Preamble

Few recent transformations simultaneously affect technology, the economy, finance, currency, and geopolitics. Artificial intelligence touches upon them all. That is why it cannot be treated merely as a matter of innovation or markets; it has become a governance issue that will impact—at least for the next decade—economic, social, and financial flows and organization across different regions of the globe.

The promise. AI could be the first general-purpose technology since electricity and computing with the potential to boost productivity across almost every sector. It can accelerate scientific research, lower the cost of accessing knowledge, and automate cognitive tasks in ways no previous wave of technology has achieved. For aging, slow-growth economies—such as those in Europe—it could be one of the few paths to sustaining the welfare state without increasing the tax burden; yet, it also carries the risk of excluding these demographics from the global economy's most productive sectors.

However, the risk is real, and it does not lie solely where people usually look for it. Public debate focuses on the risks inherent to the technology itself, and one of these deserves emphasis: the weakening or flawed design of human oversight regarding systems that begin to act proactively rather than merely responding. This risk is inextricably linked to financial factors. Multiple compression pressures hyperscalers to quickly monetize infrastructure CAPEX, and the fastest route is to commercialize more autonomous and "aggressive" AI agents with fewer human control points. Financial pressure can thus act as an accelerator for technological risk. Furthermore, in the absence of regulation or a clear liability framework for damages caused by systems that slip out of control, this risk remains unpriced: companies do not bear the cost, while markets ignore or undervalue it. It is a classic externality, the cost of which will ultimately fall upon third parties. Therefore, the most immediate risk lies in the way the promise of AI is being funded, priced, and governed.

• In financial markets, a handful of companies command an unprecedented share of global savings, and the market capitalization of a single firm exceeds nearly half of the entire listed energy sector.
• In the monetary system, debt-funded investment in AI compounds the challenges posed by high public deficits, the gradual erosion of the dollar’s ​​status as a reserve currency, and currency interventions that no longer involve consultation among allies.
• In geopolitics, computing power has come to be viewed as an instrument of state power. The rivalry between Washington and Beijing, reliance on Taiwan, control over critical minerals, and the role of Gulf sovereign wealth funds have turned AI into a theater of strategic competition.
• In the real economy, demand driven by stock market expectations is already straining power grids, water supplies, and household utility rates—even before any productivity gains have materialized.

The point of convergence is sovereignty. Governments worldwide are planning networks, allocating land and water, granting incentives, and steering industrial policy based on a single signal: the market valuation of a select group of companies. At the same time, citizens' pensions, insurance, and savings are more or less indexed to that very signal. Without any democratic decision mandating it, the State has become the implicit guarantor of a private wager.

This is what this text terms the mortgaging of sovereignty. It is not a matter of transferring power to technology companies, but something more subtle: the subordination of public decision-making to a financial projection that no government controls—and the consequences of which, should expectations fail to materialize, will fall upon the citizens.

The central question, therefore, is not whether AI will fulfill its promise. It is whether states can reap its benefits without, in the process, surrendering their capacity to protect those they govern. It is within this tension that the sovereignty contract plays out.

To the Thesis

Artificial intelligence is not, in itself, the most immediate risk. The risk lies in how governments regulate it and markets price it. A valuation that capitalizes decades of future profits tends to shape monetary and economic policy—and, consequently, guide investment and public decisions regarding energy, water, infrastructure networks, and debt. Where regulation fails, the market does not price the risk; where the market overvalues, regulation hesitates to intervene.

When market valuation turns into a political signal, sovereignty is not transferred to corporations; rather, it is mortgaged to a projection that no one controls. The State then appears as both the guarantor and the debtor of that mortgage.

This text argues that the social contract of sovereignty does not break at the moment of collapse. It breaks earlier—when the State ceases to arbitrate between the citizen and capital and begins to guarantee the latter at the expense of the former.

The arithmetic of claim

By 2030, the five largest companies in the semiconductor-AI ecosystem could be worth the equivalent of 20% of global GDP, yet the profits underpinning that value amount to less than 1% of GDP annually. The difference lies in the longevity the market attributes to them.

