Why Big Tech Is Spending Billions On AI

1 00:00:01,120 - > 00:00:04,080 Mark Rotheram: So the investment is absolutely crazy that's 2 00:00:04,080 - > 00:00:06,240 going into this, and it's all AI. 3 00:00:06,240 - > 00:00:09,119 So you've got a curve that we're on now that is just 4 00:00:09,119 - > 00:00:10,720 getting bigger and bigger and bigger. 5 00:00:10,720 - > 00:00:11,919 And you can see it happening. 6 00:00:11,919 - > 00:00:13,439 The investment is getting bigger. 7 00:00:13,439 - > 00:00:16,399 Everyone that is a cloud player is investing heavily into this. 8 00:00:16,719 - > 00:00:19,839 Matt Lovell: You look at the exponential growth in data 9 00:00:19,839 - > 00:00:22,719 center and data center efficiency requirements to 10 00:00:22,719 - > 00:00:25,920 support, you know, the scale out of AI technologies, and you 11 00:00:25,920 - > 00:00:29,440 would say, where's it going to come from to keep pace with the 12 00:00:29,440 - > 00:00:31,679 rate at which technology is growing. 13 00:00:34,479 - > 00:00:36,719 Sinéad Hammond: Welcome to the AI for business podcast with 14 00:00:36,719 - > 00:00:39,840 BCN, where we put AI and automation at the centre of 15 00:00:39,840 - > 00:00:40,719 every episode. 16 00:00:40,719 - > 00:00:44,000 This is a regular debrief for business leaders to digest the 17 00:00:44,000 - > 00:00:46,960 business news and the biggest stories and development in AI as 18 00:00:46,960 - > 00:00:48,000 they hit the headlines. 19 00:00:48,000 - > 00:00:50,640 And when it feels like there's a new story and technology 20 00:00:50,640 - > 00:00:53,600 pretty much daily, we're cutting through all of the noise and 21 00:00:53,600 - > 00:00:56,079 we're giving you a rundown of what's important to businesses, 22 00:00:56,079 - > 00:00:58,960 why it matters, and what you can do to keep up with the pace. 23 00:00:58,960 - > 00:01:02,719 AI news is moving fast, huge investments, better models, and 24 00:01:02,719 - > 00:01:04,719 even robots back in the conversation. 25 00:01:04,719 - > 00:01:07,920 For UK business leaders, the challenge isn't keeping up with 26 00:01:07,920 - > 00:01:08,480 everything. 27 00:01:08,480 - > 00:01:11,680 It's knowing what matters, what to do next, and how to manage 28 00:01:11,680 - > 00:01:12,879 the risk while you move. 29 00:01:12,879 - > 00:01:16,799 Today I'm joined by Mark Rutherham, CTO at BCN, and Matt 30 00:01:16,799 - > 00:01:20,719 Lovell, CEO of CloudGuard, who together will help us make sense 31 00:01:20,719 - > 00:01:24,159 of what these headlines mean for strategy and leadership and 32 00:01:24,159 - > 00:01:28,000 how businesses can keep secure as AI spreads faster through a 33 00:01:28,000 - > 00:01:28,560 business. 34 00:01:28,560 - > 00:01:30,079 So welcome guys. 35 00:01:30,079 - > 00:01:31,680 Thank you so much for joining me. 36 00:01:31,680 - > 00:01:32,400 It's nice to see you. 37 00:01:32,400 - > 00:01:33,519 How are you both doing today? 38 00:01:34,000 - > 00:01:34,640 Matt Lovell: Very well, thank you. 39 00:01:34,640 - > 00:01:35,840 All good, Sinéad. 40 00:01:35,840 - > 00:01:37,439 Always good to see you too And Mark. 41 00:01:38,400 - > 00:01:39,599 Sinéad Hammond: It's nice to have you. 42 00:01:39,599 - > 00:01:42,400 So we're going to start the episode as we usually do with 43 00:01:42,400 - > 00:01:45,840 the top key three headlines and developments in AI over the last 44 00:01:45,840 - > 00:01:46,239 month. 45 00:01:46,239 - > 00:01:48,799 And then we'll get a bonus one from you as well, Matt, if 46 00:01:48,799 - > 00:01:49,359 that's okay. 47 00:01:49,359 - > 00:01:52,159 So, Mark, if you had to pick the three that you're most 48 00:01:52,159 - > 00:01:54,640 focused on this month, what would they be and why? 49 00:01:55,040 - > 00:01:57,920 Mark Rotheram: Yeah, so I guess one of the biggest themes, and 50 00:01:57,920 - > 00:02:01,200 it's kind of one that's been a trend for a while, is the amount 51 00:02:01,200 - > 00:02:05,840 of investment that's coming in from big tech going into general 52 00:02:05,840 - > 00:02:07,359 purpose AI. 53 00:02:07,359 - > 00:02:10,960 So it's the growth really that's been pumped into the 54 00:02:10,960 - > 00:02:15,039 entire ecosystem by Nvidia, Anthropic, pushing the 55 00:02:15,039 - > 00:02:18,319 boundaries and pushing into bigger and bigger investments. 56 00:02:18,319 - > 00:02:21,199 And I think it's really interesting that whilst we've 57 00:02:21,199 - > 00:02:25,120 seen those huge investments happen from a lot of big tech, 58 00:02:25,120 - > 00:02:29,919 just this week OpenAI paused its UK data centre projects, which 59 00:02:29,919 - > 00:02:33,439 which has kind of put another dint in the sovereign dream of 60 00:02:33,439 - > 00:02:35,759 the the UK around regulatory AI. 61 00:02:35,759 - > 00:02:39,199 So whilst we've seen some of the big players plow more and 62 00:02:39,199 - > 00:02:42,800 more money into this and uh you know that kind of growing 63 00:02:42,800 - > 00:02:45,840 capability, we're still struggling a little bit from a 64 00:02:45,840 - > 00:02:48,479 UK perspective to get the right level of infrastructure and 65 00:02:48,479 - > 00:02:51,120 investment that we need to have that sovereign AI platform. 66 00:02:51,120 - > 00:02:53,039 So that's probably the first one. 67 00:02:53,039 - > 00:02:56,240 I think hot on the heels of that, you know, we're seeing 68 00:02:56,240 - > 00:02:57,759 these investments rewards. 