Latest / AI in the UK / UK AI Hits the Frontier — And Corporation Tax to the Rescue?
Transcript
- 0:00All right, your luck. Listen up.
- 0:16Hello and welcome to AI in the UK.
- 0:18I'm Neil Footman. And I'm Matt Wilkinson.
- 0:21Professionally, we help organisations, leaders and
- 0:24individuals learn about and implement this big scary
- 0:27behemoth called AI. There's some amazing stuff going
- 0:32on with AI in the UK that doesn't always get the attention
- 0:36it deserves. With AI in the UK, we want to
- 0:39change that. Hello, a 5 month old company
- 0:59with an idea and no product has just raised more money than most
- 1:02FTSE 250 companies are worth and the UK government has a stake in
- 1:07it. Smaller than a rounding error.
- 1:09The man who founded Britain's AI Safety Institute is now paid by
- 1:13Anthropic and Microsoft, and he's calling for tax reform to
- 1:17help human workers compete with the AI those companies are
- 1:20building. And Britain's Sovereign Compute
- 1:22Strategy relies on a company whose flagship UK data centre is
- 1:27still a scaffolding yard. This is AI in the UK.
- 1:31And this week, the contradictions are of the story.
- 1:34Neil's over for you, and an ineffable moment.
- 1:38Indeed. I just want to say good to be
- 1:39back. By the way, our dedicated
- 1:41listeners will have noticed we didn't release an episode last
- 1:44week. We were both otherwise disposed,
- 1:46including Mike. You were in the States, weren't
- 1:48you? Yeah.
- 1:48I was in Tennessee and just a quick trip to Kentucky.
- 1:52Very fruitful, great experience and I very much enjoy the people
- 1:57of the South and put on a few pounds eating a few too many
- 2:01chicken and waffles. What a combination that is.
- 2:05OK, so moving on to our first story, the ineffable moment,
- 2:08Britain's bet on frontier AI. So Ineffable Intelligence, a
- 2:13London-based AI lab founded last November by former Deep Mind
- 2:17reinforcement learning lead David Silver, has raised $1.1
- 2:21billion at a 5.1 billion valuation, the largest seed
- 2:26round in European history, according to the company.
- 2:29You may recall, Matt, we did actually discuss this one
- 2:32previously when it was more of a rumour and it's now it's now
- 2:35come to fruition it seems. Sequoia and Lightspeed Co led
- 2:39the round, with NVIDIA, Google Index, DST Global and EQT
- 2:43participating. The UK government's Sovereign AI
- 2:47Fund and the British Business Bank also Co invested, with the
- 2:50BBB writing a £15,000,000 cheque.
- 2:53The sovereign AI amount is undisclosed, but the funders
- 2:56said its target range is between 1 and 10 million per investment.
- 3:01So David Silver, an intriguing character, he is a UCL professor
- 3:05and more importantly here perhaps he was central to Alpha
- 3:09Goal and A0, the deep mind systems that beat world
- 3:13champions at goal and chess by playing against themselves
- 3:17rather than studying human games.
- 3:19Just as an aside, have you, did you say you've seen that?
- 3:22That documentary is it's the thinking machine.
- 3:24Have you seen that right about Dennis Abbison deep mind that
- 3:28there's that whole sequence in it where they basically kicked
- 3:31off by having a game of Pong learn just by playing itself
- 3:35rather than being instructed ahead of time.
- 3:37So this is anyway. This is an extension of that.
- 3:39So Silver's thesis at Ineffable is a direct extension.
- 3:43Skip pre training on human data altogether and build what he
- 3:47calls a super learner, an AI that discovers knowledge purely
- 3:51from its own experience. The bet is that reinforcement
- 3:54learning applied at sufficient scale produces something
- 3:57fundamentally more powerful than today's large language models.
- 4:02The company is five months old. As yet it has no product, no
- 4:06revenue and no published road map.
- 4:08So the $1.1 billion is essentially a bet on silver and
- 4:13a thesis for the UK. Ineffable matters in two ways.
- 4:18It's the first British founded company since DeepMind itself
- 4:21with a credible claim to be in the frontier AI conversation,
- 4:25judged by who's writing the cheques.