Order-of-magnitude figures, based on a global GDP of $115–140 trillion between 2026 and 2030. Companies considered: Nvidia, Microsoft, Alphabet, TSMC, and Broadcom.
Note: Market capitalization is a stock; GDP is a flow. Market value is the present price of a stream of profits expected over decades, not an annual claim on production.

The anomaly is not that these companies account for one-fifth of the economy. It is the market assigning them a claim on future income equivalent to one-fifth of the planet's annual output, based on capturing about 2% of it. What is becoming concentrated are claims, not production.

This does not make the imbalance harmless; quite the opposite. Market capitalization directs savings via indices, serves as collateral, lowers these companies' cost of capital, and signals to governments and utilities where to invest. The consequences stem from the resource allocation driven by these prices.

Where Duration Fails

A 25x multiple represents a bet on the cycle's duration, and four forces are challenging this very suspension of economic time. None of them are ideological.

1. Physical limits. Energy, water, networks, cooling, and manufacturing capacity do not scale at the same pace as prices. For a long-duration asset, a delay equates to value destruction. Scarcity exerts pressure on the multiple through two additional channels:
   1. Resource inflation. The same CAPEX buys less capacity: Microsoft attributed roughly $25 billion of its 2026 CAPEX increase to memory chip inflation rather than new capacity. This inflation fuels general inflation, keeps interest rates high, and disproportionately penalizes long-duration assets. It also shifts economic rent toward new bottlenecks: Micron is valued at $1.2 trillion and SK Hynix at nearly $1 trillion.
   2. Model shift. Hyperscalers are transitioning from a software model—capital-light with high returns—to an infrastructure model akin to telecommunications or utilities. Amazon’s negative free cash flow is the first sign of this. Capital-intensive businesses typically trade at 15–20 times earnings, not 25–35. The market continues to price in software margins for a business that has become infrastructure-based; this mirrors the valuation error made with telecommunications in 2000.
2. Competition. AMD, hyperscalers' proprietary ASICs, and inference chips are squeezing market share and margins. If margins drop from 40% to 25%, maintaining the same market capitalization requires significantly higher revenue.
3. Efficiency-driven disruption. In January 2025, DeepSeek demonstrated competitive models trained using less powerful chips. Nvidia lost approximately $593 billion in a single day—a Wall Street record.
4. Nvidia’s revenue represents CAPEX for others. That 2% of GDP is not a final product; it is an intermediate cost for the rest of the economy. It is only sustainable if end customers derive more value from AI than they pay for it, and realizing that gain requires a separate investment that is currently underfunded (see below).

The fourth point is an internal contradiction, not a peripheral objection. Hyperscalers spent around $400 billion in CAPEX in 2025 and are projecting about $800 billion for 2026. Amazon, despite 20% revenue growth, now has negative free cash flow: minus $7.6 billion for the twelve months ending in June 2026.

If the concentration of capital at the top diverts investment away from the industries expected to monetize AI, the circuit strangles its own end-market demand.

The funding paradox. AI generates productivity gains only if user companies invest in complements: reorganizing processes, training staff, preparing data, and adapting systems. Literature on the "productivity J-curve" (Brynjolfsson, Rock, and Syverson) shows that these intangible investments are worth several times the cost of the technology itself and that the gains materialize only after a lag of years. The sources that should be funding them are all under pressure:

• retained earnings, squeezed by energy costs and AI subscriptions themselves;
• bank credit, which is more expensive due to high interest rates—especially for European SMEs, which rely on banks;
• public investment, constrained by deficits;
• capital markets, where indices channel savings toward the top and hyperscalers already account for around one-tenth of new issuances by non-financial companies in the euro area.

Infrastructure is overfunded, whereas adoption—which would validate the valuation multiples—is underfunded. The negative phase of the J-curve drags on, and the returns that would justify the CAPEX remain elusive.

In Europe, the paradox is even more acute. European savings fund American infrastructure—with $872.8 billion in US securities held in 2025—while European companies lack the funding to adopt the technology. This is precisely where the risk of being shut out of the most productive sectors materializes.