69 00:02:57,759 - > 00:03:02,240 So the fact that the models and the tools are becoming more 70 00:03:02,240 - > 00:03:02,879 capable. 71 00:03:02,879 - > 00:03:07,599 We're seeing some absolutely crazy news out of Anthropic, 72 00:03:07,599 - > 00:03:10,560 which I'm sure we'll have a chat with Matt about with the the 73 00:03:10,560 - > 00:03:13,680 capabilities of the models that are kind of not quite released 74 00:03:13,680 - > 00:03:16,159 but being looked at by security firms. 75 00:03:16,159 - > 00:03:19,520 But new models coming out that are extremely capable and 76 00:03:19,520 - > 00:03:20,319 extremely cheap. 77 00:03:20,319 - > 00:03:24,159 So capability is going up and costs are collapsing. 78 00:03:24,159 - > 00:03:26,960 Yeah, it's really interesting to see that that trend that's 79 00:03:26,960 - > 00:03:30,159 been happening since the probably back end of last year 80 00:03:30,159 - > 00:03:34,240 is continuing with all of the different providers and even the 81 00:03:34,240 - > 00:03:36,560 the open source models coming through as well. 82 00:03:36,560 - > 00:03:40,080 And I guess third one is you know, we're continuing to see 83 00:03:40,080 - > 00:03:41,520 the physical world evolve. 84 00:03:41,520 - > 00:03:45,680 So the robotic kind of army marching out of the data centers 85 00:03:45,680 - > 00:03:47,599 is is still happening. 86 00:03:47,599 - > 00:03:51,360 Uh, we've seen robots being deployed, starting in Japan, you 87 00:03:51,360 - > 00:03:53,759 know, they've started actually getting them out there now. 88 00:03:53,759 - > 00:03:57,759 And the the thing that's enabling that is these reduced 89 00:03:57,759 - > 00:04:02,080 cost AI models with with the more capable thinking and brain. 90 00:04:02,080 - > 00:04:06,080 So the the headlines are really uh a continuation of what's 91 00:04:06,080 - > 00:04:06,560 going on. 92 00:04:06,560 - > 00:04:09,840 It's just again spiralling and getting bigger every time we 93 00:04:09,840 - > 00:04:10,319 look at them. 94 00:04:10,639 - > 00:04:10,879 Sinéad Hammond: Yeah. 95 00:04:10,879 - > 00:04:14,159 We'll dive into each of those kind of a bit more together. 96 00:04:14,159 - > 00:04:17,680 And I want to get both of your opinions on all of the news that 97 00:04:17,680 - > 00:04:18,800 um Mark's talking about. 98 00:04:18,800 - > 00:04:21,439 But yeah, it's definitely a general trend of just bigger, 99 00:04:21,439 - > 00:04:24,480 faster, more easily accessible, all of those things. 100 00:04:24,480 - > 00:04:26,959 So we'll dive into that a little bit after this. 101 00:04:26,959 - > 00:04:28,560 Matt, what about you? 102 00:04:28,560 - > 00:04:29,920 What are you seeing at the moment? 103 00:04:29,920 - > 00:04:33,040 I mean, you come from a cybersecurity perspective and 104 00:04:33,040 - > 00:04:34,319 your expertise in there. 105 00:04:34,319 - > 00:04:37,600 So what would you say is kind of the headline that you're 106 00:04:37,600 - > 00:04:39,360 really keeping an eye on at the moment? 107 00:04:39,839 - > 00:04:44,720 Matt Lovell: So AI brings huge opportunities for businesses to 108 00:04:44,720 - > 00:04:48,240 accelerate and yeah, depending on your vertical, depending on 109 00:04:48,240 - > 00:04:52,800 how applicable obviously AI is to that vertical, and it applies 110 00:04:52,800 - > 00:04:57,519 to everybody uniformly, the actual adoption of it and all of 111 00:04:57,519 - > 00:04:59,360 aspects of adoption, right? 112 00:04:59,360 - > 00:05:03,279 From user awareness and user effectiveness of using the AI 113 00:05:03,279 - > 00:05:08,480 tools, the speed, as Mark said, the IO tools are evolving, but 114 00:05:08,480 - > 00:05:10,240 also the security implications. 115 00:05:10,240 - > 00:05:12,399 And it's not just the tool itself. 116 00:05:12,399 - > 00:05:15,120 We've started to see the wave of vulnerabilities come through 117 00:05:15,120 - > 00:05:18,480 in the tools, and most of them are high or critical, which is a 118 00:05:18,480 - > 00:05:19,439 real concern. 119 00:05:19,439 - > 00:05:23,680 But it's actually data sharing and what it's doing because 120 00:05:23,680 - > 00:05:28,160 organizations are reaching into the tools and starting to truly 121 00:05:28,160 - > 00:05:31,279 understand the capability, and the capabilities accelerating at 122 00:05:31,279 - > 00:05:33,199 exponential rates, as we've said. 123 00:05:33,199 - > 00:05:37,279 Understanding how you do that securely when user awareness is 124 00:05:37,279 - > 00:05:38,800 already probably too low. 125 00:05:38,800 - > 00:05:43,120 The organizations investing in helping people on that journey, 126 00:05:43,120 - > 00:05:45,680 you know, whatever their position is in an organization, 127 00:05:45,680 - > 00:05:48,560 however, AI may be able to make them more effective, and I think 128 00:05:48,560 - > 00:05:49,920 across the board it can. 129 00:05:49,920 - > 00:05:53,600 But being mature enough as an organization to share data, 130 00:05:53,600 - > 00:05:57,040 share data securely, understand how that works within workflows, 131 00:05:57,040 - > 00:06:01,120 within supply chains, you know, within partnerships, and who 132 00:06:01,120 - > 00:06:03,759 you're sharing data and how they're then managing that data 133 00:06:03,759 - > 00:06:07,759 onward, it's a real maturity curve for organizations don't 134 00:06:07,759 - > 00:06:09,759 seem very well prepared for right now. 135 00:06:09,759 - > 00:06:12,240 That's what we are fundamentally seeing. 