- 4:27And it's the second direct investment from the sovereign AI
- 4:30fund in two months after AI infrastructure startup Callosum
- 4:34in April, a pace that's unusual for any government backed
- 4:38vehicle and the strongest early signal yet of how the fund
- 4:42intends to operate. As always, so so much going on
- 4:46here. This, this story was being
- 4:49bandied around a lot on LinkedIn, people getting quite
- 4:51excited about this and you can understand why this is a serious
- 4:54investment involving serious companies, serious investors.
- 4:59It's also now clearly we we don't deliver this point.
- 5:02The amounts of money invested by the British Business Bank and
- 5:05Sovereign Fund are smaller but significant for other reasons
- 5:08that they're actually British Business Bank has a mandate to
- 5:12invest in companies like this now and the Sovereign AI fund is
- 5:16now up and running and has appears to be from from early
- 5:19indications quite nimble in its operations.
- 5:22So to talk about the company itself ineffable this thesis
- 5:26that his idea is a fascinating one.
- 5:29I would say I'm not going to pretend to understand the
- 5:32technical side in tremendous detail, but we do know that
- 5:35increasing doubts have been raised about whether large
- 5:39language models are going to be able to make the leap to
- 5:43artificial general intelligence on their own.
- 5:45I think that was more of a fringe position not long ago and
- 5:47now it's relatively mainstream. But this specific idea, So
- 5:52Silver, so we should say about him, he's not a fringe figure.
- 5:55He built, he's one of the people who built modern reinforcement
- 5:59learning, and he argued in a paper last year that the
- 6:04knowledge extractable from human data in maths, coding and
- 6:07science is approaching a limit and that wall isn't theoretical
- 6:11anymore. You'd have to say that the omens
- 6:16are good. Sequoia flew to London to court
- 6:18him and NVIDIA put in a record 1/4 of a billion dollars.
- 6:23I mean, this seems to me like a pretty serious endorsement of
- 6:28Silver's thesis. How do you feel?
- 6:31Yeah, Silver is obviously an impressive, impressive mind.
- 6:36And, you know, it's achieved some fantastic things.
- 6:38The only caution I really have here is that, you know, a lot of
- 6:42his wins using reinforcement learning are in those domains
- 6:45with perfect rules and clear reward signals.
- 6:48Go chess, protein folding. There are real scoreboards that
- 6:52you can test against. You know, real world discovery
- 6:55doesn't have that same scoreboard and the simulation to
- 6:59real world transfer problem is the known weakness of this
- 7:02approach. Silver and Sutton's own paper
- 7:04that you just mentioned concedes that physical world experience
- 7:07is constrained by how long actions take to play out.
- 7:10Meanwhile, the large language model paradigm he's betting
- 7:12against keeps improving. GPT 5.5 Claude's recent
- 7:17reasoning gains with mythos from adding.
- 7:19You know, and they, and those gains, a lot of those game gains
- 7:22come from adding reinforcement learning on top of the language
- 7:25models, not abandoning the language models themselves.
- 7:28So it feels like maybe this is going to go somewhere.
- 7:30I don't know, but it's it's a big bet, but it's exciting
- 7:35nonetheless. It is a big bet.
- 7:36I mean, I feel very enthusiastic.
- 7:38That point you raised, though, is one that has occurred to me
- 7:40as well. As if if these are coming from
- 7:42essentially a standing star and they're not being based on
- 7:45existing training models, then it does seem like a big gap to
- 7:50make up. And as far as I understand these
- 7:52things, you would assume silver is obviously very well aware of
- 7:56that as an issue and, and would be able to respond to it in some
- 7:59way. I, I need to dig into the
- 8:00technology more. But having a bit of that
- 8:04magnitude is, I mean, it's, it's not, we'd say it's not an
- 8:08unusual amount of money to be ploughing into very promising AI
- 8:13startups. It is an unusually big amount
- 8:15here. So 1.1 billion for a 5 month old
- 8:19company with no product. Now, we've kind of seen this
- 8:21before with Mira Murati when she she, you know, previously of
- 8:26Open AI and she received funding running into billions for a
- 8:30startup that similarly had no clear road map.
- 8:33I mean, from my, from understanding, this has a
- 8:35clearer, clearer road map than hers did.
- 8:38I remember reading at the time that it wasn't entirely clear
- 8:41what what the company was going to do.
- 8:42I'd say at least this one does have a clear thesis.
- 8:46All that said, a billion is still a huge amount of money.