Social metabolism: competing for basic resources

The competition between citizens and data centers for energy and water is no longer a projection; it comes with a price tag and a bill.

PJM, United States. In the largest U.S. electricity market, the capacity price surged from $28.92 to $329.17 per MW-day over two years. The independent market monitor attributed 63% of this increase to demand from data centers—amounting to roughly $9.3 billion paid by all consumers. Consulting firm ICF estimates that residential rates could rise by 30% to 60% by 2030.

Ireland. Data centers consumed 23% of the country's electricity in 2025, up from 5% in 2015. Total residential consumption stood at 28%. The regulator had already effectively suspended new grid connections in the Dublin area back in 2021.

Water. Competition is local and becomes visible during periods of drought. In Georgia, a QTS data center (owned by the Blackstone Group) consumed approximately 110 million liters without being billed, due to misconfigured meters. This occurred during a severe drought, while county residents were banned from watering their lawns and were complaining of low water pressure. The company later paid $147,000, with no fine imposed. In Arizona—which is required to cut its Colorado River allocation by 18% by 2026—several municipalities are already imposing limits on data center consumption. In terms of aggregate volume, U.S. data centers consumed approximately 66 billion liters in 2023, with projections pointing to 144–276 billion liters by 2028. Between 70% and 80% of the cooling water evaporates and does not return to the local system.

Globally, consumption remains far below that of agriculture. The problem is not the total volume, but the concentration in basins already under water stress, where there is direct competition with households.

Fossil fuel regression. This is the most compelling evidence. In the United States, 34 units across 15 coal and fuel-oil power plants—totaling 16.5 GW—have postponed their closures, citing rising demand, largely driven by data centers. They emitted approximately 65 million tonnes of CO₂ equivalent in 2023—more than the entire state of Massachusetts. Georgia Power delayed the closure of the Bowen plant until 2038 after securing commitments from 70 data centers. Dominion no longer plans to close the Clover plant before 2045. In some instances, the Department of Energy itself mandated continued operation via emergency orders; keeping Michigan’s Campbell plant running cost around $615,000 per day, a cost shared by regional consumers.

There is a caveat here that reinforces our argument rather than weakening it. Grid Strategies estimates that utility companies may be overestimating data center demand by as much as 40% because projects are often proposed simultaneously across multiple regions. Consequently, demand that is partly speculative is already extending the lifespan of aging fossil fuel infrastructure. This is the valuation of assets as a political signal in its most concrete form.

Corporate power purchase agreements (CPPAs) exacerbate this imbalance. Even when they fund new capacity, they secure the best projects and priority grid connections, while grid costs are shared by everyone. Thus, the pursuit of stock market expectations finds its way into household utility bills well before any productivity gains materialize.

Consumer electronics. The pressure extends to mainstream digital goods. In the second quarter of 2026, contract prices for memory rose between 49% and 89% in a single quarter; storage used in smartphones rose by more than 100%. IDC forecasts price increases of up to 8% for computers and 3% to 8% for smartphones in 2026, alongside a decline in PC sales of between 4.9% and 8.9%. Low-cost brands with thinner margins are the hardest hit, whereas Samsung and Apple have secured memory supplies for the next 18 to 24 months. Consequently, the impact falls primarily on lower-income households and emerging markets.

While the impact is quantified in terms of components and analyst forecasts, it has not yet been isolated in official consumer price indices.

Cost allocation is a policy choice. In Virginia, a legislative proposal shifts distribution and capacity costs from households to data centers exceeding 25 MW. This demonstrates that the competitive landscape is not a technical inevitability. With every rate-setting decision, the state reveals whether it is protecting the citizen or stock market expectations.

Debt, interest rates, and global transmission

AI investment has shifted from being funded by profits to being funded by debt, precisely when several major economies are facing fiscal crises. Both are competing for the same pool of savings, and the risk associated with one is beginning to spill over into the other.
Scale.