136 00:06:12,639 - > 00:06:14,000 Sinéad Hammond: Yeah, definitely. 137 00:06:14,000 - > 00:06:16,639 And I think that's something that we've talked about on this 138 00:06:16,639 - > 00:06:18,000 podcast before, haven't we, Mark? 139 00:06:18,000 - > 00:06:21,759 We've talked about kind of the curve and trend of technology 140 00:06:21,759 - > 00:06:24,000 being available and then how businesses are actually 141 00:06:24,000 - > 00:06:27,360 preparing and embracing that as it goes forward. 142 00:06:27,360 - > 00:06:29,360 So I think that's going to be a really interesting thing to 143 00:06:29,360 - > 00:06:31,519 talk about, and something that'd be really useful for um 144 00:06:31,519 - > 00:06:32,399 listeners today. 145 00:06:32,399 - > 00:06:34,879 So let's dive into them one by one in a bit more detail. 146 00:06:34,879 - > 00:06:38,079 We talked about big tech doubling down on AI investment. 147 00:06:38,079 - > 00:06:42,959 You mentioned Nvidia, Anthropic, OpenAI, like who's 148 00:06:42,959 - > 00:06:44,959 doing this and what does it signal to you? 149 00:06:44,959 - > 00:06:47,199 And this is kind of an open question to both of you, I 150 00:06:47,199 - > 00:06:47,600 suppose. 151 00:06:47,600 - > 00:06:51,360 Like, what what does this mean for the way that AI is 152 00:06:51,360 - > 00:06:53,600 progressing in the landscape right now? 153 00:06:54,000 - > 00:06:57,759 Mark Rotheram: Yeah, so I think the easy way to visualize it is 154 00:06:57,759 - > 00:07:01,360 becoming an absolute dominant player globally. 155 00:07:01,360 - > 00:07:05,839 Yeah, so we've got Nvidia becoming a multi-trillion dollar 156 00:07:05,839 - > 00:07:09,680 company as their demand for chips is just exploding. 157 00:07:09,680 - > 00:07:12,480 You know, they are the people that are building the roads that 158 00:07:12,480 - > 00:07:14,240 everything runs on effectively. 159 00:07:14,240 - > 00:07:17,120 So that in itself is huge. 160 00:07:17,120 - > 00:07:20,160 You know, last year, last year, the year before they became the 161 00:07:20,160 - > 00:07:22,959 most valuable company in the world and and they are just 162 00:07:22,959 - > 00:07:23,839 continuing on. 163 00:07:23,839 - > 00:07:27,439 But hot on their heels, we're seeing all of the big players. 164 00:07:27,439 - > 00:07:32,720 So Amazon putting another 35 billion into their strategy, you 165 00:07:32,720 - > 00:07:32,800 know. 166 00:07:32,800 - > 00:07:35,759 They they actually have their own chip line as well that has 167 00:07:35,759 - > 00:07:38,639 got massive consumption for their AWS platform. 168 00:07:38,639 - > 00:07:42,319 Anthropic, you know, they they have signed massive deals, you 169 00:07:42,319 - > 00:07:44,319 know, they don't have their own infrastructure, they're working 170 00:07:44,319 - > 00:07:48,399 with Broadcom and Google, you know, very much focused on how 171 00:07:48,399 - > 00:07:50,879 they can buy as much capacity as possible. 172 00:07:50,879 - > 00:07:54,720 Meta, they're on the bandwagon now, billions again going into 173 00:07:54,720 - > 00:07:59,360 AI, and even Intel, you know, Intel have started working with 174 00:07:59,360 - > 00:08:03,360 Google and with Elon Musk in the the Terrafab chip project, 175 00:08:03,360 - > 00:08:06,560 which again, if if you look into what's happening with Terrafab, 176 00:08:06,560 - > 00:08:10,000 that's predicted to be the biggest chip manufacturer 177 00:08:10,000 - > 00:08:12,160 overtaking everything else that's happening in the next few 178 00:08:12,160 - > 00:08:12,879 years as well. 179 00:08:12,879 - > 00:08:16,800 So the investment is absolutely crazy that's going into this. 180 00:08:16,800 - > 00:08:20,959 And over the next year, we're seeing three of the biggest 181 00:08:20,959 - > 00:08:25,759 players, Anthropic, OpenAI, and SpaceX, who owns XAI, all 182 00:08:25,759 - > 00:08:29,519 IPO-ing, you know, effectively going open up to the public to 183 00:08:29,519 - > 00:08:34,000 buy stocks and shares, and that will be um the biggest round of 184 00:08:34,000 - > 00:08:37,039 global investment in history, and it's all AI. 185 00:08:37,039 - > 00:08:39,919 So you've you've got a curve that we're on now that is just 186 00:08:39,919 - > 00:08:42,399 getting bigger and bigger and bigger, and you can see it 187 00:08:42,399 - > 00:08:42,799 happening. 188 00:08:42,799 - > 00:08:45,679 The investment is getting bigger, the companies, everyone 189 00:08:45,679 - > 00:08:48,639 that is a cloud player is investing heavily into this. 190 00:08:48,639 - > 00:08:52,000 So, yeah, so what it means for us is all that money's gonna 191 00:08:52,000 - > 00:08:53,200 come from somewhere, yeah. 192 00:08:53,200 - > 00:08:58,240 They're expecting adoption, they're expecting enterprise all 193 00:08:58,240 - > 00:09:02,320 the way down to small, medium corporate, SMB, and person 194 00:09:02,320 - > 00:09:06,000 people to invest and spend money in their models. 195 00:09:06,000 - > 00:09:08,559 And and to that point was just said, it's gonna be all around 196 00:09:08,559 - > 00:09:09,120 adoption now. 197 00:09:09,120 - > 00:09:10,399 Who's gonna get the market share? 