You’ve got to admire somebody who’s willing to say that AI is “underhyped”, which is what Fierce Network’s story on the NVIDIA GTC Conference says is the view of CEO Jensen Huang. Is it even possible to have something underhyped these days? I wonder, but the comment gives us a reason to look at the role of AI in networking, both at the enterprise and operator levels. There may even be some implications we can draw regarding AI overall, hidden in the details.
To get an important point out of the way, what company CEOs say at conferences is propaganda, and NVIDIA is under pressure on Wall Street right now. They need to look confident about AI opportunities because they need the Street to see their future market as growing. Thus, you have to assume that no public forum is going to miss a chance to convey an optimistic vision. The question here is whether AI is really under- or over-hyped, meaning whether the future holds more and more of it, or if it’s a flash in the media pan. Huang can’t make an opportunity real, only realize one that’s already waiting in the wings. Reality here, as usual, depends on real buyers.
I chat with 88 operators, and of that group I’ve had AI comments recently from 67. Every one indicated a belief that AI had value in their business, but here’s where we have to make an important point, which is that an application for any technology isn’t the same as a justification for it. You can use a brick to hammer in a nail, but hammering is a poor justification for buying bricks. To make it even more subtle, you can demonstrate that AI has “value” in creating a research report, but is the overall value enough for you to be willing to pay for AI, and pay enough that providing it is profitable?
The story says “Nvidia is working with 150 telcos around the world, including 90% of the top 50, and they are rapidly adopting AI across their business for internal productivity, customer experience and improving performance and performance-per-watt on the wireless network….” I agree; I’ve chatted with some of them. The question remains whether these applications of AI are really going to transform their business, or even end up justifying continued AI spending.
The top application of AI today, in every single vertical is the “assistant”, which uses AI as a personal productivity tool. Almost two-thirds of the people who use assistants admit to me that they’d never pay for what they get; either they’re using a free tool or their company is picking up the tab. A quarter the comments I get on the technology suggest that people actually hide assistant use because they don’t believe their management would approve it, and other surveys published recently suggest the same thing. Every operator told me they use assistants somewhere, but none said it was transformational.
The second-place enterprise AI application is the support chatbot, which 58 of the 67 operators said they used. This application, at least, gets a positive check from operators’ executive suite, so while 44 of the 58 said that support chatbots had proved more expensive than expected, had not generated as positive a customer response as had been hoped, or both, nobody said they were dropping the plans. But 38 CFOs, when asked what percent of bottom-line growth they could attribute to adoption of this application, could not respond with a number, and 3 said it was “minimal”.
Spectrum efficiency, bandwidth conservation, and network reliability all get positive marks, but again none of them were cited as offering any significant improvement to the bottom line. Even proposed opex reduction attributable to AI was an application CFOs were unwilling to say would generate a significant benefit. Add all the applications cited by NVIDIA together, and you get what a few CFOs said could be “a percent or two” of improvement.
But is this enough? It may well be enough to justify operator AI interest. The operator CFOs admit that they were approving AI projects whose ROI was far lower than their target. Two said that they had or would approve a project with a single-digit ROI. The reason is that cost reduction, if provable, is a major target for operators who have little faith in new revenue opportunities.
Enterprises are also eager to find ways of reducing the cost of network equipment and services, and in particular to reduce the number of operations errors that can impact QoE or security. CFOs of enterprises are likely (by a 2:1 margin) to accept a project ROI lower than their normal target for network AI that’s directed at these missions. A slightly lower percentage feel the same way about AI chatbots applied to their own pre- and post-sale support missions.
This leaves us at an important point. AI reality isn’t necessarily, or even likely to be, AI transformation or revolution. Yes, there are good things that operators, and enterprises, can do with it. Many will be done, but will they justify the level of investment already made, the almost-a-hundred-billion boost in cloud capex, for example? Much less, justify its increase in the future? Not so far. People won’t pay enough for what they’ve proved AI can do. But it can do more, can be transformational. We have to learn how to make that happen, and just a pretty song at a conference isn’t the answer.