• Scale. Hyperscalers issued approximately $170 billion in bonds through the end of June 2026. Morgan Stanley projects nearly $570 billion in AI-linked debt for the year.
• European savings. More than a third of these issuances were denominated in currencies other than the dollar. Amazon made its debut in sterling with a £4.25 billion issuance in September. The ECB tallied around €40 billion in hyperscaler bonds—nearly one-tenth of all new issuances by non-financial corporations.
• Risk pricing. The cost of insuring Oracle’s debt against default has more than tripled since its September 2025 issuance.

The ECB has found no evidence of other issuers being widely crowded out—for now. The risk lies not in the present, but in the overlap.
Deficits hovering around 6% of GDP in the US, alongside fiscal strains in France, the UK, and Japan, are competing for the same global pool of savings as AI capital expenditure. Strictly speaking, these are not two overlapping bubbles; they are two competing claims on the same future earnings, and neither is guaranteed to be covered.

If the cycle breaks, the state—which ought to cushion the shock—will lack the necessary fiscal space. An implicit sovereign "put" exists, but it relies on available funding.

The reversal of risk perception. By November 2025, parts of the market were already treating certain companies as safer borrowers than sovereign states. Microsoft holds a AAA rating—higher than that of the United States—and about 5% of investment-grade French corporate bonds traded at yields lower than those of French government debt. Bond issuances by Meta ($30 billion, with $125 billion in orders) and Alphabet ($25 billion, with around $90 billion) priced with smaller spreads than those of similarly rated companies.

Em 2026, os juros soberanos atingiram máximos de várias décadas e os prémios das hyperscalers começaram a alargar.

The market is beginning to price in the risk associated with hyperscalers, albeit gradually. Demand for bond issuances dropped from nearly five times the supply in February to less than two times in July. The discount required on new issues rose from around 2 basis points for 2025 maturities to 12 basis points for 2026, and 78 of the 91 bonds issued during the year were trading at yields above their issuance levels by the end of July.

Sovereign risk contagion. AI risk spills over into public debt through two channels.

1. A mechanical vector. When Amazon or Alphabet issue 30-year debt, they compete directly with the Treasury for the same long-term capital from insurance companies and pension funds. Buyers often hedge interest rate risk by selling government debt, which adds further pressure. This contributes to the rise in long-term interest rates—alongside other factors such as confidence in the Federal Reserve—though its exact weight has not yet been isolated.
2. A contingent vector. If the cycle breaks, the State will absorb the losses through bailouts, automatic stabilizers, and lost tax revenue. The risk that companies do not bear is, therefore, implicitly embedded in sovereign debt.

Hence the paradox: the market assigns lower risk to the guaranteed entity than to the guarantor. Companies whose collapse would force states to intervene finance themselves at lower spreads, while the states that would have to intervene pay interest rates at multi-decade highs.

Global transmission. The shock would not be confined to the US. Every national defense—whether of the yen, the euro, or public finances—transmits pressure to the others, as the events of August 2026 demonstrated.

Defending the yen with euros. On August 1, 2026, the US Treasury intervened alongside Tokyo to prop up the yen. Instead of selling dollars, it sold euros from its exchange stabilization fund, notifying the ECB only after the operation was complete. At Jackson Hole, European central bankers expressed concern regarding Washington’s predictability, though there were no signs of risk to dollar swap lines.
The episode involved small amounts but held great significance. The issuer of the reserve currency protected its own currency using a third party's currency, without prior consultation. European monetary sovereignty became dependent, in part, on the convenience of those holding euros.

Japan as a transmission vector. For years, investors borrowed in cheap yen to buy US assets, including tech stocks. In August 2024, the unwinding of these positions triggered a global market drop within days. And if Tokyo sells US bonds to defend the yen, long-term interest rates—which underpin AI valuation multiples—will rise.

The dollar: real erosion, slow pace. The dollar accounted for 57.13% of global foreign exchange reserves in the first quarter of 2026, down from around 71% in 2000. The immediate threat is not the loss of reserve currency status, but the impact on the term premium: a less reliable dollar raises the cost of money, and long-duration assets are the first to suffer.

