With contributions from Vivien Lee, Aaron Ma and Eric Savoie
Economic webcast
Our monthly economic webcast for September 2026 is now available: “Tariffs and rising yields.”
The latest macro developments
It has been a busy time for the economy and markets, with a renewed U.S.-Canada trade war rightfully capturing attention, and a tug of war in the bond market between investors and the U.S. Treasury Department. Those two subjects each receive their own section later in this report.
But first, let us briefly touch on two other items of macroeconomic significance, relating to China and Canada.
Weak Chinese economic numbers
The latest monthly Chinese economic numbers were again underwhelming (see next chart). Retail sales missed expectations at just 0.6% year-over-year (YoY) growth. Industrial production similarly failed to meet expectations with a mere 4.5% YoY gain. Fixed asset investment and residential property sales both remained outright lower on the year. China’s composite Purchasing Managers’ Index (PMI) also remained below the important 50 threshold for a second-straight month.
Figure 1: China’s economic activity softened in recent data releases
As of July 2026. Inflation-adjusted retail trade derived using national Consumer Price Index (CPI). Sources: China National Bureau of Statistics (NBS), Macrobond, RBC GAM
None of this is great, obviously. But we are heartened by two things:
As we have repeatedly extolled, China’s longer-term prospects still seem fairly bright: the housing market continues to heal, the country’s exports are diversifying nicely away from the U.S. and there is a remarkable amount of innovation across a range of important sectors.
Seemingly in response to the latest weakness, Chinese policymakers are now talking about delivering more stimulus to boost growth in the coming quarters. Vice Finance Minister Liao Min indicated on August 21 that “new coordinated fiscal and financial policies” are being developed for the second half of 2026 to help growth reach the government’s targets. Similarly, on August 28, China announced a plan to extend maximum mortgage terms from 30 years to 40 years, alongside additional support for property developers.
Temporary Canadian economic revival
After two concerning quarters of economic contraction in Q4 2025 and Q1 2026, the Canadian economy rebounded enthusiastically in Q2 2026, expanding at a 3.3% annualized rate (see next chart). Furthermore, the marginal decline originally announced for Q1 2026 GDP has been revised away. This means the country did not actually breach the technical two-quarter threshold that many use as a shorthand to denote recession. Note that we had argued the underlying economic figures were not weak enough to validate such an interpretation and the latest revisions strengthen our assessment.
Figure 2: Canadian economy rebounded strongly in Q2 2026
As of Q2 2026. Sources: Statistics Canada, Macrobond, RBC GAM
Of course, the news of higher tariffs between the U.S. and Canada complicates and somewhat diminishes the near-term outlook for Canada – and reduces the urgency for the Bank of Canada to raise rates – all discussed in the next section. We were already slightly below-the-consensus in our Canadian growth forecasts. This prompts a further modest downgrade.
-EL
U.S. and Canada trade war
On August 22, the Trump administration followed through on its threat to impose 50% tariffs on US$20 billion of imports from Canada. It had looked like the two countries were close to striking a deal to avoid the new tariffs and reduce existing levies (the U.S. lowering Section 232 sectoral tariffs in exchange for Canada further unwinding retaliatory measures). But talks fell apart at the 11th hour.
Canada has announced retaliatory measures covering a similar dollar value of U.S. imports, which will take effect September 8. The White House is reportedly evaluating further tariffs and possible non-tariff measures in response to Canada’s counter-tariffs, raising the risk of a tit-for-tat trade war.
The new U.S. tariffs, implemented under Section 338 of the U.S. Tariff Act of 1930, cover a diverse range of goods: dairy, alcohol, wood and paper products, agricultural products, plastics, chemicals, textiles, furniture, electronics, machinery and metals. Overall, about 5% of Canada’s exports to the U.S. are impacted. At a 50% rate, that adds 2.5 ppts to the effective U.S. tariff rate on imports from Canada, based on announced tariff rates and historical trade patterns. Canada previously had one of the lowest announced tariff rates among major U.S. trading partners. Its rate is now slightly higher than Mexico’s, but still below the weighted average.
