Abu Dhabi discovered another kind of power
In 2006 an Emirati company was forced out of American port terminals by a congressional revolt. Twenty years later two Abu Dhabi firms have…
← Back to latest
Essay
The balance sheet is strategic infrastructure
American technology companies will spend around three quarters of a trillion dollars this year on infrastructure they are no longer financing entirely from their own cash flow. What fills the gap is a financial system, and the question of which country has the better one has quietly become a question about which country gets the future it is paying for.
Listen to this briefing
The balance sheet is strategic infrastructure.
American technology companies will spend around three quarters of a trillion dollars this year on infrastructure they are no longer financing entirely from their own cash flow. What fills the gap is a financial system, and the question of which country has the better one has quietly become a question about which country gets the future it is paying for.
In July, Alphabet reported its first quarter of negative free cash flow since it went public as Google in 2004. The company had not stopped making money. It had started spending faster than it earned, which is what happens when capital expenditure across the five largest American technology companies rises from $405 billion in 2025 towards an expected $750 billion this year, and towards something near $1.2 trillion in 2027 on Goldman Sachs's estimates.
The interesting part is where the difference comes from. Investment-grade bond issuance by those companies was $108 billion in 2025, about a quarter of their capital spending. It reached $194 billion in the first half of this year alone and is expected to fund roughly a third of the total, rising towards 35 per cent in 2027. Alongside the borrowing sits a larger commitment that appears nowhere in reported leverage. Microsoft, Meta, Oracle, Amazon and Alphabet have signed about $1.09 trillion of future payments under leases that have not yet begun, against roughly $285 billion of lease liabilities recognised on their balance sheets. The two figures are not directly comparable, since uncommenced commitments are undiscounted payments spread over many years while recognised liabilities are stated at present value. The scale is still the point. Oracle alone has disclosed $260 billion of uncommenced commitments against $37.89 billion recognised, largely for data centres due to begin between 2027 and 2029 on terms of fifteen to nineteen years.
Silicon Valley is normally described as America's technological advantage. The four essays before this one suggest it sits on top of a second advantage that gets discussed far less: a financial system capable of turning a claim about the future into an asset that exists now.
Wall Street is part of the technology stack
The first essay in this series traced how a data centre gets built when nobody wants to own it, and found the answer in credit rather than in engineering. Alphabet's guarantee of another company's lease payments lowered a borrowing cost by a few hundred basis points, and that spread decided which project was viable. The second found that the capacity to absorb losses while waiting sorted the industry into companies that could keep going and companies that had to keep raising.
Neither mechanism works without the machinery around it. Apollo, Blackstone, KKR, Brookfield, the insurance industry, the securitisation market, the investment-grade bond market and the pension funds behind all of it existed before the current build-out and were looking for long-duration assets. American technology companies did not create that capacity. They inherited it, and it is why a lease signed in 2026 can become concrete poured in 2027.
What that machinery is unusually good at is distributing risk widely enough that no individual holder has to be confident. The economics of frontier AI remain unresolved. American capital markets have committed hundreds of billions of dollars anyway, and the apparatus described above is busy pricing those risks every day. Whether the prices are right is a different question, and one nobody can answer yet.
China's machine does something else well
China has venture capital, equity markets, corporate debt and investors willing to fund uncertain technologies. What it lacks is an American-scale network of deep investment-grade debt markets, private credit and institutional capital able to spread this particular kind of risk across enough balance sheets to make it bearable. The comparison is one of degree and architecture.
Where China's system is stronger is in sustaining investment that commercial signals would otherwise stop. Once Beijing decides a sector matters, state banks, provincial governments, guided funds and state-owned enterprises can keep capital flowing through years of poor returns. That machine produced dominant positions in solar manufacturing, batteries, electric vehicles and shipbuilding. Applied to semiconductors, Big Fund III was established with ¥344 billion of registered capital, around $47.5 billion, and has since shifted its emphasis away from fabs and towards the bottlenecks that could be used to choke them: lithography tools, inspection systems, etching, design software.