China as the catalyst. Beijing could trigger the chain reaction in four ways:

• export controls on rare earths, gallium, and germanium;
• an escalation regarding Taiwan, on which TSMC depends;
• open and low-cost models that erode pricing power, such as DeepSeek in 2025;
• a gradual reduction of its US debt holdings.

Where capital comes from and where it goes

Where it comes from

The capital sustaining the loop was not drawn from a competing sector. It was drawn, above all, from the savings of third parties: pensioners, policyholders, European savers, and Gulf oil revenues.

A preliminary clarification: capital does not shift between sectors like water between vessels. When Nvidia rises, no sector pays for the increase; the buyer pays the seller. What can be tracked is the shift in relative weight, net investment flows, and the actual money funding CAPEX.

Relative weight. The technology sector reached 39.4% of the S&P 500 in June 2026, surpassing the record set during the dot-com bubble. The ten largest companies accounted for 40.8% of the index in May, compared to 26.6% at the peak of the 1990s and around 19% a decade ago. Financials, energy, industrials, and small-cap companies lost weight. They did not lose money; they lost the share of savings automatically allocated to them.

The rotation is not a one-way street. In the first quarter of 2026, amid AI fatigue, consumer staples funds rose more than 13% before technology regained the lead.

The irony of oil. Part of the asset the market values ​​as the "new oil" is financed by the old. Abu Dhabi’s MGX, alongside BlackRock, acquired Aligned Data Centers for $40 billion, effectively converting oil revenues into computing power.

Europe provides the funding without calling the shots. Europeans invested $872.8 billion in long-term U.S. bonds in 2025. European savings thus finance a strategic U.S. program, even as Europe debates its own lag in AI.

The chain ends with households. In May 2026, the Financial Stability Board warned of the migration of private credit risk toward insurance companies and pension funds. In July, U.S. insurance regulators launched inquiries into insurers' exposure to data center financing.

The same citizen pays three times over: as a saver whose retirement is tied to this cycle; as an energy consumer bearing the cost of grid infrastructure; and as a taxpayer acting as the backstop for an implicit bailout. The capital fueling this loop was not diverted from another sector; it was drawn from the savers' own future.

Where it goes: the consumer is worth more than the producer.

Since the launch of ChatGPT in late 2022, Nvidia and the hyperscalers have added approximately $14 trillion in market capitalization. Major energy groups, combined, added about $0.6 trillion—a ratio of more than 20 to 1.

2026 figures from CompaniesMarketCap; 2022 baselines are approximate and subject to confirmation prior to publication. GE Vernova—listed since 2024 and currently valued at $264 billion—is excluded from the comparison.

One company versus an entire sector. Nvidia alone is worth approximately 45% of the entire global listed energy sector, which comprises 802 companies with a combined value of $12.6 trillion.

The physical core has stagnated. Oil and gas—still the foundation of the global energy system—have seen virtually no appreciation over the past four years, despite the massive energy demands of AI.

Energy values ​​rise only when they serve AI. The segments that have seen exponential growth—nuclear power and electrification—are those the market views as suppliers to data centers. They are growing rapidly, albeit from small starting points.

The market has shifted from valuing energy as the physical foundation of the economy to valuing it as a derivative of the AI ​​thesis. The energy consumer has generated twenty times more value than the producers, even though the consumer cannot exist without them. This represents a quantified reversal of the comparison with oil.

The rationality the market does not price in

Outside the financial circuit, the expansion of AI has a rationale, but none of its rationales is the one the market is pricing in.

Strategic. For nation-states, AI represents power, and cost is a secondary concern. During the Cold War, such a program would have been funded by public money. Today, it is funded by private savings via the stock market: a privatized form of strategic Keynesianism, where the objective is the state's, but the risk is borne by savers.

Corporate. For every hyperscaler, under-investing is riskier than over-investing. It is a prisoner's dilemma: rational for each individual player, yet resulting in over-investment for the group as a whole.

Historical. The British railway boom of the 1840s ruined investors but left behind a network that served the country for a century. What is rational for society can be irrational for those buying in at current prices.