Figure 3: U.S. announced tariff rates, top 20 trading partners
As of 08/24/2026. Based on Bloomberg Economics’ estimate of average tariff rate on U.S. imports from the indicated country computed using 2024 trade composition (HS-10/partner level) and tariff changes implemented to date. Sources: Bloomberg, RBC GAM
Those announced tariff rates don’t necessarily align with actual tariff collection. With American importers substituting away from tariffed goods or exporters reclassifying products to avoid tariffs, the effective tariff rate has fallen short of the announced rate. Based on customs revenue, Canada’s effective tariff rate has been about half of the announced rate (see the gold line in the chart below). Accordingly, we don’t expect this rate will rise by the full 2.5 ppts shown above, as imports of products targeted by new tariffs are likely to decline. Indeed, Canada only supplies about 4% of U.S. imports of the targeted Section 338 goods, suggesting Americans have plenty of other options for those products. As a result, we don’t think the new tariffs will meaningfully increase U.S. inflation.
Figure 4: Section 338 tariffs raise effective tariff rate on U.S. imports from Canada
As at 08/24/2026. Sources: Bloomberg, RBC GAM
On the opposite side, about 80% of Canada’s exports of Section 338 products go to the U.S., suggesting it will be difficult to find new markets for all those goods. Those exports and their upstream supply chains account for about 0.5% of Canadian GDP and employment. If significant layoffs ensue, Canada’s unemployment rate could rise by about 0.4 ppts – back to levels seen earlier this year, and close to the 2025 average.
Canada’s retaliation will also impact the domestic economy. New counter-tariffs ranging from 15-50% (the weighted average is around 30%) cover C$27.6 billion of imports from the U.S. Targeted products include dairy, textiles, metal products, appliances, machinery and transportation. Slightly less than one-third of Canada’s imports of those products come from the U.S., again suggesting some ability of importers to substitute away from targeted goods. Nonetheless, we think the counter-tariffs will add 0.1-0.2% to Canadian inflation in the coming months, based on pass-through of last year’s retaliatory tariffs.
The Canadian economic impact of new U.S. tariffs will be softened by fresh fiscal support for targeted industries. The federal government announced C$7.5 billion in new funding, including enhanced employment insurance for affected workers and liquidity support for businesses. Netting out the economic damage from U.S. tariffs and Canadian counter-tariffs and support from government programs, we think Canada’s GDP will be about 0.2% lower than otherwise if tariffs are sustained.
Our working assumption is that these tariffs will be in place for two quarters and then lifted as some sort of accord is struck, but there are risks on both sides. There are plenty of instances of the White House backing down on tariff threats. Note the brief spike in the blue line above when the administration imposed 10-25% tariffs on all imports from Canada on March 4, 2025, before backing down two days later and exempting USCMA-compliant goods.
Also, given the focus on cost of living heading into the midterms and the unpopularity of tariffs in general, there is political pressure to negotiate a solution. It seemed like the two sides were close to reaching an agreement before, although that was also the case last fall and a deal remains elusive. There are also questions about the legality of the Section 338 tariffs, but this is unlikely to be addressed in the short run.
Conversely, there is also a risk that this devolves into a tit-for-tat trade war that adds to economic damage on both sides of the border. President Trump has already threatened to double auto tariffs and impose new tariffs on auto parts starting January 1. Reports suggest the administration is also evaluating other retaliatory measures in response to Canadian counter-tariffs.
An escalating trade war would be lose-lose for both sides. While we doubt such a scenario would be sustained for an extended period, we can’t dismiss the risk of temporary escalation.
The threat of additional tariffs and reduced protection for USMCA-compliant goods – which were previously exempt from tariffs but are not shielded from the Section 338 duties – creates renewed uncertainty for Canadian exporters. This comes at a time when surveys suggested businesses were beginning to move on from trade policy concerns, with investment intentions picking up (see the first chart below). Manufacturing sentiment has also risen to a 4-year high (see the second chart below).
We’ll be watching to see whether these measures turn over amid renewed trade tensions.