198 00:09:10,399 - > 00:09:11,279 How are they gonna get it? 199 00:09:11,279 - > 00:09:13,840 How are they gonna get that adoption going, and how are they 200 00:09:13,840 - > 00:09:17,759 gonna get the money back from these multi-billion pound 201 00:09:17,759 - > 00:09:19,039 investments that they're all making? 202 00:09:19,279 - > 00:09:20,159 Sinéad Hammond: Yeah, absolutely. 203 00:09:20,159 - > 00:09:23,039 And I suppose maybe a question for you there, Matt, is 204 00:09:23,039 - > 00:09:26,320 obviously there's the one side of it where investment and 205 00:09:26,320 - > 00:09:29,759 competition is ramping up and everyone's trying to get that 206 00:09:29,759 - > 00:09:30,960 share of the market. 207 00:09:30,960 - > 00:09:35,679 What tends to then happen in terms of a risk perspective and 208 00:09:35,679 - > 00:09:38,879 what do businesses kind of need to be aware of when they are 209 00:09:38,879 - > 00:09:39,279 doing this? 210 00:09:39,279 - > 00:09:42,000 I know, Mart, you mentioned adoption, and we've sort of 211 00:09:42,000 - > 00:09:44,159 talked about guidelines and things in the past. 212 00:09:44,159 - > 00:09:47,039 But from your perspective, Matt, what what tends to happen 213 00:09:47,039 - > 00:09:50,000 when technology is accelerating at a speed like this? 214 00:09:50,399 - > 00:09:52,000 Matt Lovell: Uh lots of things, Sinead. 215 00:09:52,000 - > 00:09:54,000 I think um, where do we start, right? 216 00:09:54,000 - > 00:09:58,159 Let's let's just look at uh the dependencies, right? 217 00:09:58,159 - > 00:10:01,840 So I obviously come with my data center hat on here from my 218 00:10:01,840 - > 00:10:02,879 previous life. 219 00:10:02,879 - > 00:10:06,799 And I would say, you know, you look at the exponential growth 220 00:10:06,799 - > 00:10:10,559 in data center and data center efficiency requirements to 221 00:10:10,559 - > 00:10:13,840 support, you know, the scale out of AI technologies, and you 222 00:10:13,840 - > 00:10:18,399 would say that, you know, rare earth materials, power, uh, 223 00:10:18,399 - > 00:10:22,000 cooling, you know, just physical location space, where are you 224 00:10:22,000 - > 00:10:24,960 gonna cite that for optimized performance on the globe, the 225 00:10:24,960 - > 00:10:27,759 global, you know, obviously communication fabric? 226 00:10:27,759 - > 00:10:30,720 You then look at obviously the deployment of all of those, the 227 00:10:30,720 - > 00:10:33,840 build of all of that, you know, compute and memory, storage 228 00:10:33,840 - > 00:10:34,879 capacity, et cetera. 229 00:10:34,879 - > 00:10:38,559 Where's it gonna come from to keep pace with the rate at which 230 00:10:38,559 - > 00:10:39,679 technology is growing? 231 00:10:39,679 - > 00:10:42,240 You then think about business risk, right? 232 00:10:42,240 - > 00:10:46,159 So adoption, doing it securely, having the right training in 233 00:10:46,159 - > 00:10:48,639 place, having the right guardrails in place, et cetera. 234 00:10:48,639 - > 00:10:52,000 But we're starting to see the next wave, which is 235 00:10:52,000 - > 00:10:56,399 organizations leveraging AI capability in workflows. 236 00:10:56,399 - > 00:11:00,320 What happens if there's any disruption to service 237 00:11:00,320 - > 00:11:01,200 availability? 238 00:11:01,200 - > 00:11:04,559 Okay, because that's absolutely critical to a business. 239 00:11:04,559 - > 00:11:08,240 We've seen in the last couple of months businesses not even 240 00:11:08,240 - > 00:11:12,240 realizing that their workflows are now fully automated and 241 00:11:12,240 - > 00:11:13,360 leveraging AI. 242 00:11:13,360 - > 00:11:17,200 AI is a huge part of that workflow process, and the 243 00:11:17,200 - > 00:11:19,440 availability of that is business critical. 244 00:11:19,440 - > 00:11:21,600 Is it in their business continuity planning? 245 00:11:21,600 - > 00:11:22,879 Is it in their disaster recovery? 246 00:11:22,879 - > 00:11:26,960 If there's even sporadic or intermittent disruption to 247 00:11:26,960 - > 00:11:29,840 availability, if performance degrades, for example, and we've 248 00:11:29,840 - > 00:11:33,840 seen that because we can't forecast the huge swathe of 249 00:11:33,840 - > 00:11:38,000 people switching between the AI tooling and the effectiveness of 250 00:11:38,000 - > 00:11:42,320 AI tooling, we've seen in the last six months the huge 251 00:11:42,320 - > 00:11:45,759 acceleration of Anthropics platform and Claude and the 252 00:11:45,759 - > 00:11:49,600 capabilities of Claude, not just in obviously the AI context, 253 00:11:49,600 - > 00:11:51,279 but in code and in cowork. 254 00:11:51,279 - > 00:11:54,639 And people realizing that there's different levels of 255 00:11:54,639 - > 00:11:56,000 effectiveness between the tools. 256 00:11:56,000 - > 00:11:59,519 So if you're AWS, if you're your Google Gemini, you know, 257 00:11:59,519 - > 00:12:01,840 you're looking at that and you're OpenAI and you're 258 00:12:01,840 - > 00:12:04,240 obviously uh AIX as well. 259 00:12:04,240 - > 00:12:07,840 You know, how are these people able to plan for these huge 260 00:12:07,840 - > 00:12:10,799 peaks in activity and how does it influence the user 261 00:12:10,799 - > 00:12:11,360 experience? 