of what NVIDIA has done and the opportunity it's sitting on. Huang's engaging, semi-folksy manner mixed with a good dose of proud boasting only served to bolster his image…and that of his company. 

So, what's not to love? 

To be sure, constant, self-serving AI promotion began to wear thin after a while; every recommendation to embrace AI’s transformative power or reminder of AI’s escalating trajectory ultimately lead to more business for NVIDIA. But that doesn’t mean the company isn’t right and it’s exactly the message it should be telling from a shareholder perspective. That still leaves a handful of bigger misses and ignored realities. 

• Sovereign Economics: In his fireside chat with French President Macron and Mistral AI’s Arthur Mensch, Huang’s concise explanation of Sovereign AI’s importance (roughly, the need to control your intelligence) garnered a theatre-full of vigorous nods. Everyone implicitly understood and agreed. I expect the reaction would be similar if you replaced “intelligence” with “national defence” or “food supply” or “industrial supply chain.” Yet, for economic reasons, countries routinely outsource these. This raises a number of questions. Is AI that different? Does every country actually need its own AI assets, built solely for its own use? Can they afford to? NVIDIA execs claim that there are 70 countries with Sovereign AI plans underway. That’s undeniably impressive but comes with the risk of deployments that aren’t fully utilized or as cost-effective as non-Sovereign or semi-Sovereign AI strategies. 

• Costs and RoI: On the topic of returns, it was hard to ignore the degree to which AI rollout costs were ignored in much of the conversation at GTC Paris. Nobody was denying the massive costs of new, high-end AI compute solutions but the recommendation to deploy the latest and greatest (and most expensive) hardware seemed powered by a “build it and they will come” mentality. To be fair, NVIDIA is putting in an immense effort to ensure that deployed AI capacity gets used; via promotion, and examples, and toolkits, and resources, it’s smartly driving demand alongside supply. But once we get past a phase where FOMO drives many investment decisions, returns will come into focus and costs will come into the spotlight. 

• Regulation: The topic of regulation surfaced in a few specific instances at GTC Paris. Telcos, for example, were flagged as a good fit for Sovereign AI since they know how to operate in a regulated business. Separately, Huang espoused a view that AI applications should be regulated, not the technologies behind them. Both points seem logical enough. Assuming that telcos can just navigate the regulatory waters of supporting national AI agendas vs. get bogged down in a regulatory swamp, however, seems more hopeful than realistic, potentially setting them up to invest based on an overly-optimistic view of the opportunity. Similarly, on the topic of regulating AI’s applications, common concerns like algorithmic transparency were mostly dismissed as foolish or just wrong – a welcome message for AI fanboys, but not great when engaging regulators

• 6G and Future RA: Long before it took a leading role in the formation of the AI RAN Alliance, NVIDIA’s interest in innovating the RAN industry was clear. You can check out a Youtube video of the company taking the stage at MWC Los Angeles in 2019 to talk up its vRAN work with Ericsson, among other things. GTC Paris news, in turn, included a reminder of its 6G Developer platform and the 200+ organizations across 30+ European countries leveraging it. There was also the launch of NVIDIA’s AI Blueprint for telco network configuration promising RAN parameter optimization. This all came hot on the heels of a new, scaled-down RAN platform launch (the ARC-Compact) at Computex in late May. Did all of this get a major nod in the opening keynote? Nope. Was the new RAN platform on display in the company’s exhibits? Nope. Show messaging obviously needs to be tailored to the audience but GTC Paris highlighted the risk that a focus on AI Factories and Sovereign AI will overshadow RAN support as a strategic priority for NVIDIA. That’s not great for any operators who want to engage with NVIDIA on RAN and suggests we’ll need to hear more on the front from NVIDIA given the potential RAN opportunity and the investments being made.