Tom Hulme & Stan Boland: Lessons from Jensen Huang & How to Fix the UK Tech Ecosystem
Summary
Britain’s deepest startup constraint is not merely founder supply but the five to 10 world-class operators required behind every exceptional founder. Oxford, Cambridge and Imperial together graduate only about 500 computer scientists or roboticists annually; Tom argues that output should rise 5x, while Stan would staple a Tier 2 visa and family rights to every relevant graduation certificate. The investable bottleneck is an ecosystem failing to retain and compound scarce technical talent.
Stan argues that Britain is roughly $12 billion a year short of the venture funding its population should support. US funds raised about $76 billion last year; the comparable UK figure would be $15.4 billion, versus only $3.7 billion actually raised. Tom emphasizes the shortage of investable founders and operators, while Harry argues that capital already crowds into excellent teams, inflating prices and weakening terms; all three point to slow European decision-making and a poor founder experience.
The proposed answer is concentrated capital for prospective global champions, not indiscriminate startup funding. Cambridge-founded Wordware might have raised $5 million at a $20 million pre-money valuation in Britain, but moved to San Francisco and raised $30 million at a $220 million post-money valuation—capital that raises expectations toward a multibillion-dollar outcome. The discussion’s shared call is that building market number three or four merely manufactures acquisition targets; Britain needs companies capable of becoming global number one or two.
A scaled British Business Bank could anchor roughly $4 billion of annual fund commitments and require private investors to match them 50/50. Aatish proposes flexible economics—pensions might pay a 0.5% fee for less carry while BBB pays 3% for more—so the blended fund still works around “2 plus 20.” Over 10 years, government could accumulate $40 billion of venture fund assets, although Tom warns that power-law returns make adverse selection existential: taxpayer money must reach the best managers, not “the worst investors making the worst investments.”
Britain should specialize where it has an unfair advantage: AI applications, fabless semiconductors, hardware, fintech and potentially defense. Semiconductor design captures roughly 75% of sector value, yet Europe has about 2% of that market; Bristol retains unusual full-custom microprocessor expertise. Conversely, UK energy can represent 17% of data-center costs versus 4% in the US, making indiscriminate infrastructure investment a losing proposition.
The tax system may be preserving weak companies while exporting the people who create compounding networks. Stan calls the roughly $7.5 billion of annual R&D credits spread across 55,000 companies “classic helicopter money” and would redirect much of it into actively managed venture; Harry suggests tapering credits to stop decade-old recipients becoming zombies. Both treat non-dom departures as a serious leading indicator because wealthy residents also employ people, angel-invest and seed alumni networks such as GoCardless and Monzo.
AI commoditization shifts the debate from foundation-model supremacy toward applications, brands, hardware and inference silicon. Stan says model depreciation has accelerated from “weeks” to “almost days,” strengthening China’s manufacturing position and companies such as Synthesia, Harvey, BYD and DJI. Tom would buy OpenAI at $300 billion because its roughly $12 billion run rate, memory and consumer brand create emerging switching costs; Harry would put the money elsewhere because agents may consume models through APIs, where Claude can be “as good if not better.”
The 10-year bull case is a Britain that becomes Europe’s talent magnet and reaches roughly $500 billion of tech value, though public listings may no longer be the right scoreboard. Stan’s larger national ambition is $4 trillion over 20 years, supported by about $100 billion of additional capital and a public “national ticker.” Tom expects the defining AI-native companies to emerge from today’s adversity but cautions that, like Stripe, they may provide private liquidity rather than list on the LSE.
Deep dive
1. Britain’s rate limiter is the operator bench behind each founder
Stan grounds his policy prescriptions in operating experience: he helped take ARM public, raised about $330 million across several companies and sold businesses for roughly $1.3 billion, including exits to Broadcom and NVIDIA. Tom helped establish GV Europe in 2014; it has backed more than 50 companies across 12 countries and invested over $500 million in Britain.
Britain is roughly retaining as much AI talent as it produces, Stan says, but only because losses to the US are offset by arrivals from elsewhere in Europe. That flat net figure disguises the opportunity: the UK “could be 10x better” by making itself Europe’s default place to form a company.
