Summary

Stan Boland’s central call: “We need to flood the UK with venture capital.” US VCs raised $76B last year; pro rata to population the UK should raise $15.4B and actually raised $3.7B — a ~$12B annual hole. His causality is the opposite of conventional wisdom: capital comes first and companies rise to meet it, with China as proof — from “pretty much nothing” 20 years ago to “clear global number two” by studying the US model and deploying capital at scale.
Harry’s live disagreement is the spine of the episode: he sees no shortage of money, only of founders — capital scarcity means the few great companies get bid to nuts prices (“5 on 30… then LightSpeed and General Catalyst come in and suddenly it’s 6 on 80”) and Europe’s slow, absentee funding experience loses founders anyway. Tom Hulme’s synthesis: don’t distribute capital evenly, concentrate it in potential global #1s — Wiz was assembled, heavily capitalized on day one, and sold for $32B (“like 7% of Israel’s GDP”) in five years.
The mechanism both back: a 10x’d British Business Bank fund-of-funds, never direct investing. Stan wants BBB from ~$424M/year to ~£4B on a 50/50 GP match, with creative fee/carry splits, treated as public-sector net worth rather than spending — £40B on the balance sheet in a decade (“no large fund of funds has ever lost money”; worst ~6% IRR, best mid-20s, always above gilts), then a “Thatcher moment” offering the stock to individual pensions.
Kill the zombie subsidies: Stan calls EIS/VCT funds “a freaking disaster” (all returns between 80 cents and $1.20 on the dollar) and R&D tax credits — $7.5B/year across 55,000 SMEs, ~2x the US rate — “classic helicopter money” propping up companies that limp on forever; redirect it into active venture that doubles down on winners and kills losers.
Where the UK can win: top and bottom of the stack, not middleware. Stan’s pick is fabless chip design — 75% of semiconductor value, where Europe holds an “insane” 2% share, with Bristol’s Inmos-descended full-custom capability. Tom’s caveat is brutal: the UK has the highest electricity costs in the Western world (“if the blended cost of training an LLM is 20% energy, we’re already kind of losing”), and defense is the sector where talent, buyers, and a nearby war align — Anduril’s $8B oversubscribed round shows the appetite.
Both are newly bullish China. Tom changed his mind: foundation models are “the fastest depreciating assets in human history — like weeks. It’s almost days now,” so value accrues to the application layer and to hardware, where China dominates manufacturing and devices become “the conduit for commoditized AI.” Stan adds that US geopolitical self-dislocation makes Europeans far more open to Chinese business than a year ago — “probably not good for the US.”
Tradeable quick-fire: on OpenAI at $300B Tom says yes (a consumer company at a $12B run rate, memory creating switching costs, ~20x forward), Stan says no (agents call APIs, “Claude are as good, if not better”), Harry would buy — and would “buy the shit out of Revolut” at $45–60B. Ten-year single stocks: Tom takes a uranium ETF, Stan takes Rolls-Royce (up ~3x this year, “the beginning of a journey” on defense engines). Most underinvested: hardware (Tom) and semis (Stan).
Jensen Huang, from 18 months inside NVIDIA: a good human but a control freak who rewrites product name, schedule, and pricing ten minutes before customer meetings, insulated by a buffer layer — “a human shield” — running a culture that’s “brutal… but not malevolent,” where he’ll “rip them to shreds and leave them whimpering in the corner.” On inference, Stan expects Jensen to keep re-architecting — “it wouldn’t surprise me” if NVIDIA reinvents as an in-memory compute company — with inference investment up 57x in a year.

Deep dive

1. Talent is the first rate limiter — staple a visa to the degree

Stan’s opening diagnosis: the UK is no longer the magnet it was. In AI it mints about as much talent as it keeps — losing to the US, recovering from the rest of Europe — “but we could be 10x better.” His fix for China’s and India’s graduates who arrive with no intent of staying: “you should have stapled to your graduation certificate a Tier 2 visa and rights to stay and a right to bring your family across.”
Tom’s refinement: founders “smash through walls” wherever they are — Melanie built Canva in Perth with “no right to build a $50 billion business” there — but “for every one good founder, you need five or 10 world-class operators, and I think that’s the biggest gap for us.” Oxford, Cambridge and Imperial — three of the world’s top ten universities — graduate only ~500 computer scientists and roboticists a year between them; “we should 5x that number.”
Tom’s Stripe lesson on metrics: the Collisons tracked the share of Series A companies transacting online that use Stripe (high-80s percent). Government should track the percentage of graduates who choose to stay — “great founders focus on leading indicators, not lagging.”

