AI Growth Driving Cloud Provider Investment || Tech Field Day News Rundown: May 7, 2025

Amazon Internet from Space Palo Alto Networks buys protect AI and for teams up with Nvidia for operational technology, cloud costs are higher than expected, but value to businesses, even higher chat bots, are they giving you good health advice? And Huawei, they've got AI chips to take on Nvidia and beat the US restrictions. While we're on ai, AI growth is driving cloud provider investment.

Join us on the Tech Field day rundown for all of the news that is hot this week. Welcome to the Tech Field Day rundown, where each time we meet, we run down the news of the week with variable degrees of snarkiness. I'm your host Alistair Cook, and joining me today is my guest co-host Chris Kronman.

Chris, welcome to the show on National Tourism Day. Yeah, thanks. I'm glad to be here.

I'm touring the, uh, the, the news. Excellent. Well, we must also acknowledge National Packaging Design Day and make sure we wrap this up with a beautiful bow, uh, and be very careful on National Barrier Awareness Day.

Do not crash through the barriers and end up over the side of the bridge. That does not end well. Jeff Bezos says there's plenty of room for winners in space delivered internet.

The first 27 of Amazon's project, Coupa satellite satellites launched into low earth orbit to be joined by 1600 more in the next year. Has starlink already taken this market, or is there still room for Coupa? Yeah, I think it's an interesting question and, and Jeff went on to say that, um, Amazon is more focused on unserved and underserved communities around the world.

I do note, though, that it's not just a two horse race, right? ViaSat and Hughes Net are out there when the traditional, um, geo space tesat up in Canada has geo and middle Earth orbit satellites. Uh, one web is out there with 648 low earth, uh, orbit satellites.

Um, they mostly partner with telcos though, so that's not, so not a service you can just buy at home like you can with, uh, starlink. And there's others too. Iridium is out there doing, um, narrowband, uh, LB band service for backup often to other satellite providers, which is super interesting.

And then of course, yeah, there is starlink and it's, it looks like a tough row to hoe for Bezos. And, and these guys, because, you know, 27 satellites is great. The total that they think they're gonna launch is over 3000.

Um, and they think they're gonna get half of that up by mid 2026 with as many as five launches this year. But starlink already has 8,000 plus, uh, low Earth Orbit satellites out there. They already have 5 million users, and they got that through 250 launches.

Uh, and of course, they have an advantage in the launching space because, uh, launch delays, um, can cause big problems as they already have for, for Amazon service. Uh, so it'll be really interesting to, to see how this plays out. I I do think that there's, you know, not a, a, you know, winner takes all in the satellite, uh, communication space, but I also think that there's a big headstart for starlink.

Palo Alto Networks announced that the RSA conference that they have acquired AI Intelligence startup protect AI reports in April, suggested the price would be around $700 million. Although Palo Alto isn't disclosing the terms, it seems security is increasingly being handed over to AI products and the announcements were thick and fast at RSA. Is this AI washing or does AI make a difference in the arms race between attackers and defender?

I think AI is going to be a, a really important part of building a defense strategy for fast changing environments. We know that attackers are using AI tools to actually, uh, generate new types of attacks, and clearly defenders need to be moving as fast as they can. The fun part here is that protect AI is not really about putting AI into a, a solution into a security solution, so much as protecting the AI itself.

Uh, we had discussions, uh, the week before last at AI Infrastructure Field Day around the safety and security of AI infrastructure and AI applications. And this is the, the safety part, sorry, the security part, not the safety part. The security part is where protect AI plays.

It's around scanning the models to make sure that there are no vulnerabilities in the models that are being deployed out, uh, identifying where there are vulnerabilities and reporting those banks for corrections. So, uh, the Protect AI is expanding the coverage that Palo Alto Networks has for protecting your applications to specifically protecting the AI and the generative AI components of your application. So I think this is very much an awareness that generative AI applications are another attack vector that can be used against you.

And we've seen through things like prompt, uh, poisoning of models and prompt engineering and prompt escape that there are definitely some challenges with AI applications and with securing those applications. It's definitely a specialist skill that needs to be built into core mainstream security products. We don't really want to have this collection of point solutions for just the, the developing security issues in ai.

We do want to see this comprehensive view of security across our organization so we can have a consistent security stance on the entire organization. So I think we'll continue to see this. We have already seen some acquisitions of niche AI startups by larger organizations to build out that portfolio approach to securing and operating AI infrastructure.