262 00:12:11,360 - > 00:12:14,080 And if you've got workflows that are dependent upon that and 263 00:12:14,080 - > 00:12:17,200 suddenly they are degraded, what's your plan B? 264 00:12:17,200 - > 00:12:20,000 And therefore, you know, how are we plotting this out? 265 00:12:20,000 - > 00:12:22,799 How are we then playing that through and measuring the user 266 00:12:22,799 - > 00:12:25,759 experience those workflows are delivering to our business 267 00:12:25,759 - > 00:12:26,320 customers? 268 00:12:26,320 - > 00:12:26,639 Okay. 269 00:12:26,639 - > 00:12:29,840 They're the things that we now need to be thinking about and 270 00:12:29,840 - > 00:12:30,960 safeguarding against. 271 00:12:30,960 - > 00:12:34,399 And that's what typically happens with rapid adoption of 272 00:12:34,399 - > 00:12:35,039 technology. 273 00:12:36,080 - > 00:12:37,759 Sinéad Hammond: Mark, I saw you nodding along there. 274 00:12:37,759 - > 00:12:40,080 I didn't know if you'd got anything that you had to say 275 00:12:40,080 - > 00:12:42,639 back to that as well, based on maybe some of the clients that 276 00:12:42,639 - > 00:12:45,519 we work with and an experience that you're having at the moment 277 00:12:45,519 - > 00:12:48,639 as people are embracing more of this technology. 278 00:12:49,039 - > 00:12:50,240 Mark Rotheram: Yeah, no, Matt's right. 279 00:12:50,240 - > 00:12:53,840 I mean, there's there's loads and loads of considerations and 280 00:12:53,840 - > 00:12:57,200 issues, and and you know, we talked a little bit about 281 00:12:57,200 - > 00:13:01,759 sovereign AI and and the state of the UK from a what can we 282 00:13:01,759 - > 00:13:02,799 actually have here? 283 00:13:02,799 - > 00:13:05,200 And that's something that really needs a lot of thought 284 00:13:05,200 - > 00:13:08,159 and and consideration when we're working with our customers, 285 00:13:08,159 - > 00:13:12,799 because everyone would love to have AI native in the UK with 286 00:13:12,799 - > 00:13:15,120 all the bells and whistles, and it's just not going to happen 287 00:13:15,120 - > 00:13:15,919 anytime soon. 288 00:13:15,919 - > 00:13:23,120 You know, we've we've seen a £150 billion investment into AI 289 00:13:23,120 - > 00:13:25,840 come to the UK and then go without anything happening. 290 00:13:25,840 - > 00:13:26,159 Yeah. 291 00:13:26,159 - > 00:13:30,879 So I think when we're working with our customers, it is how do 292 00:13:30,879 - > 00:13:36,320 we provide agentic capabilities in a safe, secure, robust 293 00:13:36,320 - > 00:13:37,840 manner that you can rely on. 294 00:13:37,840 - > 00:13:39,600 And you know, it is challenging. 295 00:13:39,600 - > 00:13:42,799 We are finding ways of doing that, but I think you know it's 296 00:13:42,799 - > 00:13:45,919 it's coming at a not in the way that we would love to do it. 297 00:13:45,919 - > 00:13:48,720 It's it's in a way that we we have to put compromises in. 298 00:13:48,720 - > 00:13:51,759 And yeah, and I I can see it as being in that position for 299 00:13:51,759 - > 00:13:56,559 years, um, as our infrastructure just lags behind America and 300 00:13:56,559 - > 00:13:59,120 China and and some of the other places across Europe where 301 00:13:59,120 - > 00:14:01,600 they've got that capacity and they've got that investment. 302 00:14:02,240 - > 00:14:05,039 Matt Lovell: Equally, sorry, just to come in, I think it will 303 00:14:05,039 - > 00:14:08,320 expose, but it will also accelerate our you know our 304 00:14:08,320 - > 00:14:09,759 thinking and our capabilities. 305 00:14:09,759 - > 00:14:12,320 Because on the one hand, you know, we're saying we're gonna 306 00:14:12,320 - > 00:14:15,679 see, yeah, I mean, data center electrical consumption on a on a 307 00:14:15,679 - > 00:14:19,360 very broad, high-level global basis is somewhere between 9 and 308 00:14:19,360 - > 00:14:22,879 11% of all consumption globally, right? 309 00:14:22,879 - > 00:14:24,960 So that's already a huge number. 310 00:14:24,960 - > 00:14:29,120 Although AI and the technology it's using is hyper-efficient in 311 00:14:29,120 - > 00:14:32,480 most cases, it is still going to be consuming proportionally 312 00:14:32,480 - > 00:14:34,320 more power as an aggregate. 313 00:14:34,320 - > 00:14:39,120 How are we going to support that when we are more reliant, 314 00:14:39,120 - > 00:14:41,919 more dependent upon renewable energy? 315 00:14:41,919 - > 00:14:45,679 And we know very clearly that you know the wind isn't always 316 00:14:45,679 - > 00:14:48,799 you know high and the sun isn't always shining, certainly for 317 00:14:48,799 - > 00:14:50,799 large parts of the day, depending on what part of the 318 00:14:50,799 - > 00:14:52,000 globe you might be in. 319 00:14:52,000 - > 00:14:54,720 And therefore, how are we gonna bridge that gap, right? 320 00:14:54,720 - > 00:14:58,159 And that will accelerate nuclear fission technology and 321 00:14:58,159 - > 00:15:02,639 local nuclear, more consistent, more dependable, more reliable 322 00:15:02,639 - > 00:15:04,480 connections that bridge those gaps. 323 00:15:04,480 - > 00:15:07,600 Yes, we're gonna focus on renewable energy for sure, but 324 00:15:07,600 - > 00:15:10,960 how are we also gonna keep to you know the global warming 325 00:15:10,960 - > 00:15:14,960 thresholds that we absolutely need to commit to and even bring 326 00:15:14,960 - > 00:15:15,519 them down? 327 00:15:15,519 - > 00:15:18,240 So, how are we gonna use this technology to understand it? 328 00:15:18,240 - > 00:15:20,960 You're seeing it in the medical industry, you're seeing it in 329 00:15:20,960 - > 00:15:22,320 research and development already. 