2. The $12B capital hole — and which way causality runs

Stan’s numbers: the US created $20 trillion of decacorn value in 50 years; the UK ~$170 billion — “two orders of magnitude off.” US VCs raised $76B last year; pro rata the UK should be $15.4B and raised $3.7B — “we’re short about 12 billion in venture capital.”
Harry’s flat disagreement, as a working investor: “there is simply not the supply of entrepreneurs if I were to keep my bar as high as it needs to be.” Stan’s rebuttal inverts the chicken-and-egg: “if we put capital in place here, great companies will rise to the occasion” — China proved it, going from nothing to “clear global number two” in 20 years while European AI investment is “diddly squat. You almost can’t see them.” His punchline: “the lack of capital crimps the ambition of companies, and therefore the best founders go to the States.”
Harry’s second wound: Europe’s funding product itself — weeks-long processes, absent partners — versus founders like Wordware’s likely Filip Kozera reporting the US was “super fast… they totally got me.” Stan’s answer stays the same: more capital lets the best VCs win, return, and ratchet the whole system’s performance up.

3. Concentrate capital in global winners — Wordware and Wiz

Tom’s caveat on “flooding”: by definition most seed companies shouldn’t be funded, and every rejected founder calls it a funding gap — evenly distributing capital “is damaging for talent concentration.” Stan agrees, and his own portfolio makes the case: Wordware, two Cambridge CS grads who could have raised “5 million on a 20 pre” in the UK, went to San Francisco and raised $30M on a $220M post — Harry notes those investors are underwriting a $10B outcome. “That stratospheric raising of expectations is part of the US playbook.”
Tom’s contrarian market view: “I actually think there’s probably gonna be less money in the market for venture in two years than there is today” — legacy funds raised in the past now face a much higher cost of capital and need ~20% IRR to break even — and “we’re over-fitting to history”: AI is creating a steeper power law than ever, so the job is a handful of global champions.
The template is Wiz: “$32 billion. That’s like 7% of Israel’s GDP… built in five years,” assembled without a clearly identified problem — a world-class team, heavily capitalized on day one. “The businesses we wanna build look more like Wiz… build a global business on day one.”
Stan’s corollary: “it’s almost pointless building a number three or number four in the marketplace today” — undercapitalized also-rans have no choice but to sell to US acquirers, so the UK never gets companies “that stand up on their own two feet.”

4. Rebuild the British Business Bank — 10x the fund-of-funds, never go direct

Unanimity on one point: government picking companies would be “an absolute disaster.” Tom’s reasoning: venture’s decade-long feedback loop is “like the worst learning loop ever” — “I don’t know if I’m any good at it still” — so new capital must be deployed by experts.
Harry, giving “not many shits anymore”: most of BBB’s fund-of-funds portfolio “is just dire… these funds should not be in existence,” and honestly only three to five UK players are very good. Stan’s counter-scaling: BBB’s ~$424M/year into fund-of-funds “needs to be 10x’ed” — 4B/year at a 50/50 match (a billion-dollar fund knows half comes from BBB), with creative structuring: pension funds pay a 0.5% fee for higher carry, BBB pays 3% for lower, “net-net we’re still at two plus 20.”
The fiscal unlock, per Stan: BBB investment counts as public-sector net worth, not current spending — do it for a decade and you hold £40B of fund assets (“no large fund of funds has ever lost money”; 6% IRR at worst, mid-20s at best, “always higher than gilt yields”). Then “a Thatcher moment”: offer the stock to individual pensions, because “people in their 20s and 30s should be owning assets in the future of the country.”
Both believe capital would import talent: US partners with track records would raise in London if the money existed, principals below partner level “would do a bloody good job” running new funds, and conditions could pull companies in — “we’ll fund it, but you’ve gotta move to London.” Tom adds the LP-side fix: aggregate the UK’s 90 subscale local pension funds into a world-class investment office that “can do something like Yale” — his fear being adverse selection, “the worst investors making the worst investments,” so UK PLC must engineer its way into the best funds.

5. Pick your battles: top and bottom of the stack, and the energy handicap

Stan’s stack map: Europe can win at the application layer (especially where European moats exist) and at the silicon layer — but not middleware, which is easier built in the States. The specific prize is fabless chip design, “where 75% of the value in the semiconductor space is… what NVIDIA is, what Qualcomm is, what Broadcom is” — Europe holds ~2% of that market (“it’s insane, honestly”), and Bristol’s Inmos-descended full-custom microprocessor capability is one of maybe two such pools in Europe.
Tom’s discipline: stop pretending “we can be experts at everything” — and face second-order facts, like “probably the highest electricity or energy costs in the whole of the Western world… if the blended cost of training a large language model is 20% energy, we’re already kind of losing.” Harry corroborates from a data-center CEO: “in the US my energy costs 4%. In the UK, if I set up today, it’s gonna be 17%.”
Honest geography, per Tom: “building a startup in Europe is doing it on ultra hard mode… if you do it in London, it’s slightly easier mode” — so lean into pockets of specialization rather than aggregating everything to the national level.