Tom’s sharper diagnosis is that “for every one good founder, you need five or 10 world-class operators.” Exceptional founders can break geographic rules—Melanie Perkins built Canva in Perth—but a repeatable ecosystem needs the executives who recruit, sell, scale operations and later seed the next generation.
Oxford, Cambridge and Imperial collectively graduate only about 500 computer scientists or roboticists per year; Tom wants 5x that output. Stan’s immigration rule is equally direct: relevant graduates should receive a Tier 2 visa “stapled to your graduation certificate,” plus the right to bring family, because many Chinese and Indian graduates currently leave for America.
2. Capital scarcity and founder scarcity form a disputed causal loop
Stan’s benchmark is stark: US technology created roughly $20 trillion through decacorns over 50 years, while two UK outcomes created about $170 billion. US venture firms raised around $76 billion last year; population-adjusted Britain should have raised $15.4 billion, versus $3.7 billion in reality—a gap near $12 billion.
Harry’s pushback—worth keeping: day-to-day investors cannot find enough founders clearing a genuinely global bar, while capital crowds into obviously strong teams and inflates their prices. He sees rapid jumps from “five on 30” to “six on 80,” even the removal of liquidation preferences, suggesting too much money chasing too little exceptional supply.
Tom reverses the usual causality. Rather than waiting for success to attract capital, he argues that abundant capital raises ambition and summons supply, pointing to China’s sustained investment over 20 years; without it, British companies are stunted and founders reasonably ask, “Why don’t I just jump on a plane and form the company in the US?”
Harry adds that US funding is a better product: founders report fast decisions, engaged partners and investors who immediately understand the pitch, while European processes take weeks. Stan expects greater capital and competition to improve that service, because the best founders will select the best VCs and outsized returns will reinforce them.
3. Capital should concentrate behind prospective global champions
Neither guest advocates funding every seed company. The discussion’s condition is sophisticated concentration: identify founders who can absorb more capital without becoming overcapitalized, then let them recruit the best people and pursue a larger ambition. Distributing money evenly would weaken talent density as well as returns.
Wordware carries Stan’s argument. Its Cambridge-trained founders might have raised $5 million at a $20 million pre-money valuation in Britain; in San Francisco they raised $30 million at a $220 million post-money valuation. That financing forces a multibillion-dollar objective—and Stan argues the ownership math may imply investors are really underwriting a $10 billion company.
Tom sees AI producing an even steeper power law. Wiz, acquired for $32 billion after roughly five years, represented about 7% of Israel’s GDP; it began with a world-class team, not a fully specified problem, and was heavily capitalized from day one. The resulting wealth can recycle through employees, founders and local investors.
Harry’s conclusion is categorical: “It’s almost pointless building a sort of number three or number four.” Undercapitalized followers usually sell to US acquirers, preventing Britain from retaining jobs and ownership; capital-intensive, winner-takes-all markets instead require timely large checks behind candidates for global number one or two.
4. Britain can win at the top and bottom of the technology stack
The chip-focused guest divides the stack into semiconductors and hardware, middleware and tools, then applications. Europe’s most credible openings sit at the edges: differentiated AI applications with defensible European advantages, and technologies “attached to the metal” that sell measurable architectural value to business customers.
Semiconductor design is the clearest specimen. The guest says roughly 75% of semiconductor value sits in design—the model used by NVIDIA, Qualcomm and Broadcom—yet Europe holds around 2% of that global market. Fabrication can remain with manufacturers such as TSMC; Britain should build fabless design companies.
Bristol retains rare full-custom microprocessor capability originating with Inmos 40 or 50 years ago, making global winners plausible if financing follows. Tom agrees on specialization: Britain cannot claim expertise everywhere, so it must identify unfair advantages and build the supporting talent, capital and infrastructure around specific locations.
Energy exposes the limits. One data-center operator told Harry that energy represents 4% of US costs but would be 17% in Britain; Harry says energy can be 20% of large-model training cost. The opportunity is therefore chip design, edge hardware and applications—not pretending expensive UK power is competitive for every data center.
