Nearly Right

Britain aims for 5% of AI chip market with £1 billion whilst rivals invest hundreds of billions

Lord Browne's semiconductor proposal exemplifies a recurring pattern: technical brilliance paired with commercially comfortable ambition

Lord Browne wants Britain to capture 5% of the global AI chip market—$35 billion in annual revenue by 2033. His case rests on real strengths. Cambridge-based Arm designs over 90% of smartphone chips worldwide. British universities remain research powerhouses. The country invented the computer itself.

Then come the numbers that matter. Britain will double its chip design workforce from 12,000 to 24,000 over a decade, adding 1,500 students annually. Taiwan employs 150,000 technical workers in semiconductors alone. South Korea is training 36,000 new experts whilst Samsung hires 80,000 more. Britain proposes £1 billion over ten years. South Korea committed $470 billion. The United States passed a $52 billion CHIPS Act.

This is not a competition. This is a ritual—the familiar British performance where world-class technical capability meets commercially comfortable ambition, where bold targets arrive with resources that couldn't possibly achieve them.

Britain's semiconductor history: from pioneering to peripheral

The technical record dazzles. Britain built the first general-purpose electronic computer, the first electronic memory, early parallel processing systems. Then came the commercial reckoning.

Inmos received $176 million in government backing in 1978 and developed the transputer—ahead of its time, technically sophisticated. It passed through Thorn to STMicroelectronics and disappeared. Ferranti invented the gate array and dominated the world semi-custom market, then failed to invest in CMOS during the late 1970s. GEC closed profitable chip factories during the 1971 downturn. The government's Alvey Programme in the 1980s met its technological objectives whilst GEC and Plessey continued losing ground internationally.

Newport Wafer Fab, one of Britain's last fabrication plants, was sold to Chinese firm Nexperia, forced into resale to an American company after belated government intervention. No British manufacturer stepped forward to buy it. Not one.

The pattern transcends semiconductors. Britain invented key internet protocols but doesn't host major internet companies. British scientists pioneered artificial intelligence; American firms dominate the industry. Britain excels at the breakthrough, the elegant solution, the technical triumph. What it cannot sustain is the brutal 20-year slog from invention to market dominance—the patient capital deployment, the tolerance for expensive scaling, the institutional commitment to manufacturing.

What building semiconductor dominance actually costs

Taiwan offers the clearest case study. In 1974, the government appointed the Industrial Technology Research Institute to build a semiconductor industry. In 1976, it convinced RCA to transfer technology. When TSMC launched in 1987, the government provided 48% of startup capital—$100 million in seed money.

But money was only the beginning. The government built Hsinchu Science Park from scratch: roads, power, water, telecommunications, all subsidised. Semiconductor firms received tax breaks and reduced-cost utilities. Taiwan deferred consumption for decades, allowing trade dependence to surge from 27% of GDP in 1960 to over 103% by 1983. The demonstration factory ITRI established achieved higher yields than RCA's American facilities—immediate proof that Taiwanese capabilities matched US standards.

The results: TSMC now carries a market capitalisation exceeding $550 billion and employs over 150,000 people. Its fabs achieve 95% yields—19 perfect chips from every 20 produced. Taiwan's semiconductor sector generates $115 billion annually, representing 20% of the country's GDP. TSMC alone constitutes 30% of the Taiwan Stock Exchange.

South Korea followed the same script. Samsung's founder established Samsung Semiconductor in 1978 with explicit government encouragement. By 1983, South Korea became the third country to develop 64K DRAM chips. By 1992, it topped the DRAM market. Current commitments dwarf that history: $470 billion in private investment over 23 years, supplemented by $19 billion in government support. The K-CHIPS Act provides 25% tax credits for facility investments, 50% for research.

The arithmetic is brutal. A single cutting-edge fab costs $20-24 billion to build, $1 billion annually to operate. It consumes power equivalent to 50,000 homes yearly and vast water supplies. TSMC uses 5% of Taiwan's entire energy—more than Taipei. This isn't an industry where modest bets yield proportional returns.

The arithmetic of ambition

Britain's National Semiconductor Strategy arrived in May 2023 after repeated delays. It commits £1 billion over ten years. The first three years receive £200 million.

Industry reaction was withering. Simon Thomas, chief executive of Paragraf, called it "frankly flaccid." Scott White of Pragmatic Semiconductor questioned whether the sum was "enough money to make a difference." Andrew Buss at IDC Europe said the funding is "not nearly enough to be viable if the UK wants to be one of the main global players."

The comparison table writes itself:

Britain accounts for 0.5% of global semiconductor sales. It lacks end-to-end manufacturing capability. The compound semiconductor cluster in South Wales competes against Dresden, Grenoble, Leuven, and Eindhoven—established centres with vastly greater resources.

Now examine the target. The AI chip market reached $123 billion in 2024 and projects to $295-700 billion by 2030. Browne proposes capturing 5%—$35 billion annually. But this isn't a market where polite intentions matter.