- 8:49Now, what I would say about this is in any other sector, this
- 8:52would be evidence, I think, of a very clear bubble.
- 8:57But what we're dealing with here is a sector where just to get to
- 9:02the table requires such a huge amount of money for all sorts
- 9:06for, for getting the best talent, for getting the compute
- 9:09power, and critically, that it's reserving the power that they
- 9:13need in order to actually make a goal of this.
- 9:15So it's kind of a, it's not just a bet on the company, it's it's
- 9:19like a, an advance on money they'll require for that, for
- 9:22gaining the, the compute and the processing power they need.
- 9:25So I to me, it doesn't sound ridiculous.
- 9:30It sounds like sort of the the ballpark figure that companies
- 9:34of this stature and this ambition required just to get a
- 9:37seat to the table. Yeah, even by frontier AI
- 9:40standards, this is at the outer edge.
- 9:42You know, 5.1 billion valuation and a five month old company
- 9:46with no team disclosed no product yet.
- 9:48No, no disclosed road map. Of course that doesn't mean to
- 9:52say that he hasn't disclosed that to the investors yet.
- 9:55It's it's a very interesting investment in everything that we
- 9:58talk about here. We're having to infer from a
- 10:00category, not from anything ineffable as detailed and even
- 10:03in the name, it's kind of, you know, it talks to sort of just
- 10:06how ineffable this is. This does seem to be a bet on
- 10:10silver's track record. And that's a lot of weight for
- 10:12for one person to carry. Realistically, 1.1 billion is
- 10:15about what it takes to start of Upper Frontier Lab.
- 10:18You know that the barrier to entry is so high that only a
- 10:21handful will ever exist at this level.
- 10:24And so, yeah, it's, it's, it's interesting, it's exciting.
- 10:28But just looking at the costs to, you know, to, to put it up,
- 10:33put together a training run, you know, being about 100 million,
- 10:37it's, it's not going to go that far.
- 10:40No, indeed. I mean, well, the other, the
- 10:42other part of this story is where all that investment comes
- 10:45from. So clearly the, the, the lion's
- 10:46share of it is coming from the kind of big names, big US names
- 10:50you'd expect that typically enter into funding runs on this
- 10:53Sequoia NVIDIA. I think what also makes it
- 10:55interesting though, certainly for me is the involvement of the
- 10:58Sovereign AI Fund, which has been getting a lot of attention,
- 11:02certainly in my LinkedIn rabbit holes and there's a lot of
- 11:05people talking about and very excited about it, I think.
- 11:08So when we discussed it before, we were a little sceptical about
- 11:11the amounts of money involved and whether it would be able to
- 11:14operate in a way that is conducive to success in that
- 11:18world. Now I have to say, I mentioned
- 11:20to you just before we start recording us, I heard an
- 11:22interview with the UKAI minister Kanishka Narayan on the Rest is
- 11:28money. And I have to say he made a very
- 11:31convincing case to me. I mean, he was essentially
- 11:33saying that no, that that these amounts of money are not, we
- 11:37can't go straight in with, with the amounts of money that that
- 11:39NVIDIA has access to. This sovereign AI has been very
- 11:42deliberately structured to be effective in this, to be
- 11:46effective in this area in the VC world.
- 11:49It's not going to be slowed down by endless bureaucratic process
- 11:53in review, he said. It's designed to give very rapid
- 11:56responses when when investments required, which is apparently
- 11:59one of many key complaints about for UK companies attempting to
- 12:03get significant investment in the UK.
- 12:06You just look at the people involved.
- 12:07This is not headed up by senior civil servants.
- 12:11The involvement is James Wise from Balderton Capital, who's a
- 12:15chair and Josephine Kant is running ventures.
- 12:20These are these are serious individuals in this area and
- 12:22they, they know what they're doing.
- 12:24Yes, the amounts of money are not going to be of the same
- 12:28magnitude. I think if, if Can you imagine
- 12:31the response in the UK if it was revealed that the government
- 12:35was, was our government funded investment body was putting 10
- 12:38billion into something like this or whatever it is, you know,
- 12:41quarter of a billion, half a billion, it would be, there's
- 12:44not the stomach for that, but. Nice some of that months ago.