330 00:15:22,320 - > 00:15:25,279 We are making major breakthroughs because the 331 00:15:25,279 - > 00:15:28,960 technology is finding answers that we've been searching for a 332 00:15:28,960 - > 00:15:29,919 long period of time, right? 333 00:15:29,919 - > 00:15:33,360 The analytical capability will take us there, but it will need 334 00:15:33,360 - > 00:15:35,039 focus, is what I'm saying. 335 00:15:35,039 - > 00:15:38,399 What are the principal problems in scaling this technology out? 336 00:15:38,399 - > 00:15:41,519 And those are the priorities to fix, and then how is it going 337 00:15:41,519 - > 00:15:45,039 to help us fix the other priority issues around the world 338 00:15:45,039 - > 00:15:48,320 without causing greater levels of global warming, which in 339 00:15:48,320 - > 00:15:50,240 itself will accelerate other problems? 340 00:15:50,559 - > 00:15:52,639 Mark Rotheram: Yeah, and I think that's where the you know if 341 00:15:52,639 - > 00:15:55,919 you look at the the biggest trend that wasn't here a year 342 00:15:55,919 - > 00:16:00,159 ago, that's emerged recently, is data centers in space. 343 00:16:00,159 - > 00:16:03,840 Yeah, and that that is one answer, but it's not an answer 344 00:16:03,840 - > 00:16:05,039 that's going to be here today, is it? 345 00:16:05,039 - > 00:16:07,519 It's gonna be a a year, two, three years. 346 00:16:07,519 - > 00:16:10,320 We already have the first one, I think the first ones were 347 00:16:10,320 - > 00:16:11,600 launched last year as a test. 348 00:16:11,600 - > 00:16:14,480 But I think that's that is one of the big answers, isn't it? 349 00:16:14,480 - > 00:16:18,240 But that again, it it flies in the face of sovereign AI and 350 00:16:18,240 - > 00:16:19,840 sovereign kind of approaches. 351 00:16:19,840 - > 00:16:22,559 And I think that's that's where we're being pushed, if I'm 352 00:16:22,559 - > 00:16:23,039 honest. 353 00:16:23,039 - > 00:16:27,120 As we want to embrace AI, we're gonna have to accept that it's 354 00:16:27,120 - > 00:16:30,159 not all gonna be sat here in our data centre in the backyard. 355 00:16:30,159 - > 00:16:34,480 It's gonna have to be European, global, or you know, 356 00:16:34,480 - > 00:16:36,559 stellar-based uh as we go forward. 357 00:16:37,120 - > 00:16:39,679 Sinéad Hammond: Do you think that'll then change the way that 358 00:16:39,679 - > 00:16:43,279 businesses have to think about their procedures and their 359 00:16:43,279 - > 00:16:44,639 policies and things as well? 360 00:16:44,639 - > 00:16:47,840 Because you know, typically, especially since things like 361 00:16:47,840 - > 00:16:51,360 GDPR and you know policies that are in place, a lot of that 362 00:16:51,360 - > 00:16:54,799 where the data is kept has been part of that conversation. 363 00:16:54,799 - > 00:16:56,320 It's been part of business process. 364 00:16:56,320 - > 00:16:58,960 But it sounds like based on what you're saying there, those 365 00:16:58,960 - > 00:17:00,480 sorts of things are gonna have to be reviewed. 366 00:17:00,480 - > 00:17:04,240 So just by implementing any type of AI, particularly more 367 00:17:04,240 - > 00:17:08,960 agenc and more kind of advanced level, it is going to change 368 00:17:08,960 - > 00:17:09,279 that. 369 00:17:09,599 - > 00:17:11,839 Mark Rotheram: I think one one of the most fundamental things 370 00:17:11,839 - > 00:17:14,640 that we're gonna grapple with over the next 12 months is where 371 00:17:14,640 - > 00:17:17,119 data is processed, not necessarily where it's stored, 372 00:17:17,119 - > 00:17:18,240 it's where it's processed. 373 00:17:18,240 - > 00:17:22,480 So we look at you know what's happening with Microsoft Copilot 374 00:17:22,480 - > 00:17:23,599 is a really good example. 375 00:17:23,599 - > 00:17:26,160 Microsoft Copilot had been out for a few years. 376 00:17:26,160 - > 00:17:30,079 It was innovative, baked into the ecosystem, and then up came 377 00:17:30,079 - > 00:17:34,319 clawed co-work, and it's brilliant, better than co-pilot 378 00:17:34,319 - > 00:17:35,359 in a lot of ways. 379 00:17:35,359 - > 00:17:37,039 So Microsoft have done a really good thing. 380 00:17:37,039 - > 00:17:39,839 They've they've last year they said, look, we're gonna let you 381 00:17:39,839 - > 00:17:44,240 use uh the Opus model and the Sonic models within Copilot, but 382 00:17:44,240 - > 00:17:46,960 there's a data processing agreement uh because the data 383 00:17:46,960 - > 00:17:49,759 will be processed by Anthropic in their data centers wherever 384 00:17:49,759 - > 00:17:50,480 they might be. 385 00:17:50,480 - > 00:17:53,599 Um earlier this year we saw another flip where they're gonna 386 00:17:53,599 - > 00:17:57,680 further embed that anthropic capability into Copilot. 387 00:17:57,680 - > 00:18:00,640 So, you know, that started earlier this month with the 388 00:18:00,640 - > 00:18:04,720 first co-pilot kind of foundry businesses getting access to it, 389 00:18:04,720 - > 00:18:07,599 but it comes with that compromise of your data will be 390 00:18:07,599 - > 00:18:09,599 processed globally by anthropic. 