Nvidia commands 65-98% market share across different AI chip segments. Its data centre revenue hit $18.4 billion in a single quarter, up 409% year-over-year. AMD projects data centre GPU revenue exceeding $4.5 billion. Intel, despite a $20 billion Ohio facility and decades of manufacturing expertise, struggles to compete. Even established giants with multi-billion-dollar R&D budgets find the market brutally competitive.

Britain's entire ten-year strategy couldn't build one competitive fab. That single facility would cost $20 billion—20 times the UK's total commitment.

Why modest targets guarantee modest outcomes

Browne's document identifies the right insight: "In today's gold rush, the real fortunes belong to those who build the shovel, not just those digging for digital gold." True. Nvidia, supplying chips, captures far more value than OpenAI, producing generative marvels. The diagnosis is perfect.

Then comes the treatment. Britain will avoid competing with Nvidia in data centres and instead focus on "specialised applications"—robotics, factory automation, medical devices, autonomous vehicles. This sounds strategically prudent until one asks: who else is targeting these sectors?

Apple designs custom AI chips for mobile. Google has invested billions in tensor processing units. Amazon deploys its own Trainium chips. Tesla develops semiconductors for autonomous vehicles. Every "specialised application" attracts massive investment from companies with vastly greater resources.

Specialisation without the capital to dominate isn't strategy. It's a rationale for accepting permanent minor status whilst calling it focus.

The workforce target reveals the same psychology. Adding 1,500 engineering students annually to reach 24,000 total workers over a decade represents careful expansion that won't disrupt existing priorities. South Korea trains tens of thousands annually. Taiwan employs 150,000. These aren't different strategies—they're different ambitions entirely.

Britain proposes comfortable targets that feel achievable because they demand little change. That comfort is precisely what guarantees they'll achieve little.

Can industrial strategy overcome structural timidity?

Britain's semiconductor strategy raises the larger question: can government direction overcome four decades of institutional choices that systematically favoured finance over manufacturing?

The country opted out of the semiconductor race in the 1980s, embracing neoliberal policies that prioritised market mechanisms over strategic direction. Taiwan, South Korea, and Japan chose differently—developmental state models that subordinated short-term market signals to long-term industrial positioning. They accepted costs Britain refused: deferred consumption, directed investment, patient capital that expected returns measured in decades.

Recent analysis of Britain's semiconductor landscape reveals additional complications. Chinese investors hold equity positions in 36 of the country's 61 major semiconductor firms, typically with stakes below 1%—small enough to avoid regulatory scrutiny yet representing pervasive influence. Between 2014 and 2024, Chinese firms accounted for 7% of greenfield investments. Huawei maintains stakes in companies like XMos and operates significant R&D in Cambridge.

Meanwhile, 15% of supplier facilities sit in the European Union, 19% of customers are EU-domiciled. Brexit introduced supply chain vulnerabilities whilst excluding British firms from most European Chips Act subsidies, creating competitive disadvantage with no offsetting benefit.

The strategy document acknowledges that "the UK's semiconductor industry will never be wholly sovereign" and "completely onshoring chip production is unlikely to be economically viable." Admirable realism. But it raises uncomfortable questions about what "capturing 5% of the global market" actually means if sovereignty is unattainable and competitive manufacturing is economically unviable.

The proposed solution—deep integration with US chip manufacturers and access to American sub-3-nanometre fabrication technology—reveals Britain's emerging position. TSMC is building fabs in Arizona, bringing Taiwanese expertise to American soil through CHIPS Act incentives. That's technology transfer between equals. Britain's relationship increasingly resembles customer rather than partner.

The pattern continues

Britain's semiconductor strategy is remarkable for how perfectly it exemplifies the country's post-industrial pattern. Identify genuine technical strength. Acknowledge strategic importance. Propose government coordination. Set targets that sound impressive without requiring uncomfortable resource commitments. Allow politicians to claim action whilst avoiding the difficult choices that building genuine capabilities would demand.

Taiwan deferred consumption for decades to build semiconductor dominance. South Korea directed chaebols into electronics with massive ongoing subsidies. The United States, after watching its manufacturing base erode, committed $52 billion specifically to reverse decades of offshore outsourcing. Each approach demanded accepting significant costs and risks.

Britain proposes £200 million over three years and strategic partnership with nations that built capabilities through sustained commitment.

Browne concludes by suggesting that "if the UK leans in fully, the new age of AI may be written not just in code, but in silicon, bearing a very British imprint." The conditional clause does considerable work in that sentence. If.

The evidence suggests Britain isn't leaning in fully. It's leaning in comfortably, which across 50 years of semiconductor history has proven insufficient. Britain possesses technical capability—demonstrably, world-class capability. What it has never demonstrated is institutional willingness to commit resources at the scale manufacturing dominance demands.

The AI chip market will be written in silicon. The question isn't whether Britain has the brains to contribute. The question is whether it has the stomach for what contribution requires: sustained capital deployment, patient institution-building, acceptance that technical elegance without manufacturing scale buys you a footnote in others' success stories.

The current strategy provides an answer. Just not the one its authors hoped to give.

#artificial intelligence