- 12:46Fix the potholes. Well, indeed, but the
- 12:49involvement of the government connected fund, it brings in all
- 12:54the other things to play like, you know, the access to the
- 12:56priority for government procurement, all that kind of
- 12:59stuff. So, so for me, I, I think it
- 13:02does, I mean at the time will tell, but I mean the early, the
- 13:04early signs for me are good. Yeah, I, I mean, it's, it's hard
- 13:11to sort of sort of feel that this is a muscle of the state
- 13:14rhetoric. You know, the sort of and agree
- 13:16with that. I mean, it's, you know, the, the
- 13:18two bits of funding, you know, the sovereign AI check as well
- 13:21as BB BS 15,000,000. It the whole lot's going to be
- 13:26worth less than 3% of the 1.1 billion round.
- 13:30It feels to me a bit like that the the investment is more that
- 13:35the funds. Real value is soft power.
- 13:38You know, how does it, you know, how does it work with
- 13:40relationships, future deal or flow, you know, options, you
- 13:44know, options in the next round, you know, really about being
- 13:47able to maybe grease the wheels behind the scenes to help make
- 13:50things happen. You know, we just have to
- 13:53question what's what's that soft power worth?
- 13:56You know when Sequoia or NVIDIA are the check writers.
- 14:00But it's it's interesting for sure.
- 14:02I think it's, yeah, You know, I think it's not to be
- 14:05underestimated. Sure, money talks, but that that
- 14:08kind of soft power and and the access to government I think is
- 14:12very valuable as well. Speaking of government or
- 14:15government adjacent. And we'll move on to our next
- 14:17story, which has got a very healthy dollop of politics
- 14:20involved in it. Matt what what's, what's that
- 14:23one about? Yeah, so our former Prime
- 14:26Minister Rishi Sunak has become a paid advisor to both Anthropic
- 14:32and Microsoft. And he used a BBC Newsnight
- 14:35interview to call for the gradual abolition of employer
- 14:38National Insurance and replace it with taxes on corporate
- 14:42profits, which he argues will be swollen by AI productivity
- 14:46gains. His argument is a simple
- 14:49asymmetry. When a business hires a person,
- 14:51it pays 15% employer National Insurance on earnings above 5000
- 14:55lbs. When it deploys an AI agent to
- 14:58do equivalent work, it pays nothing comparable.
- 15:00That gaps in access gives firms an immediate financial incentive
- 15:05to choose software over people. You know, he told Newsnight
- 15:08presenter Faisal Islam. That CEO is privately
- 15:11describing, you know, flat is the new up businesses growing
- 15:15revenue without growing headcount and that the effect is
- 15:19most visible in law accountancy in the creative industries.
- 15:23The World Economic Forum reported in March that US entry
- 15:26level job postings had fallen 35% in 18 months, citing AI as a
- 15:32primary driver. The British Chambers of Commerce
- 15:34flagged a similar pattern in the UK.
- 15:37Sunak acknowledged the concerns of graduate seeking entry level
- 15:40work are justified. The the interview marks perhaps
- 15:43the first time a major UK political figure has explicitly
- 15:47framed the tax system as creating a structural
- 15:50disadvantage for human workers against AI workers.
- 15:53Sunak also referenced his collaboration with Deputy Prime
- 15:56Minister David Lammy. They appeared together at the UK
- 16:00India Shaping Tomorrow AI Showcase as evidence of cross
- 16:03party consensus on attracting AI investment even as the parties
- 16:07disagree on almost everything else.
- 16:10He praised the AI Security Institute, which he established
- 16:13in its original form as the AI Safety Institute in 2023 amid
- 16:18speculation about its future funding under Labour.
- 16:21The proposals fiscal logic runs in One Direction until it
- 16:24doesn't. If AI systems replace workers
- 16:27and carry no equivalent employer National Insurance burden VAT
- 16:31liability equivalent the revenue gap widens.
- 16:35Sunak's answer is to shift the base to corporate profits.
- 16:39What he doesn't address is the transition and who pays if
- 16:42profits take time to materialise at the scale needed.
- 16:45Yeah, it's it's really interesting that he's that he's
- 16:49raising this and it does seem to be that he might have a few
- 16:55conflicts of interest here. You know, the his argument does
- 16:58has to have to stand or fail on its own merits merits.
- 17:01And structurally it seems pretty sound.
- 17:03You know, employer National Insurance attacks on the act of
- 17:06hiring a person and we've seen some of the impacts of
- 17:09increasing that you know, AI agents carry no equivalent.