Redistributive. The return that would justify these valuations stems primarily from the substitution of capital for labor.

The disconnect is this: investors are paying for financial returns on a project whose underlying logic is state power and labor substitution—neither of which guarantees those returns.

The narrow window

The redistributive logic holds a decisive paradox: both possible outcomes undermine current valuations.

• If AI fails to replace labor at scale, the capital expenditure (CAPEX) will not be recouped, leaving the debt used to finance it uncovered.
• If AI succeeds, it erodes the wage income used to purchase goods and services from client companies—thereby undermining the very final demand meant to pay for AI.

The market is pricing in a narrow window: AI replaces enough labor to generate massive profits, yet not so much that it destroys the demand that pays for them. It is upon this narrow window that a market capitalization equivalent to one-fifth of global GDP rests.

The sovereignty contract

In the contractarian tradition, individuals leave the state of nature and surrender their power to a sovereign in exchange for order, justice, and protection. The minimum obligation of such a sovereign is the protection of life, which includes access to the means of subsistence.

A return to a natural state driven by price. When citizens compete for energy and water against entities with vastly superior bargaining power, and the State arbitrates based on market signals, there is a partial regression to the state of nature. It is not Hobbes’s "war of all against all," but an unequal competition for basic goods, mediated by price rather than violence.

The "will of all" replacing the "general will." Rousseau distinguished the volonté générale—oriented toward the common good—from the volonté de tous, a mere sum of private interests. Market valuation represents the "will of all" regarding capital, weighted by wealth. When public decision-making becomes indexed to this figure, that sum usurps the place of the general will.

A scale imbalance. The sovereign retains the obligation to defend the res publica, yet the financial power to which it submits has grown far beyond its capacity to arbitrate or guarantee it. The US stock market is currently valued at approximately 2.2 times the nation's GDP—surpassing the peak of the dot-com bubble and the levels seen in Japan in 1989.

US Federal Reserve series (Z.1); other measures show a figure of around 200% for the year 2000. Part of this rise is structural: multinationals book profits generated worldwide on the stock market, whereas GDP measures only domestic production

The politically decisive figure is the comparison with the State's fiscal capacity. With federal revenues hovering around 17% of GDP, the US stock market value is equivalent to approximately 13 years of federal tax revenue. The State—which would be called upon to stabilize the system—possesses, in any given year, a tiny fraction of the financial value it would need to protect. It is no longer merely a matter of influence; it is a matter of scale.

The State as guarantor. Networks are scaled up, land is ceded, and incentives are granted because the market has validated demand. Citizens' pensions are indexed to that same price. If the cycle breaks, the State will be forced to intervene—not by choice, but because networks, pensions, and credit are already tied to the outcome. Costs are socialized, while the capital gains accumulated beforehand remain concentrated.

When a recession becomes the greatest in history. A collapse of the AI ​​complex, in isolation, would produce a severe recession—like the telecom crash of 2001—but one cushioned by monetary and fiscal policy. The Depression of 1929 required a paralysis of rescue mechanisms, and it was the gold standard that caused that paralysis. Today, an extreme scenario would require an equivalent convergence of factors:

1. the bursting of the AI ​​investment bubble and the collapse of associated debt;
2. a sovereign debt crisis that deprives states of fiscal space;
3. a loss of credibility for the dollar that prevents the Federal Reserve from creating money without triggering inflation.

With all three occurring simultaneously, the sovereign loses the means to fulfill its promise of protection at the very moment that promise is called upon. The signs to watch for are concrete: CAPEX cuts by a hyperscaler, rising debt costs, Japanese capital repatriation, an increase in the US term premium, and the trajectory of residential energy tariffs.

The social contract between the sovereign and the people does not break at the moment of collapse; it breaks earlier—when the State ceases to act as an arbiter between the citizen and capital and instead becomes a guarantor of the latter vis-à-vis the former. The collapse merely makes that rupture visible.