While we're on announcements from the RSA conference, ForeScout Technologies announced an integration to NVIDIA's Bluefield dpu, formerly known as smartnick. The ForeScout on-Premises sensor runs directly on the Bluefield DPU to offload things like deep packet inspection or anything else that requires a lot of compute work. Uh, offloading those from the CPU, the announcement particularly highlights operational technology and and IOT use cases as there more pro, uh, proliferation of hardware offload at the edge.

Chris? Yeah, I think so. I, I remember when the DPU first got announced by Nvidia, uh, at, at a, at the tech field day, at least after the first time I saw it.

And there was a lot of conversation around what are the use cases here, right? I mean, it looks neat. It's whizzbang, this is really cool.

We've had smart nicks before. What, what's new here? It turns out there has been a lot.

But, you know, one of the questions was in a data center environment, you know, can't you just throw more compute at it? Why not just rack another server or put an appliance in there? You're not saving a ton by just moving stuff to the nick because you are adding cost, you're adding, you know, uh, power, uh, things like that.

And so in that context, yes, I mean, the edge makes a lot of sense, and especially where IT and OT are converging, right? Which is where this is focused at on that critical infrastructure where you've got operational technology, um, and it's a place that ForeScout plays really, really well. And a place where minimizing the footprint of whatever you're putting out there makes a ton of sense, right?

You don't have infinite rack space, you don't have infinite power. You can't just load appliances up when you're out on a factory floor or wherever else. These things may be happening in a plant, uh, water treatment facility, whatever it might be, right?

Uh, and so ForeScout taking that leadership in it, iot, I, iot, I mt, uh, ot, kinda all, all of the, uh, the, the edge applications where you're dealing with a lot of data and in different, um, protocols, right? So being able to do deep packet inspection of industrial network protocols is something that they specialize in. And I note that that's been shown and recognized, uh, giga Ohms most recent operational technology security report listed ForeScout as a leader and a fast mover in the platform play quadrant.

You know, seeing that this is a comprehensive play there, and I think combining that with the DPU at the edge makes a ton of sense for critical infrastructure. All of that said, you know, kind of going back to what I said earlier, you know, this isn't the only place that DPU are gaining traction. There's a ton of tools and frameworks out there.

Um, the partner list is a laundry list of kind of all the other players in, uh, networking and storage and security. Uh, and to some of our earlier points, AI applications are becoming a big place where this offloading makes a ton of sense. So, so yes, the edge is hot for, uh, um, this offload, but I think there's lots of other places where it plays as well.

Speaking of other places, uh, spending more money than you expect, yet still feeling like you are saving money, seems odd. Yet that is what a survey of public cloud providers revealed. The CIOs surveyed reported spending an average of 30% more than they expected, and more than half would still get approval for more increases.

Clearly, these organizations are seeing value from their public cloud spend, should they be looking for Better financial governance of their cloud costs. I think the survey was really interesting because they surveyed 3000, uh, CIOs at enterprise organizations. So not small organizations.

They're, they're larger organizations, and they found that overwhelmingly these customers were spending more than they expected on cloud. Now, that doesn't really come as a surprise to anybody who's been around for a while. The bit that was surprising was that they still felt they were getting more value than they were actually spending on, they're getting bitty better value for their money in the cloud.

And that, again, four out of five were, were going to be able to get an increase in spend through their governance, within their organization fairly easily, provided they could show that it was gonna continue to deliver more value. There were a few, about 30% who were saying that those increases were dependent on market conditions, which we know are a little uncertain at the moment, and hopefully we'll resolve out. But I think this highlights something that we've seen is that there's a, a maturity coming through in the way organizations are using public cloud, and that the financial operations finops movement is heading public cloud.

And what we're seeing here is the link to spending more money might deliver more value in our organization, but we need to know where to spend that money. The idea of finops is to be able to do that identification. If we spend money here, we're gonna get more revenue there.

If we spend excess money in another place, we're not gonna get more revenue. So while we're spending more money there, this mature approach to the cloud, the public cloud is a really good tool set. Uh, you only have to pay for what you use, but you pay for everything you use.

So managing that, so we spend the amount of money we need to, to get maximum value rather than just trying to spend as little as possible. I think this maturity and understanding that there is huge business value to be received from using public cloud technologies for the sorts of use cases where it has beneficial. And then rolling back, we've definitely seen some rolling back from everything should be in the cloud to, we use the cloud as one of the tools that we have in our, our, uh, in our workshop and other tools like on-premises, environments, software as a service and co-located, uh, data centers are all tools we might choose to use.