- 17:13If you want to slow the substitution effect of, you
- 17:16know, reducing the disadvantages of humans is a logical
- 17:19intervention. The framing of this as a pro
- 17:22worker policy, not just a tax cut argument, is actually more
- 17:26sophisticated than most of what UK politicians have said about
- 17:29AI and labour. What are your thoughts, Nails?
- 17:33I'm not, so I'm, I don't know what I was thinking a bit about
- 17:36the conflict of interest. I mean, what, what do you feel
- 17:39the conflict of interest might be?
- 17:41Effectively? And I know, yeah, he's he's now
- 17:43representing Anthropic and Microsoft in some capacity.
- 17:49But as I understand it, I mean, this argument would basically,
- 17:56if, if followed through, would result in higher levels of
- 17:59corporation tax, which is presumably not something
- 18:02Anthropic and Microsoft are pushing him to advocate for.
- 18:06Is that I'm understanding this correctly.
- 18:09It would. It would abolish National
- 18:12Insurance as an incentive employer, National Insurance as
- 18:14an incentive to make employment cheaper and funnel that over to
- 18:19income tax. Sorry, not income tax,
- 18:21corporation tax. It would increase corporation
- 18:23taxes to generate revenue. That's my understanding is that.
- 18:28Yeah, and I think that there's a, there is a subtlety in this.
- 18:31If we look at the future, we can imagine a future where AI agents
- 18:36start replacing jobs and all of a sudden we have a, a larger
- 18:40group of people out of work disenfranchised.
- 18:43There will be clash now. So I, I, I think one of the
- 18:47things that's, that's really crucial here is that we have to
- 18:50try to make sure that we're keeping people in work.
- 18:52We know that in the recent increases to National Insurance
- 18:56or employer, National Insurance actually paused hiring in a
- 18:59number of industries, particularly those where the
- 19:02labour was less well paid. So that that National Insurance
- 19:05was a bigger hit to to the wage bill than shall we say, you
- 19:09know, a high paid consultancy where actually that is a tiny
- 19:12fraction of the total salary. So we have to be very careful to
- 19:16actually make sure that we're incentivizing the productive use
- 19:20of human capital as well as when we we start moving into
- 19:23deploying AI. Yeah, yeah, I remember here in a
- 19:27particularly cynical take after the first Rachel Reeves first
- 19:31budget, when she jacked up employers National Insurance,
- 19:35some commentators suggested that she was intentionally trying to
- 19:39make human labour less attractive in order to bump up
- 19:42productivity figures at companies who had been reluctant
- 19:45to employ technology. I'll assume if if that was ever
- 19:49true, and who knows, I'll assume that day has passed.
- 19:52I mean, look, I I think it's great that someone of Rishi
- 19:57Sunak's stature is talking seriously with actual policy
- 20:01proposals. I think this has been very much
- 20:03lacking in the discourse in the UK.
- 20:06This suggestion makes sense to me on several levels.
- 20:11It would, at a stroke make employing people more attractive
- 20:16than employing AI. There are certain sectors where
- 20:18that could make a very quick difference.
- 20:20I mean, you sort of referred to. Hospitality say that would make
- 20:26an immediate and and meaningful difference to employment figures
- 20:29in in those industries and and perhaps in retail say as well.
- 20:33I think the so the short term measure and medium term measure,
- 20:36very sensible. You'd have to say as a longer
- 20:37term measure, it's probably something of a stick in plaster
- 20:41because these models and these tools are just getting so much
- 20:45better so quickly that even the additional incentive of, you
- 20:51know, a 15% reduction in costs on human labour is probably
- 20:55going to be attractive for a given period of time.
- 20:59And then less so because, you know, output from some of these
- 21:02things, as we know, hits a kind of a marginal cost of 0 and
- 21:06insert certain elements. But overall, I am very much for
- 21:11this. I think it's, I think it's
- 21:12really positive that someone of his stature is addressing this
- 21:14seriously with serious policy proposals.
- 21:18Yeah, I agree. I think there's two things
- 21:20really that we've seen already that certain AI companies, I
- 21:25think Anthropic, are starting to increase that cost of the token,
- 21:28the way they manage, the way they charge for AI.
- 21:32Anyway, we know that we're probably coming to close towards
- 21:36the end of this cheap token, the free token, if you will.