5. A scaled British Business Bank could mobilize dormant wealth
Europe is not short of money, Aatish argues; pensions, insurers and roughly 1,100 London family offices simply allocate too little to venture. BBB currently puts about $424 million annually into funds, “a drop in the ocean” against the $15.4 billion population-adjusted target.
Aatish’s proposed reset is approximately $4 billion a year from BBB with 50/50 matching: a manager raising a $1 billion fund could expect $500 million publicly anchored but must secure the other half privately. This could energize principals ready to launch firms and perhaps attract proven US partners to London.
Fund economics should flex around investor constraints. If pensions refuse a 2% fee, the proposal is 0.5% paired with lower participation in carry, while BBB might pay 3% and receive more carry; net economics can still resemble “2 plus 20.” The objective is to make public and private money mesh, not impose identical terms on every LP.
Under the government’s public-sector-net-worth treatment, a guest says these commitments can be recognized as financial assets rather than current spending. Ten years at $4 billion produces a $40 billion portfolio; he claims fund-of-funds returns range from roughly 6% at the low end to the mid-20s at the high end, but Tom insists manager selection must prevent severe adverse selection.
6. Defense combines urgency, spending scale and dual-use spillovers
Harry asks why defense differs from health. Tom explains that Britain still has a dominant state buyer and Europe is fractured among national procurement systems, but multiple services and regiments can become customers; innovation is now being forced, and proximity to Ukraine creates both urgency and a real testing environment.
Three shifts matter to Tom: talented founders now view defense as socially necessary, European governments face an existential threat, and capital appetite has expanded—Anduril’s recent financing at $8 billion was described as oversubscribed. He sees room for next-generation European primes rather than only components for incumbents.
The addressable spending may reach $2–3 trillion across Europe over five to eight years, including multiple layers of the supply chain. One guest cites cyber, UAVs and drones as possible dual-use products; DJI illustrates how commercially scaled hardware can also become strategically important.
7. London’s listing problem begins with too few scaled companies
Only one London-listed technology company exceeds $10 billion, Harry notes: Sage, a roughly 30-year-old ERP vendor. His diagnosis is supply—British companies suffer broken cap tables, wrong hires, weak product focus, limited ambition or inadequate growth capital, then sell before they are ready for an IPO.
Harry also adds a sentiment problem. Founders repeatedly hear that the LSE offers weaker valuation and liquidity, including friction from stamp duty, while US exchanges “roll out the red carpet.” Those negative stories become “SEO for our minds” and reinforce the default choice of America.
Listing location is nevertheless an imperfect metric. Harry would prioritize headquarters, employee concentration and where intellectual property is created over the exchange venue; a US listing can be acceptable if high-value employment and innovation remain in Britain. Stan adds that domestic ownership matters because otherwise the eventual wealth compounds abroad.
Stan proposes a national target: create $500 billion of tech value within 10 years and $4 trillion within 20, compared with roughly $100 billion today. He estimates the first milestone requires about $100 billion of additional capital, or $10 billion annually, and likes the idea of a Norwegian-style public ticker that lets citizens watch national technology wealth accumulate.
8. Tax changes are exporting people with multiplier effects
Non-dom reform presents a conflict between principle and pragmatism. Tom supports equal taxation in principle, yet sees wealthy residents leaving who were also angel investors and employers; he treats those departures as a leading indicator rather than waiting for lagging revenue data to prove the damage.
Aatish argues the combined impulse—ending non-dom treatment, changing inheritance and capital-gains taxes, and adding private-school fees—is simply too large. New residents arrive with pre-existing wealth, so Britain needs a provision that makes staying feasible while preserving enough fairness for ordinary taxpayers to feel part of the same mission.
Harry reports being told that Treasury models are effectively static: raise a tax rate to X and they calculate Y revenue without adequately modeling departures or behavioral responses. The guests do not deny the fairness problem, but reject assuming that taxpayers, capital and employment remain fixed after the rules change.