6. Defense is where the stars align

Tom (a reservist) gives three reasons defense is different now: the best people want in “because for the first time they actually think there’s existential threat”; the buyer isn’t monolithic — multiple services and regiments, each “being forced to innovate”; and “geopolitically, we are close to a war zone… and we have a point of view in that war.” Anduril’s $8B oversubscribed round shows the capital appetite; the UK’s role in Ukraine makes it “an amazing place to test new technologies” and “an opportunity to build next-generation primes.”
Stan sizes it: “probably 2 to 3 trillion going to be spent over the next five to eight years in Europe” across all layers, not just final product. Tom’s dual-use point: “you could argue DJI is one of” the biggest defense companies — cyber and UAV tech will flow back to civilian use.
The contrast case: digital health, “where the UK has one major customer, none else,” versus fintech, where the UK’s global position has earned GV a disproportionate number of investments.

7. The LSE is a supply problem — set a £4 trillion national goal with a ticker

Stan on the listing debate: only one London-listed tech company worth more than $10B — Sage, “a 30-year-old ERP company” — while the US minted $20.5T to the UK’s ~$100B. That’s not an exchange problem but a pipeline problem: companies get stunted or sold to US buyers before IPO scale. Tom adds a sentiment layer — “these stories are kind of like SEO for our minds” — plus the stamp-duty liquidity drag; and given the choice of HQ, employees, IP or listing venue, “I’m taking the first three.”
Stan’s national target: pro rata the UK should have created $4T and created $0.1T — so set “a national goal of creating 4 trillion of wealth in tech” over 20 years, half a trillion by year 10, requiring ~$100B of capital, ~$10B/year more than today — exactly the venture gap.
Against Harry’s Daily Mail nightmare — “your taxpayer dollars going to fund Tom or Sarah’s venture fund… and a Porsche” — Tom offers two cultural fixes: Sequoia names its meeting rooms after its LPs “to remind everyone who they’re in the service of,” and Norway’s sovereign wealth fund runs a literal public stock ticker (“the happiness of the country does go up and down depending on the ticker,” Harry confirms from interviewing its CEO). Stan: “if we have this 4 trillion goal, it’d be a great idea to have a national ticker as we climb our way towards it.”

8. Non-doms, the Treasury’s static models, and the multiplier effect

Tom the pragmatist on non-dom removal: “do I think everyone should pay equal tax? Yes, in principle, but practically speaking, I would rather that talent was in the UK” — the departures of angel-investing, job-creating friends are a leading indicator to take seriously. Stan: the cumulative shock — non-doms, inheritance tax, capital gains, private-school fees — “is too much… we are shooting ourselves in the foot,” and probably doesn’t “even make economic sense for the Treasury.”
Harry’s spiciest disclosure: an unnamed “most famous politician” told him “we have to get rid of the Treasury” because its models are static — raise tax rate X, get Y, with no behavioral variability. Stan: “Unbelievable… that’s not good.”
Tom’s multiplier evidence, as told: GoCardless — angel-backed partly by non-doms, it employed hundreds, one founder left to build Monzo, another became a VC, senior alumni spawned further companies. And the cycles are compressing: “previous, it might be five or 10 years… now it might be 18 months, 24 months.” Stan hedges the politics: “there’s bound to be some trickle-down economics, but there is also this need for fairness” — the balance has just “swung too far.”

9. Kill the zombie subsidies — and stop losing grads to quant funds

Stan on EIS/VCT funds: manager quality is low and the incentive is capital preservation — “all the returns are somewhere between 80 cents and $1.20 on the dollar. It’s ridiculous… those funds are a freaking disaster” — he’d abolish them (angel SEIS/EIS still makes sense).
His most unpopular position, flagged as such: R&D tax credits — “$7.5 billion a year… 55,000 companies,” ~2x the US rate, with no quality check — are “classic helicopter money” that “goes to companies that don’t need it or companies who shouldn’t have it,” keeping zombies limping year to year. Redirect it into active venture: “when things are going well, you double down. If things are not going well, you kill it.” Harry, initially pro-credits, concedes a taper by company age makes sense.
Stan’s four-point US-vs-UK summary: talent retention, early-stage mentoring quality, “the excess of props for companies that are not making it,” and the capital shortfall — “we need to flood the UK with venture capital.” Harry’s half-joking takeaway: “just put Stan in for the BBB lead.”
On Project Europe CEO Kitty’s claim that quant funds are “the biggest enemy of talent” — throwing £250K at grads — Stan is skeptical of scale, but Harry counters that ~1,000 people equals two full years of CS/robotics graduates. Tom’s lever: expand entrepreneur relief (cut from £10M to £1M) — even making entrepreneurs’ capital gains exempt — since quant comp is taxed as income.