- 21:38If we can reduce cost to employ humans, we hopefully we can slow
- 21:42down the change long enough that the the people can retrain and
- 21:45find new jobs. I think that's what we really
- 21:47need to be looking at. How do we manage that transition
- 21:50to whatever the future is going to look like?
- 21:53Because none of us can really hold a crystal ball out up and
- 21:57accurately say how and when this is going to happen.
- 21:59But those these models are getting so, you know, so
- 22:03impressive so quickly and trying to manage that transition is
- 22:05going to be really crucial. Yeah, and it's just as a final
- 22:09note on this, this this broader discussion around how to
- 22:12structure taxation and and how to structure government is
- 22:16finally, it feels like it's just starting to gain about momentum.
- 22:19I mentioned an interview with Tom Blofield, the the founder of
- 22:23Monzo in our our last podcast. He had some really interesting
- 22:25stuff to say. I mean, he, he's very much of
- 22:28the, the opinion that this is all going to be transformative
- 22:31in the short to medium term. He was saying he thinks income
- 22:34tax will be abolished within five years.
- 22:37Is that is a bold prediction, but who's who's to say is wrong?
- 22:40And actually some of his other ideas about how this could be
- 22:44addressed and, and feasible future funding models for here
- 22:47in the UK is something that comes up in our next story.
- 22:50Well, in the discussion around that, fortunately enough.
- 22:53Moving on to story #3, so this is yet more news about data
- 22:59centres in infrastructure. So BT and N Scale have announced
- 23:03a big data centre push. BT and N Scale announced on
- 23:08April 23rd that they will build sovereign AI dentist data
- 23:11centres at 3 existing BT sites across the UK.
- 23:15N Scale will build up to 14 megawatts of capacity powered by
- 23:19NVIDIA hardware. BT is providing the estate and
- 23:23the network. The three sites have not yet
- 23:25been disclosed. 14 megawatts is a modest deployment by
- 23:29hyperscaler standards and a long way from the 31,000 GPU's
- 23:34originally promised under a separate N scale partnership.
- 23:36Announced last September, the deal, known as Stargate UK, was
- 23:41a joint project with Open AI and NVIDIA.
- 23:44On April 9th, Open AI paused it. The company cited high
- 23:49industrial energy costs and the UK regulatory environment,
- 23:53including all the stuff about the copyright rules.
- 23:56After backlash from the creative industries, UK industrial
- 24:00electricity prices remain among the highest in Europe.
- 24:04The Lauten question also remains open.
- 24:07The Guardian reported in March that Nskill's flagship UK site
- 24:10near Loughton in Essex was still operating as a scaffolding yard
- 24:15with no construction started. The wider question why this
- 24:20matters fiscally, was put into sharp focus last month by Tom
- 24:23Blofield, the Monzo Co founder. On the Rest Is Money podcast.
- 24:28Blomfield argued that as AI displaces white collar work,
- 24:31governments will need to shift the tax base from labour to
- 24:34compute. Don't think we'll tax human.
- 24:38Labour will tax compute meaning systems like data centres and
- 24:42then we will use the proceeds to pay for government.
- 24:45For context, income tax and National Insurance combined
- 24:49raised over 550 billion in 202526, close to 60% of total UK
- 24:56tax receipts. So that dovetails quite nicely
- 24:59with our previous stories. You can see he's thinking big
- 25:03Tom Blofield, I mean he, he's involved in AI, think it's Y
- 25:07Combinator or similar on the West Coast of the States now.
- 25:10And he says the technology sees just reinforces this belief that
- 25:14in the short to medium term, there's going to be a huge
- 25:17disruption, particularly in the knowledge work sector that will
- 25:21potentially devastate tax income.
- 25:24And the way to do it is to move away from taxing income and to
- 25:28move towards tax increasing corporation tax.
- 25:32Now what he said was that in the case of AI that that could mean
- 25:35having more data centres here and actually taxing the compute.
- 25:38But we can move on to that in a moment.
- 25:41On the question of the data centres themselves, what did you
- 25:45feel when you read this story? Did you feel like just we're
- 25:47going over same old ground here and do you believe it when you
- 25:50see it or you thought this is actually something that has real
- 25:54promise and could happen? Well, for me, this looks a bit
- 25:57like BT trying to reinvent itself.
- 26:02We know that BT has been a bit of struggling a little bit.
- 26:04You know, it was, it was known as British Telecom.