391 00:18:09,599 - > 00:18:13,920 So we're already seeing it happen, and I think as we build 392 00:18:13,920 - > 00:18:17,440 out bespoke solutions, agentic solutions, we're having those 393 00:18:17,440 - > 00:18:18,160 same conversations. 394 00:18:18,160 - > 00:18:22,000 If you want access to the best models, the best we can do is 395 00:18:22,000 - > 00:18:24,480 European processing, but that comes at a tax. 396 00:18:24,480 - > 00:18:26,880 More often it's global processed. 397 00:18:26,880 - > 00:18:31,440 So we have to think about PII, GDPR, how we control that, how 398 00:18:31,440 - > 00:18:35,119 we can do things before it gets sent out, and how we we monitor 399 00:18:35,119 - > 00:18:36,079 and manage all that. 400 00:18:36,079 - > 00:18:39,839 It's not impossible, it's just like I say, more considerations 401 00:18:39,839 - > 00:18:41,039 that businesses need to have. 402 00:18:41,039 - > 00:18:43,599 They need to know that they're making those decisions. 403 00:18:43,599 - > 00:18:46,480 They need to know what the security boundaries look like, 404 00:18:46,480 - > 00:18:49,440 where things are being processed, and and the the 405 00:18:49,440 - > 00:18:50,960 what-ifs related to that. 406 00:18:51,200 - > 00:18:54,720 Sinéad Hammond: Yeah, we say this every episode: any changes 407 00:18:54,720 - > 00:18:57,759 that come with technology or a business decision, it's not just 408 00:18:57,759 - > 00:19:01,440 like a an IT, it hasn't been an IT department decision for a 409 00:19:01,440 - > 00:19:01,920 long time. 410 00:19:01,920 - > 00:19:04,640 Um, so it's all these knock-on effects that go with that. 411 00:19:04,640 - > 00:19:08,720 You've mentioned in the headlines as well around AI cost 412 00:19:08,720 - > 00:19:11,599 barriers being pretty much eradicated. 413 00:19:11,599 - > 00:19:15,440 You know, anyone can get a any sort of low-cost license. 414 00:19:15,440 - > 00:19:18,480 I mean, I think everybody I know has one where you you just 415 00:19:18,480 - > 00:19:21,839 kind of access AI with without any kind of barrier there. 416 00:19:21,839 - > 00:19:26,559 What does cheaper AI change for businesses, especially 417 00:19:26,559 - > 00:19:28,799 businesses without big teams? 418 00:19:28,799 - > 00:19:32,960 And what does that then look like when you know new models 419 00:19:32,960 - > 00:19:35,119 are being developed going forward? 420 00:19:36,160 - > 00:19:39,599 Mark Rotheram: Yeah, so the the race for higher performance and 421 00:19:39,599 - > 00:19:42,319 cheaper cost is continuing. 422 00:19:42,319 - > 00:19:46,720 We saw um the latest open source models from Google 423 00:19:46,720 - > 00:19:48,480 arrive, um, Gemma 4. 424 00:19:48,480 - > 00:19:51,039 That's something you can run on your phone, yeah, and you get a 425 00:19:51,039 - > 00:19:53,119 really good strong model just locally. 426 00:19:53,119 - > 00:19:56,160 Never mind having to pay for APIs, you know, it's it's part 427 00:19:56,160 - > 00:19:57,359 of hardware, it just runs. 428 00:19:57,359 - > 00:19:59,759 So these things are happening all the time. 429 00:19:59,759 - > 00:20:04,240 What it means for the businesses we're working with is 430 00:20:04,240 - > 00:20:08,799 that they can tap in to this, I call it a big brain, regardless 431 00:20:08,799 - > 00:20:10,000 of the provider and the model. 432 00:20:10,000 - > 00:20:13,599 You've got these big brains now that you can tap into that can 433 00:20:13,599 - > 00:20:17,119 spread across your context, unlike anything you've been able 434 00:20:17,119 - > 00:20:17,920 to see before. 435 00:20:17,920 - > 00:20:22,960 So you can get much more contextual intelligence at a 436 00:20:22,960 - > 00:20:26,799 very low cost that can really help power your business 437 00:20:26,799 - > 00:20:27,680 processes. 438 00:20:27,680 - > 00:20:31,200 So if if we kind of flip it into the the so what, 439 00:20:31,200 - > 00:20:34,400 historically it's been quite hard and costly to automate 440 00:20:34,400 - > 00:20:35,440 business processes. 441 00:20:35,440 - > 00:20:39,839 It's now a lot easier and a lot more powerful than it's ever 442 00:20:39,839 - > 00:20:40,000 been. 443 00:20:40,000 - > 00:20:43,920 So we can bring a more powerful capability at a lower entry 444 00:20:43,920 - > 00:20:46,799 price, a lower cost to run, that's much better than anything 445 00:20:46,799 - > 00:20:48,880 we've been able to deliver to our customers before. 446 00:20:48,880 - > 00:20:53,440 So we can turbocharge with a genetic workforce, agentic 447 00:20:53,440 - > 00:20:57,759 pipelines and workflows, businesses targeting teams, 448 00:20:57,759 - > 00:21:00,960 targeting hotspots, targeting just repetitive work or not so 449 00:21:00,960 - > 00:21:03,759 repetitive work in a way we've never been able to do before. 450 00:21:03,759 - > 00:21:08,319 So and the opportunity for every business to do something 451 00:21:08,319 - > 00:21:10,960 impactful is right there now without breaking the bank. 452 00:21:10,960 - > 00:21:14,400 That's the real impact, and it's growing month on month that 453 00:21:14,400 - > 00:21:15,440 potential impact. 