The results show both what that buys and what it does not. SMIC mass-produces at seven nanometres using deep ultraviolet machines pushed far past their design intent, which requires roughly 34 lithography steps where extreme ultraviolet tools need nine, with a substantial cost and yield penalty. Analyst estimates, rather than company disclosure, put output at seven nanometres and below in the tens of thousands of wafer starts a month and rising steeply. China can now field domestic accelerators approaching the performance class of Nvidia hardware introduced several years earlier, which is enough for large-scale domestic inference, and it remains behind at the frontier on efficiency, software and manufacturing.
The financing story here has an ironic turn. Export controls were meant to constrain Chinese compute, and they have also created a captive domestic market for the firms Beijing was subsidising. SMIC reported second-quarter results this month showing net profit up more than 260 per cent to $479.2 million on revenue of $3 billion, with Hua Hong's profit up by rather more. State patience was supposed to be the thing sustaining these companies through unprofitable years. At the moment they do not need it.
Europe and Abu Dhabi
The third essay argued that Europe's difficulty is mobilising abundant savings as long-duration risk capital and pairing it with credible demand. Europe has enormous private financial markets and finances plenty of risky innovation. What it struggles to do is get enough capital behind the same strategic objective quickly enough for the decision to matter, which is a problem of coordination across twenty-seven budget cycles rather than a shortage of money.
Abu Dhabi is the instructive contrast. The fourth essay described a state of ten million people acting with more financial concentration than economies many times its size, because the number of people who have to agree before $49 billion moves is very small. That is an advantage in speed and a liability in scrutiny, and the two are the same feature.
The counter-reading
Systems that mobilise capital quickly also misallocate it quickly, and the historical case against the American model is strong.
Investment in American communications equipment rose from around $62 billion a year in early 1996 to more than $135 billion by the end of 2000, on Federal Reserve Bank of Richmond figures, much of it long-haul fibre laid on the strength of a claim that internet traffic was doubling every hundred days. Research by Coffman and Odlyzko found that traffic grew at that rate briefly in 1995 and 1996 and roughly doubled annually thereafter. Odlyzko pointed out that the widely repeated figure would have implied internet use growing seventeen million-fold between 1994 and 2000. The market capitalisation of S&P telecommunications firms fell by roughly $700 billion between 2000 and 2002, more than 3.5 per cent of the total value of American corporate equities at the peak.
The most uncomfortable part of that paper is its conclusion. The Richmond Fed authors declined to describe the episode as a bubble, arguing that valuations reflected genuine beliefs about future fundamentals, and that when forecasts vary widely and short selling is constrained, prices come to reflect the optimists. The capital was not obviously irrational. It was wrong, at enormous scale, for reasons that were visible at the time to people nobody was obliged to listen to.
Much of the useful fibre was later bought cheaply by survivors and became part of the infrastructure the internet economy runs on, though not all of the overbuild found a use. A financial system that destroys investor capital on that scale may still end up building the right things, which raises the question of who bears the cost in the meantime. The first essay in this series followed the exposure from technology companies into private credit, insurance books and retirement assets. Those holders bought yield.
Whose future is being financed
Across five essays the same mechanism keeps appearing. Alphabet promises that a lease will be paid and a data centre gets built. Microsoft absorbs losses because Office and Azure already generate cash. Eighteen European governments borrow through a common instrument and a factory in Bavaria starts making interceptors. Abu Dhabi commits decades of capital and receives an export classification. In each case somebody made a credible promise about the future and a financial system converted it into an asset in the present.
Industrial policy tells only part of the story. Export controls, the CHIPS Act, energy permitting, tax credits, defence procurement and Chinese state direction all shape what gets built, and this series has spent most of its length describing them. What increasingly determines the outcome is whether a financial system can turn those policies and those commercial expectations into factories, chips and data centres before anybody knows whether the expectations were sound.
Which leaves the question the series has been circling. All three systems mix state direction with private allocation, and the meaningful differences are in who can coordinate capital, how concentrated that decision is, and what happens when investors and governments disagree. In China the state can override the disagreement. In Europe it usually cannot, and the argument runs through budgets and parliaments at the speed those institutions allow. In America the decision is dispersed across allocators pursuing returns, which makes it fast, adaptable and nobody's in particular. Technological power now depends heavily on institutions that convert expectations about the future into physical infrastructure long before anyone can check the expectations. Each of these systems does that differently, and each has a different answer to the question of who pays when the expectation turns out to be wrong.
Disclosure: McCyber uses AI tools from several of the companies named in this essay, including Anthropic. See About.