Figure 5: Canadian business investment intentions have been improving
As of Q2 2026. Sources: Bank of Canada Business Outlook Survey, RBC GAM
Figure 6: Canadian manufacturing sentiment is at a 4-year high
As at 08/27/2026. Sources: S&P Global, RBC GAM
The new tariffs partially realize a downside risk to Canada’s economy that the Bank of Canada (BoC) has been flagging for some time. However, the concentrated nature of the tariff impact means a blunt instrument like monetary policy is not the right tool for the job. Fiscal policy can be much better targeted to address the economic fallout. And there could be a timing mismatch between erratic U.S. trade policy and monetary policy’s long and variable lags. Setting interest rates based on tariff rates that could change tomorrow seems unwise.
That said, to the extent tariffs slow growth somewhat and delay absorption of economic and labour market slack, they could keep the BoC on the sidelines somewhat longer than would otherwise be the case. We still think the BoC’s next move is more likely to be a rate hike than a cut, but perhaps not this year.
-JN
Higher yields versus the Treasury
Bond yields have been on an upward trajectory since the year 2020. The yield curve has steepened, with longer-dated bonds experiencing the largest yield increases. The U.S. 30-year yield is now at its highest level in nearly two decades (see next chart).
Figure 7: U.S. long bond yields have risen markedly since the pandemic
As of 08/28/2026. Shaded area represents recession. Sources: U.S. Department of Treasury, Macrobond, RBC GAM
After a rangebound period, the U.S. 10-year yield has also been marching higher in recent months (see next chart). As detailed in an earlier MacroMemo, larger term premiums and higher inflation expectations are the main mechanical reasons for this increase (see subsequent chart).
Figure 8: U.S. 10-year bond yields has been rising in recent months
As of 08/28/2026. Shaded area represents recession. Sources: U.S. Department of Treasury, Macrobond, RBC GAM
Figure 9: 10-year yield change since June 30, 2026
As of 08/28/2026. Term premium and average short-term rates estimated using the Adrian, Crump and Moench model inflation expectations estimated using 10-year inflation swaps. Sources: Federal Reserve Bank of New York, Bloomberg, RBC GAM
These yield increases and the resultant steepening of the yield curve strike us as broadly appropriate given the factors at play.
Inflation: Inflation is proving persistently elevated in the U.S. While one might ascribe much of it to a sequence of unfortunate one-off events – pandemic-related distortions, then tariffs, then an energy shock – the concern is that economic actors are becoming accustomed to the new environment, potentially leaving inflation stalled at an enduringly higher level.
Monetary policy: The bond market remains jittery about new Fed Chair Warsh. His policy of low transparency and not-yet-consistent messaging is creating uncertainty about the path forward for U.S. monetary policy. His most recent speech, at the annual Jackson Hole Economic Symposium, gave a hawkish impression, prompting the market to price in a 65% chance of a September 16 rate hike (see next chart). This pushes short-term yields higher but has an ambiguous interpretation for long-dated bonds. The implication of a higher policy rate should theoretically lower the inflation rate and bolster the Fed’s credibility (theoretically compressing the term premium slightly).
Figure 10: Market expectations of a Fed hike have gone up recently
As of 08/31/2026. Probability estimated based on the fed funds futures. Sources: Bloomberg, Macrobond, RBC GAM
U.S. public policy credibility: Notwithstanding the latest gesture from the Fed, there remain concerns about the credibility of U.S. public policy. This refers both to the capacity for the Fed to tighten monetary policy appropriately given efforts to politicize the decision-making process by the White House, and also concerns about the calibre and volatility of non-monetary public policy decisions. The wider term premium is a partial reflection of such thinking.
Fiscal: Simultaneously, the U.S. fiscal deficit remains enormous, at approximately 6% of GDP (see next chart). The persistent deficits of the past few decades have in turn recently pushed the overall Treasury public debt past an eye-watering US$40 trillion (see subsequent chart). The bond market rightly has concerns about the sustainability of all this. The U.S. continues to rank last in our fiscal health index scorecard.