10. Why now is the time to be bullish on China

Tom’s change of mind, stated as such: last time on the show he called foundation models “the fastest depreciating assets in human history, like weeks. It’s almost days now.” Post-DeepSeek, if models commoditize, value accrues at the application layer (GV’s Synthesia, Harvey) and in hardware — where “China is so much better than the rest of the world,” and devices become “the conduit for commoditized AI.” He’s moved from excitement about US foundation-model dominance to “we’re playing catch-up” on the hardware layer.
Stan’s overlay: color-code AI investment by geography and “it’s basically US and China and these tiny little dots of Europe” — plus geopolitics: “America trying to dislocate itself from the rest of the world will put China in a much better position… Europeans are gonna be much more open to working with Chinese companies than they were even a year ago. It’s probably not good for the US.”
His lived experience: he sold a company to Huawei and lasted about a month — “they didn’t give me authority to buy a box of pencils” — but at his chip company (likely Icera) the biggest customers were Huawei and ZTE, harder on price but “willing to engage.” His hedge stands: “the Chinese state is something different, obviously be wary of.”
The pushback isn’t smoothed over: China doesn’t let our companies in, and “every single piece of data that a Chinese company has, the Chinese government has authority to acquire at will”; Harry adds the 20–30% state subsidy of Chinese car production “destroying” the German car market via BYD. Stan concedes the rare-materials cornering but notes Germany’s carmakers “have their own challenges.”

11. Jensen lessons and the inference bet

Harry’s puzzle: if everyone sees the shift from training to inference, why isn’t NVIDIA cornered? Stan, who worked for Jensen for ~18 months: GPUs are being architecturally reworked for inference, and “it wouldn’t really surprise me” if Jensen reinvents NVIDIA as an in-memory compute company, organically or by acquisition — “he’s got very big ears and tracks what’s happening with enormous study.”
The management portrait, verbatim texture intact: “firstly, he’s a good human… he is, however, a bit of a control freak” — rewriting “product name, schedule, pricing and resources… with 10 minutes to spare” before customer meetings, insulated by a buffer layer of executives (“a human shield”). The culture is “brutal… in a way that is not malevolent”: he’ll “rip them to shreds and leave them whimpering in the corner to lick their wounds,” then forget it — “it’s not for everybody… but honestly, you’ve got to admit it’s worked.”
Stan’s silicon thesis beneath it: LLMs “are not the end of the story,” but large matrix-vector multiplies at high speed are a constant — and inference investment grew ~57x in the last year, with chain-of-thought and test-time compute pushing token generation up even at the edge.

12. Quick-fire: OpenAI at $300B, the one-person-unicorn myth, decade picks

Would you buy OpenAI at $300B? Stan says yes, though his reasoning is that agentic demand flows through APIs where “Claude are as good, if not better” — “it’s not obvious to me that the consumer chat interface is the winning interface. I’d probably put the money elsewhere.” Tom says no, despite describing OpenAI as a consumer company with a ~$12B run rate, memory creating switching costs, a million ChatGPT signups in an hour, and at ~20x forward “a reasonable place to put money” (“for my kids at school, LLMs are ChatGPT”). Harry would buy — and separately “would buy the shit out of Revolut” at the $45–60B marks people call pricey.
Tom’s contrarian belief: the first one-person billion-dollar business is “absolutely ridiculous.” Yes, Bolt.new hit a $40M run rate in three months — but “we’ve seen distillation of foundation models. We’re gonna start to see distillation of business models” — hyper-successful businesses get copied “ridiculously quickly,” so the solo-founder moat “is a myth.”
Ten-year buy-and-holds, both avoiding tech: Tom takes a uranium ETF (climate change is real, nuclear/SMRs are “the predictable cleanest energy source we have,” and an ETF avoids single-name risk); Stan takes Rolls-Royce — already ~3x this year, but “we’re at the beginning of a journey” on defense air-engine demand. Most underinvested areas: hardware for Tom (“you need it to be difficult to be valuable… contrarian and right”) and, one layer down, semis for Stan. Most-admired politicians: Lee Kuan Yew (“specialization”) for Tom; for Stan, nobody in the current government inspires enthusiasm, though Darren Jones “has got the potential to be great.”
Ten-year outlook: Stan expects the government to be forced into change (“things have not gone well”) and the UK to hit the $500B tech-value milestone and become Europe’s magnet. Tom the optimist: the decade’s most impactful companies will grow in the UK from adversity — as Salesforce did from 1999 and Airbnb/Uber from 2008 — “they won’t be names we know today,” they’ll be AI-native, and “it’s not clear to me they’ll list at all… some of our best portfolio companies, like Stripe, aren’t listing anytime soon.”

Vol.166 Macro Talk 82 | How Will Reciprocal Tariffs Reshape Global Trade?A Chinese Executive on Trump’s Tariffs and Companies in Mexico