Sources: Goldman Sachs via Yahoo Finance; Reuters; Federal Reserve Bank of Richmond; Andrew Odlyzko, University of Minnesota; Financial Times; South China Morning Post; Tom's Hardware; company filings.
Enable JavaScript and use a browser with spoken audio support to listen.
Read aloud by your browser. Voice and availability depend on your device.
In July, Alphabet reported its first quarter of negative free cash flow since it went public as Google in 2004. The company had not stopped making money. It had started spending faster than it earned, which is what happens when capital expenditure across the five largest American technology companies rises from $405 billion in 2025 towards an expected $750 billion this year, and towards something near $1.2 trillion in 2027 on Goldman Sachs’s estimates.
The interesting part is where the difference comes from. Investment-grade bond issuance by those companies was $108 billion in 2025, about a quarter of their capital spending. It reached $194 billion in the first half of this year alone and is expected to fund roughly a third of the total, rising towards 35 per cent in 2027. Alongside the borrowing sits a larger commitment that appears nowhere in reported leverage. Microsoft, Meta, Oracle, Amazon and Alphabet have signed about $1.09 trillion of future payments under leases that have not yet begun, against roughly $285 billion of lease liabilities recognised on their balance sheets. The two figures are not directly comparable, since uncommenced commitments are undiscounted payments spread over many years while recognised liabilities are stated at present value. The scale is still the point. Oracle alone has disclosed $260 billion of uncommenced commitments against $37.89 billion recognised, largely for data centres due to begin between 2027 and 2029 on terms of fifteen to nineteen years.
Silicon Valley is normally described as America’s technological advantage. The four essays before this one suggest it sits on top of a second advantage that gets discussed far less: a financial system capable of turning a claim about the future into an asset that exists now.
Wall Street is part of the technology stack
The first essay in this series traced how a data centre gets built when nobody wants to own it, and found the answer in credit rather than in engineering. Alphabet’s guarantee of another company’s lease payments lowered a borrowing cost by a few hundred basis points, and that spread decided which project was viable. The second found that the capacity to absorb losses while waiting sorted the industry into companies that could keep going and companies that had to keep raising.
Neither mechanism works without the machinery around it. Apollo, Blackstone, KKR, Brookfield, the insurance industry, the securitisation market, the investment-grade bond market and the pension funds behind all of it existed before the current build-out and were looking for long-duration assets. American technology companies did not create that capacity. They inherited it, and it is why a lease signed in 2026 can become concrete poured in 2027.
What that machinery is unusually good at is distributing risk widely enough that no individual holder has to be confident. The economics of frontier AI remain unresolved. American capital markets have committed hundreds of billions of dollars anyway, and the apparatus described above is busy pricing those risks every day. Whether the prices are right is a different question, and one nobody can answer yet.
China’s machine does something else well
China has venture capital, equity markets, corporate debt and investors willing to fund uncertain technologies. What it lacks is an American-scale network of deep investment-grade debt markets, private credit and institutional capital able to spread this particular kind of risk across enough balance sheets to make it bearable. The comparison is one of degree and architecture.
Where China’s system is stronger is in sustaining investment that commercial signals would otherwise stop. Once Beijing decides a sector matters, state banks, provincial governments, guided funds and state-owned enterprises can keep capital flowing through years of poor returns. That machine produced dominant positions in solar manufacturing, batteries, electric vehicles and shipbuilding. Applied to semiconductors, Big Fund III was established with ¥344 billion of registered capital, around $47.5 billion, and has since shifted its emphasis away from fabs and towards the bottlenecks that could be used to choke them: lithography tools, inspection systems, etching, design software.
The results show both what that buys and what it does not. SMIC mass-produces at seven nanometres using deep ultraviolet machines pushed far past their design intent, which requires roughly 34 lithography steps where extreme ultraviolet tools need nine, with a substantial cost and yield penalty. Analyst estimates, rather than company disclosure, put output at seven nanometres and below in the tens of thousands of wafer starts a month and rising steeply. China can now field domestic accelerators approaching the performance class of Nvidia hardware introduced several years earlier, which is enough for large-scale domestic inference, and it remains behind at the frontier on efficiency, software and manufacturing.
