Having spent last week on the network operator/telco space, it’s time to take a look at the enterprise side, and in particular the trends and developments in cloud computing and AI. Wall Street’s survey of CIOs shows that cloud spending growth slowed significantly so far this year, perhaps to a third of last year’s rate (which itself was slower than the year before). Enterprise comments (from 429 this year so far) confirm this, and also confirm that so far Microsoft seems to be leading the pack of cloud providers in growth.
What I can’t confirm about the Street view is the expectations cited. They say that 70% expect over half of their workloads to be in the cloud this year, growth of ten percent incrementally from the value today. I only heard 13% of the enterprises suggest half their workloads were running in the cloud, and only 25% said they expected the workload of the cloud to increase as a percentage of total work. I think all “attitude” or “expectations” questions asked of enterprises on topics that are getting a lot of ink have an inherent distortion created by “bandwagoning”.
The survey also says that most enterprises have already re-evaluated and optimized at least half of their cloud costs, when only 22% told me that they’d done even a third of their work in that area. I’m not sure why there’s such a difference here, but the Street data is based on only CIO comments, and the base is smaller than mine. The Street base is less vertical-diverse too, which may be the largest factor of all, since my own base is fairly representative of the distribution of enterprises across verticals.
The Street is even more optimistic on AI. The Street says that CIOs see the cloud providers as the biggest beneficiary of AI adoption, but that’s an interesting question to ask because it implies a forecast on the winning AI supplier rather than on the enterprises’ own plans for AI. Of 359 enterprises who offered comments to me on their usage of AI, 79% said that currently their greatest AI use is a cloud-hosted, generative, model, but they also said that this use was driven by individual or group exploitation of the fact that AI services could be expensed, and thus usually didn’t require much up-the-command-chain approval. Only 17% thought that cloud AI could or had been transformational in any sense. In contrast, though only 16% cited a significant self-hosted AI deployment, all but five of these companies said they’d achieved significant benefit from it.
The development most linked to these significant benefits was IBM’s AI initiatives to deploy tools to build and deploy AI agents that (to quote the IBM announcement) “represent a fundamental tipping point in the AI revolution, shifting from AI that can chat with you, to AI that can do work for you—with unprecedented autonomy.” I’ve noted in past blogs that IBM is almost unique in framing AI as a way of developing components integrated with current automated workflows, and they seem to be formalizing and reinforcing this position, especially over the last couple of months. In fact, 32% of enterprises say they’re now looking at that approach, when only 17% said that at the beginning of the year.
I think it’s pretty clear that the real business value of AI will never come by supporting simple “productivity” benefits of typical workers, by enhancing their emails and documents. It has to come from one of two places, either the improvement of current workflows in supporting business operations, or by empowering new workers doing things that have not yet been effectively supported through automation. This is the “real-time, real-world” stuff I’ve been blogging about for some time.
In that area, Nvidia had an interesting comment at its shareholder meeting. “We have many growth opportunities across our company, with AI and robotics the two largest, representing a multitrillion-dollar growth opportunity”. For sure, Nvidia is accepting that they can’t live and grow on GPU chips forever, so they’re eager to point out other spaces, but robotics is one of the links to that real-time-real-world stuff that collectively represents the single largest benefit pie on the table for IT and network vendors to dig into. We have robots in factories, largely as fixed elements in an assembly line or as part of an automated warehouse transport process, but there are plenty of areas where a more autonomous application of robots could make a big difference.
The interesting thing about robotics as an AI mission is that it’s a bit like autonomous vehicles in terms of issues and implementation. It is simply not possible to cede the collision avoidance and maneuvering tasks to a remote process; latency can kill when a second moves a vehicle nearly 100 feet or a robot punches you in the gut instead of shaking your hand. A realistic robotic mission, like an autonomous vehicle mission, is a mixture of strategic and tactical, thoughtful and instinctive.
All this means that AI control of complex real-time processes can’t realistically happen without some hierarchy of AI agents, not unlike a human-driven version of the same thing. In a well-run organization, there’s a hierarchy of responsibility and decision-making. An architect doesn’t drive rivets or pour concrete, nor does a construction supervisor. Instead, work is assigned and the real or virtual agent who gets it is expected to handle it autonomously within the specified constraints. For AI, this means having not one agent, but layers of them, all melded into a complex framework. The question is how that framework is created, or even represented.
The Digital Twin Consortium has also announced something that can tie into all this real-world stuff. As I’ve said in the past, digital-twin technology is a key to making any real-world system work right, and specifically for integrating AI with such a system. A digital twin of a complex process is what allows for hierarchical control. You could view the real-world future as a digital twin of an automated-and-robotic process set, with AI overseeing the process at multiple levels.
What we need here is something (or someone) to unify all the pieces of our real-world automation framework. We have, from market leaders, the key pieces in place, but it still seems to be left to enterprises to assemble them into a working ecosystem. Only 19 enterprises had comments that suggested to me they were even looking at this, and only two had taken steps to actually execute on any combining of the various parts.
Generally, the market is moved by those who have the most to gain. To me, the obvious candidates there would be IBM, Nvidia, the combined (now that DoJ and HPE have settled) HPE/Juniper entity, or Oracle. IBM is a leader in real AI to date, and winning here would cement their position. Nvidia needs to go beyond chips, and they’re the first player to involve robotics in the story. HPE/Juniper has all the pieces, particularly if you consider that Mist might play a role in real-world automation since networking is a real-world process. Oracle has gained a lot in helping AI with cloud services. And, of course, there are the cloud providers themselves.
Of this group, I’d bet on Nvidia or Broadcom, because they have the broadest opportunity in the real-world-real-time space. IoT process control isn’t s particular strength of the other players, and it’s not clear if they could build a spot for themselves in that market quickly. If they couldn’t, they could be spending marketing and education dollars to prepare a market for someone else.
Broadcom has a broader chip position, including chips to support IoT access. It almost surely has more stuff to engage with, but it’s not a great friend of cloud providers. Nvidia is making its money these days by selling to cloud providers, and they could use their influence to encourage them to take some important first steps, steps that we could see even in 2025 if everybody got serious. Either of the two companies has some motive to push, but as far as pushing AI and the cloud, Nvidia has the inside track, providing they make the right moves.