Um, definitely speaks to maturity that these are being made. These decisions are being made on a value to business basis. Healthcare advice from humans is expensive, and waiting lists for specialists are only getting longer.

AI chatbots on the other hand, are widely available and faster response. So naturally we asked chatbots like CHI Chat, GPT for health advice, an Oxford led study that showed that people aren't necessarily getting good outcomes from consulting AI experts. Have we confused the certainty of an AI chat bot for the education of a healthcare professional?

I, I think the answer there is yes and no. Right? I did see another survey that said about one in six American adults are already using chatbots for health advice at least monthly.

And when you combine that with the results of this, uh, study from Oxford, it's perhaps a little concerning, right? So the survey, uh, talked to, they had 1300 people involved in the uk and they had them use chat, GPT GPT-4 O, uh, as well as coheres command, R plus and Metas LAMA three. Um, what they found was that the folks who were using the LLMs that were, that were using these chat, uh, these AI chatbots were actually less likely to identify a relevant health condition, and they were more likely to underestimate the severity of the conditions that they did identify.

Um, so that's really bad that, that doesn't sound great at all for the chatbots. Now, um, a a commenter from the survey did say that the participants often omitted key details when querying the chatbots, uh, or they received answers that were difficult to interpret. And so some of this isn't the M'S fault, it's the user's fault, which we can all understand.

That doesn't change the outcome though, that you're misidentifying or not identifying health issues. And then, um, you know, ranking them as lower, uh, problematic than they, than they would be. So, you know, these standard, um, chatbots like chat GBT have even been recommended by the American Medical Association not to use them.

Um, now all of that said, though, I think there is a difference here between, you know, throwing the baby out with the bath water and, and maybe not relying on a open model or, or, you know, accessible consumer model to do all the specialized things. There is work underway by Apple and Microsoft to build, um, specifically trained models and applications for healthcare analysis and, and, and things like that. And I don't know that those are necessarily going to be inherently bad or wrong, uh, in that they're trained on more specific data, right?

Because remember all the models that we mentioned that are being used in this survey were trained on internet data. And so it's a little bit of the garbage and garbage out problem that I think we all need to be aware of, that not everything on the internet is reliable, and therefore, um, even a perfect LLM using that data can't ever be completely reliable. So, um, yeah, be careful when you're using it yourself at home right now, but, uh, I think there's gonna be advance advancements in this space that we all should pay attention to In other, in other AI news, the rundown has covered some ways that the US government is trying to limit China's access to high-end AI hardware.

Now, Huawei is close to shipping their own AI chip, the Ascend 9 0 1 D, which is being produced by the Chinese state-owned SMIC foundry. Uh, Huawei expects the 9 1 0 D to be more advanced than NVIDIA's H 100. Although the earlier nine 10 C failed to live up to the hype, China may well end up self-sufficient for AI processors making the US restrictions irrelevant.

Will we See Huawei AI chips coming to the us though? I think this has, um, got a couple of interesting dimensions to it. So one of the things is that whenever you restrict somebody's access to, to something they want, well, they'll find ways around those restrictions, whether it's the, uh, as we've seen some of the, the smuggling of AI chips to places that are restricted, uh, but also domestically produced.

So the US policies are about the US being, uh, self-sufficient. Well, this is clearly China making moves to be self-sufficient as well. And Huawei has seen this opportunity, they've seen it for a while, since of course the, the nine 10 C chip was released a little while ago.

It was supposed to be close to the performance of the H 100, and now this nine 10 D is due sometime this month. And whilst it's hard to see Huawei rapidly overtaking Nvidia, you can imagine that there's a second mover advantage here that Huawei is in a position to develop more rapidly from a standing start than maybe, uh, Nvidia did 20 years ago. But Nvidia has an awful lot of expertise and an awful lot of money coming into fund, more development.

And so it'll be hard work for, uh, Huawei to catch up and to exceed what's being delivered by, uh, Nvidia. So seeing the Huawei chips coming into the us, possibly not other parts of the world, quite possibly. We'll see again, if there's, uh, restrictions on export or tariffs on export of chips, uh, we may see some desire to second source, but I think it is gonna be a long time.

Uh, it's gonna be a while before the Huawei chips actually are competitive with the newest of Nvidia chips. On the other hand, China can't get the newest of Nvidia chips, so you only have to be faster than the ones that are accessible in China. Uh, Huawei will probably find a large market domestically in China, and it is a huge market, uh, whether it comes overseas.

