Suppose that deals like the Nvidia/Nokia one aren’t what some on Wall Street think they are, cynical mechanisms to boost the companies in the near term. What could they do to actually build real AI benefits? Is there a real, tangible, industry-altering value that could be achieved here, or should we just go back to cynicism? Can Nvidia actually promote something helpful to AI in telecom missions? Can Nokia help them?
You have to start an exploration of this question with an exploration of what “industry-altering value” would mean. Information technology has had, since its real commercial inception 70 years ago, three waves of industry-shaking. The first was in 1964 with the advent of the first real “mainframe” computer with practical multi-tasking, mass storage, and “middleware”, including data networking. The second, a decade later, was the minicomputer, and the next a decade or so after that was the personal computer. I think it’s pretty clear that what all of these waves had in common was a closer approach of IT to work and lives.
How do we bring AI close to us, or use AI to bring IT close to us? It’s in a way a battle between the Internet model of projecting things via the network (which gave us the cloud), and the smartphone model and the historical succession of waves, which puts power directly into our hands. I think the first rule we could define here is that we can’t demand a choice, we have to support both to make AI truly transformational. Not only that, only some significant new relationship between people and technology is going to create anything telcos can monetize, and without that Nvidia and Nokia are wasting their time.
The waves show that we have to put AI in everyone’s hands, and if we expect them to live and work in an AI-dependent way, the AI has to be “Always on” and totally functional. To me, that means that everyone needs not just one AI hosting point, but multiple ones. AI needs a hosting fabric that goes from your wrist or finger, through your phone, onward through multiple hops to eventually a massive hyperscaler data center. It has to follow that path at the pace of life and work, and by doing so create a demand for life-and-work-sustaining reliability and latency.
What do we need to make this real? We need an “AI platform agent” that can see the entire fabric of AI hosting points, and can deploy a function to the place where its features indicate it’s most effective. This means that there has to be an “association of AI” (likely we’d end up with several, but that’s not optimum for obvious reasons) that defines standards for coding AI elements to reference their needs, and AI hosting points to reference their capabilities. For the latter, you’d likely need to have the ability to discover things like latency and even to test it regularly so you could rebalance things when needed. You’d also need to have “middleware” or “firmware” in hosting points so you could consistently respond to the matching of needs to capabilities, load AI elements, and interface with local devices and users. And, of course, you’d need security to ensure that this distributed mesh of AI wasn’t hacked.
I don’t think there’s a telco in the world, or a telco planner or executive, who believes something like this could be set up by a telco, or a consortium of them. I do think that any telco would likely be trusted to sustain it if they could acquire the pieces of technology involved. They could, if vendors would drive the development, and the key vendor in the process is (you probably guessed it) Nvidia.
Chip vendors play the critical role in this extravagant AI future for two reasons. First, AI chips run AI elements, obviously. Second, anything that’s going to be placed close to the user (in their hand, on their wrist) is going to rely on chip implementation or it will be the size of a suitcase, hardly the sort of thing you could make a partner in your work of life. Thus, the critical end of the chain of AI will have to reside in a chip, or be firmware linked to it. This means that you need APIs in some form to sell chips, which of course means that Nvidia has long API experience. You also need APIs to offer “facilitating services” as a telco, and that’s what I believe, and many others believe, is essential for telcos to bridge themselves into the future.
Nvidia also has experience in what I’ll call “OandO” meaning, organization and orchestration. When your service, application, device, or whatever is supposed to help someone work or live, the amount of aid you can deliver depends on your ability to understand how that person is working or living, meaning you need a view of and model of the real world that you can use to interpret questions and assess responses. A “digital twin”. Nvidia and Nokia are both actually already active in the digital-twin space, and a digital twin model would not only allow the functionality of the AI mesh to be linked to the features projected to the users, but also allow that functionality to be composed across hosting points. You can also define templates that represent the AI element composition of features, and that represent the feature composition of services.
The challenge here for both our “N” companies is that facilitation is inherently disintermediating. You’re helping someone else build on top of you, which pushes you down the value chain and reduces the net gain you could expect from a technology. In order to justify that, you have to facilitate something that builds a lot of value, that leaves plenty to trickle down from the retail level to those lower on the value chain. That means just hosting a digital twin isn’t enough, you have to think of hosting a broad, real-world, life-and-work enhancement model, one distributed widely, one that because it’s pervasive and distributed, generates traffic and requires a high level of quality to promote what’s necessarily a high level of human dependence on its operation.
Something like this isn’t only good for AI-based features and services, it’s good for AI enhancement to network operations, even for operations automation that doesn’t involve AI at all. How many network-related tasks, cloud-related tasks, go awry because they involve a sequence of inter-related steps, a single one of which can leave everything in a bad state if it’s done wrong, or even just in the wrong order? Imagine being able to ask AI to make a change, have it run the request against a digital twin to formulate the needed steps, and then even lay them out for review before they’re made. Or imagine every step being done first on the digital twin model, the result analyzed and accepted, before being applied so the real infrastructure.
For the telcos, the best part of this might be that not only does it mesh the cost-management advantage of AI with the service-revenue advantage, you can present this whole framework as “facilitating services” through a set of APIs. That validates the API as a telco revenue opportunity in the best possible way, by linking it to both the lower-cost and higher-revenue side of the profit equation.
Why don’t we hear about this sort of thing, from Nvidia and others? To me, this sort of story would be exciting, but perhaps I need to cultivate a new set of vices if that’s what excites me. Given the industry’s seemingly endless appetite for hype, fueled by an endless appetite for clicks, it could be that this sort of thing is just not what most are interested in, whether it’s essential for the future of telecom or not. I have to admit that I don’t get this sort of discussion spontaneously from the 88 telcos I chat with, though a dozen or so have expressed interest when I’ve raised it.
Why don’t we hear it from Nokia, or more significantly, why didn’t we hear it even before this deal? Nokia has actually been pushing industrial automation and digital twins, as I’ve already noted. Ericsson, Nokia’s first-world vendor rival, has an API program built around its Vonage acquisition, and even though just making telco features available through APIs is even more disintermediating than offering the complete services whose features those APIs expose, Nokia needs to do something effective in response. This could be it.
Nvidia is uniquely positioned to do this, to drive this, for another reason, which is the “AI bubble” story that’s growing on Wall Street. On Tuesday, SoftBank sold its whole Nvidia stake, and hedge funds saw blood in the water, even those who hadn’t scented blood already in AI valuation. Nvidia needs to stem this tide, and nowhere would be better than in their deal with Nokia aimed at the telecom space. OK, Nvidia, here’s your chance. Nokia is perhaps the best-positioned of the network vendors to do something like this, and Nvidia is perhaps the ideal partner. OK, Nokia, here’s you chance as well.