- 26:06You know, it was the used to advertise people.
- 26:08We are picking up the phone. And so, you know, these days
- 26:11everything's moved to mobile. They've they've had a bit of a
- 26:13struggle, I believe with their acquisition of a, you know,
- 26:16mobile operator. BT's kind of lost its way a
- 26:18little bit. In some ways this could really
- 26:20help put it back on the map. I mean, it's not doing quite
- 26:23what all birds AUS shoe company did, which is, you know, they
- 26:26were making shoes and now they've shifted over to being an
- 26:29AI data centre company and in doing so increased their
- 26:34increase the share price by about 500%.
- 26:36So that, you know, these, these are funny shifts.
- 26:38But I, I do believe that by BT trying to become an integrated
- 26:42part of the UK's compute, it actually, it actually helps it
- 26:47to, you know, to make better use of the, of the, of the, you
- 26:51know, the space it already has and actually deploy at scale and
- 26:55it already has the connectivity in place.
- 26:58So it makes it, it does make a lot of sense.
- 27:00Yes, it's small compared with, you know, the, you know, with,
- 27:04with the the Stargate project. But this, I think this is what
- 27:09credible UK sovereign compute looks like.
- 27:12It's not vapour Ware backed by AUS hyperscaler that can just
- 27:15pull out at a moment's notice. This is anchored in the UK
- 27:18economy with a company that's anchored here and and needs this
- 27:22to. Succeed.
- 27:23Yes, yes, To your final point, I mean, I think that just the the
- 27:27mere fact of it being a British domicile company BT that has
- 27:34such a such a significant part of its business here surely
- 27:37gives it greater skin in the game for the success of
- 27:39something like this than perhaps for an open AI for whom even a
- 27:44project of that scale is just one of many.
- 27:47So yes, that's good. You know why not?
- 27:50I mean, I think this this is a logical step for BT They're not
- 27:54a shoes company. They're, you know, they have
- 27:56been involved in, they've been the at the forefront of
- 28:00telecommunications in Britain for decades.
- 28:02I would say that to some extent, you know, I believe it when I
- 28:05see it. We have had several stories like
- 28:07this that have kind of come to nothing.
- 28:10I hand on heart, hope and scale does succeed with all its
- 28:14planning to do. It's, it's track record thus far
- 28:18has been mixed, but they seem to have, but I, I only know about N
- 28:24scale, what I read about in interviews and, and, and and and
- 28:27articles and so on. They seem to have a credible
- 28:30team behind them. There are some question marks of
- 28:32the CEO's experience, but he seems to be a serious
- 28:36individual. And, and they do seem to be
- 28:37committed to doing this. Committed to doing it and
- 28:40actually doing it are two different things.
- 28:42So I, yeah, bottom line, I guess I'll believe it when I see it.
- 28:47So how about this whole tax position, because this this sort
- 28:51of ties into what we were just talking about with with Rishi
- 28:54Sunak. What do you think of Tom
- 28:55Bluffield's position? Yeah, I don't think this is just
- 28:58a podcast sound bite that, you know, from Bloomfield.
- 29:01And AI published a 13 page policy document, you know,
- 29:04earlier in April making essentially the same argument
- 29:07that we will need to shift the back tax base from labour and
- 29:10payroll to capital and automated labour.
- 29:12Because if AI compresses the wage base hard enough, the
- 29:15existing, you know, the existing financial model breaks.
- 29:19Now that's a frontier AI company arguing against its own short
- 29:23term commercial interests. And Bloomfield's specific
- 29:26phrasing, data centres, inference tokens, whatever you
- 29:29want to call it, matters. He's not just talking about
- 29:32taxing the building, he's gesturing at taxing the metres
- 29:36unit of computational work which travels with the workload to the
- 29:40UK customer the way VAT does on imported digital services.
- 29:44Crucially, this connects directly to the sovereignty pipe
- 29:46point. If the compute happens on AUK
- 29:49based BT site under UK jurisdiction, the legal
- 29:53sovereignty you know that's argued for becomes the very
- 29:56thing that makes the tax enforceable and the mobility
- 29:59objection partly dissolves. Yes, yes, it does.