The financing story here has an ironic turn. Export controls were meant to constrain Chinese compute, and they have also created a captive domestic market for the firms Beijing was subsidising. SMIC reported second-quarter results this month showing net profit up more than 260 per cent to $479.2 million on revenue of $3 billion, with Hua Hong’s profit up by rather more. State patience was supposed to be the thing sustaining these companies through unprofitable years. At the moment they do not need it.
Subscribe
New essays and snapshots, by email, roughly weekly.
Europe and Abu Dhabi
The third essay argued that Europe’s difficulty is mobilising abundant savings as long-duration risk capital and pairing it with credible demand. Europe has enormous private financial markets and finances plenty of risky innovation. What it struggles to do is get enough capital behind the same strategic objective quickly enough for the decision to matter, which is a problem of coordination across twenty-seven budget cycles rather than a shortage of money.
Abu Dhabi is the instructive contrast. The fourth essay described a state of ten million people acting with more financial concentration than economies many times its size, because the number of people who have to agree before $49 billion moves is very small. That is an advantage in speed and a liability in scrutiny, and the two are the same feature.
The counter-reading
Systems that mobilise capital quickly also misallocate it quickly, and the historical case against the American model is strong.
Investment in American communications equipment rose from around $62 billion a year in early 1996 to more than $135 billion by the end of 2000, on Federal Reserve Bank of Richmond figures, much of it long-haul fibre laid on the strength of a claim that internet traffic was doubling every hundred days. Research by Coffman and Odlyzko found that traffic grew at that rate briefly in 1995 and 1996 and roughly doubled annually thereafter. Odlyzko pointed out that the widely repeated figure would have implied internet use growing seventeen million-fold between 1994 and 2000. The market capitalisation of S&P telecommunications firms fell by roughly $700 billion between 2000 and 2002, more than 3.5 per cent of the total value of American corporate equities at the peak.
The most uncomfortable part of that paper is its conclusion. The Richmond Fed authors declined to describe the episode as a bubble, arguing that valuations reflected genuine beliefs about future fundamentals, and that when forecasts vary widely and short selling is constrained, prices come to reflect the optimists. The capital was not obviously irrational. It was wrong, at enormous scale, for reasons that were visible at the time to people nobody was obliged to listen to.
Much of the useful fibre was later bought cheaply by survivors and became part of the infrastructure the internet economy runs on, though not all of the overbuild found a use. A financial system that destroys investor capital on that scale may still end up building the right things, which raises the question of who bears the cost in the meantime. The first essay in this series followed the exposure from technology companies into private credit, insurance books and retirement assets. Those holders bought yield.
Whose future is being financed
Across five essays the same mechanism keeps appearing. Alphabet promises that a lease will be paid and a data centre gets built. Microsoft absorbs losses because Office and Azure already generate cash. Eighteen European governments borrow through a common instrument and a factory in Bavaria starts making interceptors. Abu Dhabi commits decades of capital and receives an export classification. In each case somebody made a credible promise about the future and a financial system converted it into an asset in the present.
Industrial policy tells only part of the story. Export controls, the CHIPS Act, energy permitting, tax credits, defence procurement and Chinese state direction all shape what gets built, and this series has spent most of its length describing them. What increasingly determines the outcome is whether a financial system can turn those policies and those commercial expectations into factories, chips and data centres before anybody knows whether the expectations were sound.
Which leaves the question the series has been circling. All three systems mix state direction with private allocation, and the meaningful differences are in who can coordinate capital, how concentrated that decision is, and what happens when investors and governments disagree. In China the state can override the disagreement. In Europe it usually cannot, and the argument runs through budgets and parliaments at the speed those institutions allow. In America the decision is dispersed across allocators pursuing returns, which makes it fast, adaptable and nobody’s in particular. Technological power now depends heavily on institutions that convert expectations about the future into physical infrastructure long before anyone can check the expectations. Each of these systems does that differently, and each has a different answer to the question of who pays when the expectation turns out to be wrong.
Disclosure: McCyber uses AI tools from several of the companies named in this essay, including Anthropic. See About.
Sources: Goldman Sachs via Yahoo Finance; Reuters; Federal Reserve Bank of Richmond; Andrew Odlyzko, University of Minnesota; Financial Times; South China Morning Post; Tom’s Hardware; company filings.
Subscribe
McCyber intelligence briefings and structural essays, directly. No institutional spin.