Well, Huawei doesn't necessarily need to send all of their technology overseas with a huge domestic market, but I think we will see in particularly developing economies, uh, cost effective, uh, solutions coming out of China, being shipped into developing countries, including potentially these Huawei processes. It's now time for a closer look, and we can take a closer look now at how the demand for AI services is driving more growth in public cloud platforms. Amazon, Microsoft, and Google have all continued massive investment in public cloud platforms to keep up with the generative AI demand.

Andy Jassi highlighted triple digit growth in AI revenues, and Microsoft has committed to 40% more capacity in Europe alone. Google is investing $17 billion in cloud infrastructure. Is this still the beginning of the AI cloud build out, or are we in the middle?

It's a really interesting question. I think we're still towards the beginning, and I think it's gonna bleed beyond just the big cloud providers. I think this may be an opportunity for upstart folks that are focused on AI to maybe, uh, you know, I don't know if they're actually gonna get, you know, to compete with these, uh, really, really big companies, but maybe get snapped up by them, right?

Uh, I do think it's interesting if you dig into the numbers that Google seems to be the ones spending the most, and that would make sense, right? I think that their cloud position is probably third in ranked behind those other two as far as percentage of the market they're picking up. But o you know, over the course of the year, they're talking about increasing CapEx by 40%.

Uh, so they spent a little over $50 billion last year. 3 billion for the quarter. Um, so they're gonna outspend their revenue this year.

Um, it seems like they think there's some runway here. There's definitely some catching up that Google's trying to do to the other big guys, but, uh, but like I said, my interest is in these other folks that are coming in AI specific, uh, that might be able to give the bigger guys a run for their money. What do you think, Alistair?

Well, I think it was interesting that at AI infrastructure field day a week before last, one of the presenters was a company called Caruso, and they build AI data centers where there is disposable or excess power. And so they're building these data centers where the power cost is low and therefore the cooling cost is low. And this addresses one of the, the issues that I see in these massive buildouts, that there's an environmental risk here.

We're building these massive data centers on more power required more, uh, essentially more concentration of power means more, more heating of the, the country or the, the planet. Uh, I like this movement of these, uh, clouds towards where there's waste energy or low environmental impact energy and the, uh, sustainability focus for some. So definitely I, I like, there's smaller clouds, what's sometimes referred to as neo clouds, new clouds being built for specific purposes, and AI is one of those purposes.

But we do definitely see huge growth and innovation going on in the main cloud providers because the public cloud is a good place to experiment with these massive workloads that you may only need for a transitory period of time. Building out an infrastructure to fine tune a massive, uh, large language model on premises is very expensive. And if you only need it for maybe two weeks every three months, then that's not a very good use of your money.

This is why we're seeing a lot of cloud experimentation and cloud development happening in public cloud. We may see a movement back to on-premises, uh, for the inference stage where you're actually using the AI to build, so to deliver some business value inside your application. That inference stage is more often closely coupled where your data lies, whether it's on premises or, uh, sitting out in the public cloud.

But yeah, continuing growth in investment in public cloud infrastructure. And, uh, we saw quite a lot of the innovation from Google and their, their build out and uh, uh, their, their desire to own even more of the space of AI where AI infrastructure failed as well. Uh, I think we still are in the middle.

I hope we're in in the middle if we're just at the very beginning. The AI build out and the cloud is gonna be huge and, uh, all consuming. So hopefully we're in the middle of it as we're looking to the rest of the month.

Well, today, in fact, is the first day of Mobility Field Day. Uh, you can find Tom Hollingsworth our partner on this here, rundown, uh, we'll be hosting Mobility Field Day right now day. Tom, I hope you're having a great time with, uh, the huge number of companies and delegates that you have for Mobility Field Day Next week, Tom gets to take a bit of a rest as the Tech Field Day experience at Click Connect rolls in and Steven will be hosting a different group of delegates at Click Connect.

But after another week off at the end of the month, security Field Day rolls in and Tom will be back on deck again. He will be leading the charge for Security Field Day, and again, uh, some great presentations from maybe even some of the companies we've covered in this rundown. Uh, following that, I will be back in Silicon Valley for Cloud Field Day at the start of June.

So thank you for watching this episode of The Tech Field Day rundown. You can catch a new episode every Wednesday as a YouTube video or in your favorite podcast application. The rundown is streamed on Techstrong TV as well.

You can watch us over on Techstrong and Futureum Group programs. We'll be back next week Wednesday to talk about all of the IT news of the week. So from myself and the at, thanks for joining us and have a great week.