Fontes

Mercados, concentração e avaliações

• Companies ranked by market cap e Largest energy companies by market cap, CompaniesMarketCap, 3 out. 2026
• Tech stocks surpass 50% of S&P 500 index for first time, jun. 2026
• How concentrated is the S&P 500 in your index tracker?, com dados da J.P. Morgan, 2026
• Consumer Staples ETFs lead as investors de-risk in early 2026, Trackinsight, fev. 2026
• DeepSeek sparks AI stock selloff; Nvidia posts record market-cap loss, Reuters via Business Standard, jan. 2025
• Buffett Indicator: US corporate equities divided by GDP since 1947, Eco3min, com dados da Reserva Federal (Z.1) e do BEA
• Meanwhile, thirty-five years later, Global Financial Data, sobre o Japão em 1989
• List of countries by stock market capitalization, Wikipedia, 2026

Financiamento, crédito e dívida

• Bond voyage: hyperscaler capex e emissões, Aviva Investors, jul. 2026
• Global AI debt issuance to top $500 billion in 2026, Reuters via MarketScreener, jun. 2026
• AI hyperscalers drive higher US corporate bond issuance, Reuters, jan. 2026
• Hyperscaler debt binge pushes yields up as investor demand cools, Reuters, 29 jul. 2026
• Amazon raises £4.25bn in first sterling bond sale, EU Today, 10 set. 2026
• How hyperscaler issuance is reshaping the Treasury curve, TwentyFour Asset Management, 2026
• Investors see some corporate bonds as safer bets than government debt, Bloomberg, nov. 2025
• Private Credit's $135B Data Centre Bet, set. 2026
• US insurance regulators open inquiries into AI data centre financing, jul. 2026
• Universal Asset Owners: private credit and the AI build-out, jun. 2026, sobre o relatório do FSB de maio
• Nvidia's chips may be novel, but its circular financing isn't, Financial Times, 2026
• Abu Dhabi's MGX raises $50bn, jun. 2026
• Gulf sovereign funds pour into AI and infrastructure, jul. 2026
• The Crossover: Big Tech AI capex crosses cash flow, Scenarica, 2026, sobre a inflação da memória no CAPEX da Microsoft

Moeda, câmbio e dívida soberana

• US Euro Sale to Support Yen Raises ECB Coordination Questions, EU Today, 9 ago. 2026
• European central bankers question US co-ordination after yen intervention, EU Today, 31 ago. 2026
• IMF Data Brief: COFER, primeiro trimestre de 2026, FMI, 1 jul. 2026
• Foreign investors pump $132 billion into US securities in May, dados TIC, jul. 2026
• Foreign investors accelerated purchases of US financial assets last year, dados TIC de 2025
• Long bond yields at multi-decade highs, 18 ago. 2026
• France 10Y bond yield hits near 18-year high, Trading Economics, ago. 2026

Energia, água e materiais

• Projected data center growth spurs PJM capacity prices by factor of 10, IEEFA
• Power costs soar in PJM region as data center demand spikes, Reuters, 2025
• Data centers drove 63% of the largest electricity price spike in the U.S., com base no monitor independente da PJM
• PJM's rate shock and Virginia SB 253, Introl, 2026
• Data centres account for almost a quarter of Irish electricity usage in 2025, Irish Times, jul. 2026
• Ireland's datacentres now consume nearly a quarter of nation's electricity, Computing, 2026
• Data center water wars in Georgia and Arizona, Fortune, mai. 2026
• Data center used 29 million gallons during Georgia drought, com base numa investigação do Politico, mai. 2026
• Data center water use and drought in the American West, Quartz, mai. 2026
• Data center energy use is giving old, dirty power plants a new lease on life, Frontier Group
• 15 coal plant retirements delayed for AI data centers, DeSmog, dez. 2025
• DRAM prices surged by up to 89% in Q2 2026, TweakTown, com dados da SigmaIntel
• Memory shortages raise PC and smartphone prices in 2026, TechSpot, com dados da IDC

Literatura

• Brynjolfsson, Rock e Syverson, "The Productivity J-Curve", American Economic Journal: Macroeconomics, 2021