- 30:04That's essentially touches on what we were saying before that
- 30:06BT has considerably more skin in the game that they are UK based
- 30:09company and so on. I think I'd have to say again, I
- 30:13mean I think it's it's positive that we are now talking about
- 30:17these with actual proposals because it's easy to bang on
- 30:20about this is going to this is going to devastate jobs or it's
- 30:23not going to devastate jobs because technological upheavals
- 30:26always produce new jobs. Whatever your position is.
- 30:29It's all been very vague. This is pointed in the right
- 30:32direction for concrete proposals, and there is a
- 30:35definite logic to what Tom Bloomfield says.
- 30:39But the scale is daunting, daunting task to undertake.
- 30:45We mentioned a moment ago that income tax and National
- 30:49Insurance together run to almost 500 billion.
- 30:53Now if he's saying the income tax itself is going to be
- 30:57abolished in five years, that is a huge amount of money to make
- 31:03up from other sources in the meantime, even getting close to
- 31:07that. And I would say the the thesis
- 31:11of generating significant tax receipts from compute within a
- 31:15country is I don't know, is is perhaps still not entirely
- 31:19proven. It absolutely makes sense and
- 31:22it's also really encouraging. I read some of that report from
- 31:25Open AI that the cup that the big AI companies themselves are
- 31:28realising that they have to get in this conversation.
- 31:30I mean, to be honest, partly because the conversation has
- 31:34been a bit crap from from other directions.
- 31:36It's, it's there, there has been a bit of a lack of clear
- 31:39government proposals on how to deal with this.
- 31:43So I applaud the idea. I applaud the sentiment.
- 31:47It is a huge, huge task they've got to overcome.
- 31:51So this is just got to be the beginning.
- 31:52There's going to have to be people much smarter than me
- 31:57thinking of other ways that that we're going to address this and
- 31:59offset any potential big losses in income tax that could arise
- 32:04as a result of the big shift in the jobs market.
- 32:07I agree. And I and I do fear for where
- 32:09those jobs will go. I think the difference between
- 32:12this and many of the other shifts is just how quickly the
- 32:18technology is transforming what's possible.
- 32:21And so while I'm pretty confident that given time, we'll
- 32:26be able to find new jobs, we'll be able to do more, you know,
- 32:31humans will always find a way and we will hopefully we'll, you
- 32:34know, we'll end up in a, you know, a Star Trek like future
- 32:37rather than the Battlestar Galactica one or a Terminator 1.
- 32:40You know, So we've got to, you know, we've got to look at sort
- 32:42of where the future takes us. But I do feel there could well
- 32:45be sort of this, you know, we could end up in that sort of
- 32:49depression moment where there's not enough, you know, there
- 32:52isn't enough money in the economy because there aren't
- 32:54enough jobs because everybody shifts over.
- 32:56That's it. That's a real concern.
- 32:58I think one of the things that that you mentioned earlier was
- 33:01sort of taxing on compute the tokens is really important
- 33:04because I don't believe that sovereignty has to mean that the
- 33:09data centre is actually built on physical land that you own.
- 33:13If you look at Elon Musk's vision and to be, you know,
- 33:16there seems to be a sentiment growing around this and we could
- 33:19be putting data centres up into space where where we can power
- 33:23these by using solar power. You know, we don't have the same
- 33:26issues in terms of heat restrictions.
- 33:29So there could be quite an interesting shift from wanting
- 33:32to have ground based data centres versus actually having
- 33:35them in in orbit. And so thinking about where this
- 33:38could go is going to be is a really interesting thought
- 33:41experiment and one that really needs to happen by, as you say,
- 33:45very, very smart brains. Yeah, yeah.
- 33:48You can always do with more of them in my experience.
- 33:50Well, that would be a good place to wrap up for this week.
- 33:53I think we've got through an awful lot there.
- 33:55Thanks as ever for talking through the issues with me this
- 33:58week, Matt. Yeah, it's been a pleasure,
- 34:00Neil. And we will see you all next
- 34:02week. Look forward to it, take care.
- 34:05Cheers. And that's it for this week's
- 34:23episode of AI in the UK. If you found it useful, you can
- 34:28subscribe wherever you get your podcasts and we'd really
- 34:32appreciate a great rating or review.
- 34:34It genuinely helps others find the show.
- 34:38You can also follow Neil's Footman and Matt Wilkinson on
- 34:41LinkedIn for links, notes and occasional extra content.
- 34:46Thanks for listening and we'll be back next week.
- 34:49Helping you make sense of AI as it lands here in the UK.