To the historians of the future, the mid-2020s may well be remembered as the era when the world’s most advanced tech companies stopped being asset-light firms and started becoming infrastructure companies instead. For three decades, the Silicon Valley gospel was one of weightlessness. We were told that software was eating the world, and that the most successful businesses were those that could scale to billions of users with little more than a handful of engineers and a clever algorithm. But now the asset-light paradigm has been buried under a mountain of capital expenditures, marking a return to a style of industrial expansion not seen since the heyday of the American steel and railway magnates. The consequences for the US economy will be profound.
The AI Infrastructure Boom: Understanding Tech CapEx
What is CapEx and why does it matter for Tech Giants?
The term “CapEx”, or capital expenditure, has traditionally been the domain of oil companies, airlines, and car manufacturers. It is the money spent to acquire or maintain fixed assets; these can be software or hardware procurement, as long as the asset in question provides long-term value. This is in contrast to operating expenses (OpEx), which cover more day-to-day spending.
For the tech world, CapEx was historically an afterthought compared to R&D (Research and Development). You “researched” an app, then “developed” it, and then sold it a million times. But AI has changed the physics of the industry. Training a model like GPT-5 is not just an intellectual exercise; it is an industrial one. It requires tens of thousands of specialized processors working in unison for months, consuming many megawatts of power. This has forced companies like Microsoft, Google, Amazon, and Meta -collectively termed hyperscalers- to pivot their entire business models toward high-intensity capital formation.
As the chart below shows, hyperscaler investment has ballooned in the last two years, and is set to rise even more this year and next.
From Data Centers to Chips: Where the capital is flowing
The US.hyperscaler CapEx boom is in essence an enormous buildout of the physical plumbing behind AI. The biggest slice is being poured into compute: GPUs, custom AI chips such as Google’s TPUs and Amazon’s Trainium/Inferentia, CPUs, servers, storage and the dense clusters needed to train and run frontier models. Close behind is the data-centre estate itself: Land, buildings, server halls, racks, backup systems, electrical gear, cooling plant and the fit-out required to turn warehouses of concrete and steel into AI factories. A growing share is also going into the less glamorous but increasingly critical bottlenecks: Power, grid connections, substations, transformers, generators, water and advanced cooling. Another chunk is funding networking: Fiber, switches, routers and optical equipment that allow thousands of chips to operate as one system and link data centers together across regions.
Economic Impact: How investment drives US GDP growth
The impact of this investment on the broader U.S. economy has been potent. High Federal Reserve interest rates in recent years were supposed to choke off private investment and slow the economy. However, U.S. GDP has remained remarkably resilient to tight monetary policy. The reason lies in the “I” for investment of the GDP equation: Y = C + I + G + (X – M).
Gross fixed capital formation as a catalyst for expansion
As the chart below shows, tech investment in both hardware and software has played an increasingly import role in boosting US GDP recently. In Q1 2026, the most recent quarter of data, hardware and software investment combined contributed 1.34 percentage points to overall GDP growth of 2.0 percent. It should be noted that tech’s overall contribution to GDP is smaller, as part of the investment is being offset by greater IT imports.
Measuring the contribution of Tech spending to national productivity
The jury is still out on exactly how all this capital spending will influence productivity. While productivity growth has jumped in recent years compared to the prior decade -as the chart below shows- it is likely too early to attribute much of a boost to AI. GPT-4 -the product which gave birth to the current AI boom- was only launched in early 2023. AI agents such as Claude Code, which are currently revolutionising software programming, started to become popular only from late 2025.
AI may have even held back productivity growth slightly in the last couple of years as workers spent time learning the new systems and companies reorganised their workflows the famous “J-curve” productivity trend.
Instead, the jump in U.S. productivity so far in the 2020s compared to the 2010s might be better attributed to the wide adoption of previous technologies, such as videoconferencing, smartphones and cloud computing- the country’s shale boom curbing electricity costs, and a flexible labor market allowing a swift reorientation of workers to the more efficient businesses that survived the pandemic.
However, there are reasons to be optimistic that current tech spending will boost productivity going forward. Unlike the “social media era” of tech, which was largely about capturing attention and selling ads, the AI era is about augmenting human labor. Early evidence suggests that AI “copilots” are already delivering productivity gains in areas like software engineering, legal document review and medical diagnostics.
The impact is only likely to increase as the technology diffuses more widely, improves in quality, and as firms and workers learn how better to use it. Some tech leaders, such as Anthropic’s Dario Amodei have spoken optimistically of annual U.S. GDP growth rates in the double digits as a result of productivity gains; while none of our panellists yet see this scenario coming true, the gains could nonetheless be significant.
Global Comparison: US Investment vs G7 Economies
G7 Economies Ranked: Why the US is leading in business investment
The divide between the United States and most peers in the G7 has been extreme in recent years. While the US has seen a significant rise in capital formation driven by the tech sector, key euro area economies and Japan have recorded stagnating or even declining investment levels, as the chart below shows.
Despite the political and tariff uncertainty sown by president Trump, the US remains an attractive destination for business investment even setting aside the boom in the tech sector.
The US advantage is many-fold. First is capital market depth. A US startup can raise USD 100 million in a week; a European startup might take a year to raise the same amount. Second is the energy edge. The US is the world’s largest producer of oil and gas, leading to energy prices roughly half those in Europe. Then there’s the regulatory environment, which is generally more permissive than in other Western economies. A flexible labor market and world-class universities also help shift the needle in America’s favor.
The sustainability of the American Investment Model
Despite current momentum, the American model faces significant risks. The first relates to the build-out of energy infrastructure: If the US cannot bring new transmission lines and power plants onstream fast enough, the data center boom will run out of juice. Then there’s productivity: If the returns on investment from AI aren’t as high as hoped, the current U.S. investment boom could turn into a winter. Political turmoil a loss of Federal Reserve independence, interruption of democratic institutions, etc – is a further cause for concern.
A stock-market bust is another risk. Leading US. startups powering the AI boom, such as OpenAI and Anthropic, have large negative free cash flow, are borrowing large sums of money and engaged in increasingly circular investment patterns with other tech firms. The risk is that one of the major firms involved overreaches and goes bust, causing a collapse of investor confidence in the sector as a whole.
Finally, there is the geopolitical bottleneck. The entire US tech investment model is predicated on access to advanced semiconductors designed in California but manufactured in Taiwan. A conflict in the Taiwan Strait could sever this semiconductor supply chain and off access to the best chips which is why the Trump administration is so keen for TSMC, Taiwan’s main semiconductor manufacturer, to build more fabs in the US itself.
For now, our panelists don’t see any of the above risks seriously getting in the way. Our Consensus is for US outperformance to continue, with investment growth to easily outpace that seen in the rest of the G7 in the coming quarters as the above chart demonstrates. The surge in capital formation may be frothy, and it may be risky, but bar a major surprise it will remain one of the American economy’s key engines for the foreseeable future.