Harry’s multiplier example is GoCardless. Non-dom Europeans and Americans angel-invested; the company employed hundreds, one founder later built Monzo, another became a London VC, and senior alumni created more businesses. Such recycling once took five to 10 years, he says, but faster company formation might compress it to 18–24 months.
9. Passive subsidies preserve zombies while active capital reallocates talent
Stan calls many EIS and VCT managers ineffective because they optimize for capital preservation rather than exceptional outcomes. If returns cluster between roughly $0.80 and $1.20 per dollar, managers are encouraged to flip companies safely instead of letting founders “swing for the fences”; his blunt prescription is to abolish those funds.
His more controversial target is roughly $7.5 billion of annual R&D credits paid across 55,000 companies. With no quality test beyond qualifying expenditure, the program becomes “classic helicopter money”: it either pays companies that do not need support or sustains businesses that should release their talent and capital.
Harry defends credits for genuinely high-growth young companies but concedes the zombie problem. His compromise is time-based tapering, so a business cannot claim indefinitely; Stan would move much more of the budget into active venture, where investors “double down” on progress and kill failures.
Mentoring is the nonfinancial complement. Tom routinely introduces portfolio founders to experienced operators and tells startups never to impose a minimum check on exceptional angels: a £1,000 or £5,000 investor may contribute disproportionately because the commitment is personally meaningful.
10. China gains as model value migrates toward applications and hardware
Stan has changed his mind on China because foundation models can now be distilled remarkably quickly. He previously called them “the fastest depreciating assets in human history” with useful lives measured in weeks; “it’s almost days now,” which pushes value toward applications and the devices carrying commoditized intelligence.
That favors application companies such as Synthesia and Harvey, but also China’s manufacturing depth. BYD, DJI and Chinese battery expertise show what happens when hardware, supply chains and scale reinforce each other; the semiconductor-focused guest places Britain’s plausible opportunity one layer lower, in the designs powering those products.
One guest’s investment map of AI is essentially large US and Chinese clusters with “tiny little dots of Europe.” China’s 20 years of sustained technology funding built the capacity now visible, while America’s attempt to detach from parts of the world may make Europeans more open to Chinese commercial relationships.
Harry’s pushback is national-security and reciprocity risk: Chinese companies enter Western markets and collect data while Western firms face restrictions, and state subsidies may distort vehicles by 8% to 25%, according to the figures discussed. Stan separates commercial counterparties from the Chinese state—Huawei and ZTE were effective chip customers for him—but concedes that government access and influence require caution.
11. AI brands can compound, but model and business moats remain contested
Harry rejects the “one-person billion-dollar business” thesis. AI makes companies startlingly efficient—he cites Bolt reaching a $40 million revenue run rate in three months—but the same distillation that compresses models will compress business models, allowing successful products to be copied “ridiculously quickly.”
Tom would buy OpenAI at a $300 billion valuation: roughly $12 billion of run-rate revenue, a million ChatGPT sign-ups in an hour, memory-driven switching costs and a brand synonymous with LLMs for both children and older users. At perhaps 20 times forward revenue, he sees extraordinary consumer momentum and partially protected downside.
Harry would invest elsewhere. He expects much demand to come from agents and applications calling APIs, where latency, performance and price dominate and Claude can be “as good if not better”; the consumer chat window may not be the winning interface. Hardware remains valuable too—Nothing’s roughly 7 million devices could become AI conduits.
NVIDIA is not passively awaiting an inference-led decline, Stan says: it is modifying GPU architecture and could pursue in-memory compute organically or through acquisition. From working for Jensen Huang, he remembers “very big ears,” command of minute detail and a culture that was “brutal” but “not malevolent”—effective enough to make adaptation a serious base case.
The closing positioning reflects the thesis: Stan chooses a uranium ETF for nuclear demand, climate constraints and energy productivity; Tom chooses Rolls-Royce after an approximately 3x rise this year, expecting more defense and aircraft-engine demand. Ten years out, Stan forecasts $500 billion of UK tech value; Tom expects unknown AI-native champions, though perhaps no meaningful return to public listings.