454 00:21:15,759 - > 00:21:18,319 Matt Lovell: You know, one one of the personal observations I 455 00:21:18,319 - > 00:21:23,440 would make is how I travel into town on public transport every 456 00:21:23,440 - > 00:21:26,960 day, and you know, typically you would lose that that travel 457 00:21:26,960 - > 00:21:30,559 time in terms of productivity, but because the tools and the 458 00:21:30,559 - > 00:21:34,720 models are so powerful on your phone, you can be super 459 00:21:34,720 - > 00:21:38,640 productive for the journey time, coming in, going home, and 460 00:21:38,640 - > 00:21:42,559 solve problems and use you know that mind space in order to be 461 00:21:42,559 - > 00:21:43,279 really productive. 462 00:21:43,279 - > 00:21:46,240 Now, I'm not advocating everybody does that, not at all. 463 00:21:46,240 - > 00:21:49,920 What I'm saying is I choose to do that, and other people I'm 464 00:21:49,920 - > 00:21:53,039 looking around me and just being observant are clearly choosing 465 00:21:53,039 - > 00:21:55,839 to do that because the tool is so capable, right? 466 00:21:55,839 - > 00:21:58,319 And and certainly in the last three or four months, as Mark 467 00:21:58,319 - > 00:21:59,759 said, in in the tools there. 468 00:21:59,759 - > 00:22:04,160 Those tools have opened up whole swathes of new 469 00:22:04,160 - > 00:22:06,720 opportunities for us to be leveraging that. 470 00:22:06,720 - > 00:22:10,319 And I'm not advocating that we do it on a free platform because 471 00:22:10,319 - > 00:22:13,279 then we have different data constraints, and that data is 472 00:22:13,279 - > 00:22:17,039 more likely to be ending up in training a global model at a you 473 00:22:17,039 - > 00:22:18,000 know a shared level. 474 00:22:18,000 - > 00:22:20,960 So you've got to be super, super clear on the guidelines 475 00:22:20,960 - > 00:22:21,519 for people. 476 00:22:21,519 - > 00:22:24,960 But then people can be thinking about it and just using that in 477 00:22:24,960 - > 00:22:27,680 their hand in terms of accelerating processes. 478 00:22:27,680 - > 00:22:31,039 And I certainly found that incredibly powerful, you know, 479 00:22:31,039 - > 00:22:35,119 in terms of learning capability, distilling huge swathes of 480 00:22:35,119 - > 00:22:38,559 information that normally I'd just have a singular dimension 481 00:22:38,559 - > 00:22:40,799 of being able to read it, etc. 482 00:22:40,799 - > 00:22:43,519 I'm consuming it really differently, and that makes a 483 00:22:43,519 - > 00:22:44,319 fundamental difference. 484 00:22:44,319 - > 00:22:46,880 I was talking to somebody else the other day, and they were 485 00:22:46,880 - > 00:22:50,319 saying, Do you know what sector grew the quickest in terms of AI 486 00:22:50,319 - > 00:22:54,640 analytical use and progression in 2025? 487 00:22:54,640 - > 00:22:57,599 And I was like, Well, I would suggest to you, based on my own 488 00:22:57,599 - > 00:22:58,720 experiences, finance. 489 00:22:58,720 - > 00:23:00,559 And they were like, Absolutely not. 490 00:23:00,559 - > 00:23:03,839 That was like 25%, but healthcare grew 65%. 491 00:23:04,720 - > 00:23:05,359 Sinéad Hammond: Wow. 492 00:23:05,359 - > 00:23:06,960 So interesting. 493 00:23:07,599 - > 00:23:10,400 Matt Lovell: That is a major, major step forward, yeah, in 494 00:23:10,400 - > 00:23:14,079 terms of what that can bring to all of us at a global quality of 495 00:23:14,079 - > 00:23:16,240 life level, medical advanced level. 496 00:23:16,559 - > 00:23:17,279 Sinéad Hammond: Absolutely. 497 00:23:17,279 - > 00:23:20,400 I mean, we talk about it from a business level here because on 498 00:23:20,400 - > 00:23:23,359 this, you know, that's generally the kind of people that we work 499 00:23:23,359 - > 00:23:28,480 with, but it's got applications so far out of just the business 500 00:23:28,480 - > 00:23:31,279 processes and things that it's so interesting that we don't 501 00:23:31,279 - > 00:23:31,759 even touch. 502 00:23:31,759 - > 00:23:34,880 And I would have said finance or legal potentially, I might 503 00:23:34,880 - > 00:23:36,960 have said those sectors seeing a lot of growth in there. 504 00:23:36,960 - > 00:23:39,279 But yeah, it's so interesting that it's healthcare. 505 00:23:39,279 - > 00:23:42,160 And in fact, conversations about applications of AI and 506 00:23:42,160 - > 00:23:45,119 healthcare bring us nicely on to this huge topic which we've 507 00:23:45,119 - > 00:23:45,839 touched on before. 508 00:23:45,839 - > 00:23:49,200 And Mark, you've talked about a little bit earlier, robotics. 509 00:23:49,200 - > 00:23:52,160 So if you tune into our next episode, we'll be diving into 510 00:23:52,160 - > 00:23:56,559 it, robotics more, and we're talking about the CISA and how 511 00:23:56,559 - > 00:23:57,680 that impacts AI. 512 00:23:57,680 - > 00:23:59,039 So definitely tune in for that. 513 00:23:59,039 - > 00:24:01,200 However, in the meantime, that's all we've got time for 514 00:24:01,200 - > 00:24:01,920 for this episode. 515 00:24:01,920 - > 00:24:04,480 So, Mark, Matt, thank you very much for joining us. 516 00:24:04,720 - > 00:24:05,039 Mark Rotheram: Thank you. 517 00:24:05,039 - > 00:24:06,079 Cheers, thank you. 518 00:24:06,319 - > 00:24:08,240 Sinéad Hammond: If you want to catch up with other episodes, 519 00:24:08,240 - > 00:24:12,000 you can do so at bcn.co.uk or subscribe to the podcast in all 520 00:24:12,000 - > 00:24:13,119 the usual places. 521 00:24:13,119 - > 00:24:15,200 Thank you so much for listening, and we look forward 522 00:24:15,200 - > 00:24:17,119 to seeing you in the next episode.