Figure 11: U.S. federal budget deficit represents 6% of nominal GDP
As of 08/27/2026. Sources: Bloomberg, U.S. Treasury Department, RBC GAM
Figure 12: U.S. Treasury total public debt surpasses US$40 trillion
As of 08/26/2026. Sources: Bloomberg, RBC GAM
Supply: Mechanically, it is difficult to find a home for the rapidly expanding supply of U.S. public debt. This is compounded by two things. First, many countries around the world are running substantial deficits of their own, which compete with one another for funding. Second, AI hyperscalers are expanding aggressively, and having already exhausted their cashflow, are now issuing sizeable, long-dated bonds of their own. This is additional competition for sovereign debt.
Demand: At the same time, the natural demand for sovereign debt has diminished. Critically, the era of quantitative easing is over. The U.S. Federal Reserve and other central banks are no longer in the business of buying trillions of dollars of government debt.
While China still runs a large current-account surplus that must be recycled into something, that something is no longer U.S. Treasuries. Instead, it is increasingly private-sector investments. Japan – another traditional buyer – must now worry about defending its own exchange rate, which involves selling rather than buying U.S. Treasuries.
Indeed many of the world’s currency reserve managers are now more interested in accumulating gold and other assets than Treasuries. This has obliged private-sector investors to fill in the resulting void, and these rate-sensitive investors have helped to drive yields higher.
Counterforces
As it happens, there are a few counterforces that are attempting to hold down long-dated bond yields.
One underappreciated variable is simply that a long-dated U.S. bond now pays around 2% in inflation-adjusted interest – not a bad return. It makes sense that yields are high, but perhaps they don’t need to rise all that much further to keep those private-sector investors appeased.
Another helping hand comes from the fact that the U.S. Treasury has adjusted its issuance schedule over the years to rely less on long-dated borrowing and more on short-dated bonds. This puts less pressure on the long end of the curve and so prevents yields there from distending even further. Of course, that adds pressure to the short end, but the theory is that the fed funds rate can keep this from being distorted too badly.
Still, it makes for an awful lot of short-term bond issuance, as demonstrated by the fact that a startling one-third of U.S. federal debt will mature over the next 12 months (see next chart).
Figure 13: A third of U.S. debt matures in the next year
As of 08/27/2026. Sources: Bloomberg, U.S. Treasury Department, RBC GAM
The third bond yield-depressing force – and this is the new one – is that the U.S. Treasury Department recently committed to doubling its rate of bond buybacks through November 4, from US$20 billion to US$40 billion. The previously scheduled buying amounted to tidying up the bond market – buying back unloved illiquid bonds, essentially. The doubling is patently an attempt to hold down long-dated yields.
Financially, it isn’t a great deal for the government. The U.S. is borrowing at a 4% interest rate at the short end and then using that money to buy back bonds for which it was only paying out perhaps a 2% coupon in the long end (having been issued when interest rates were lower).
But will it work to hold down yields? The debate is between the fact that the extra money is a pittance in the context of a US$40 trillion bond market (0.05% of the total), versus the signaling value that the Treasury is serious about restricting bond yields (and could yet do other things to achieve its goal). Some have pointed to the US$1.0 trillion sitting in the Treasury General Account (see next chart), which might also be deployed in the service of holding down bond yields.
However, that account amounts to the U.S. government’s chequing account. Given the current rhythm of inflows and outflows, it can’t realistically be maintained at much less than US$900 billion.
Figure 14: U.S. Treasury General Account now holds US$1 trillion
As of 08/31/2026. Sources: U.S. Federal Reserve, Bloomberg, RBC GAM
Fundamentally, the reason to be skeptical about the Treasury Department’s ability to significantly restrain the level of bond yields is that it isn’t the Federal Reserve. Central banks have theoretically unlimited ammunition, which means they can be highly persuasive when they even hint about deploying their bazooka, whereas the Treasury Department is working with a BB gun.
We therefore look for current elevated U.S. bond yields to persist, though after the recent selloff they are now in the range of our one-year-out forecasts and so we no longer anticipate a significant further backup.
A not-trivial complication: for all of the bad-mouthing of U.S. bond yields that has occurred here in the context of the various forces exerting upward pressures, Germany, Canada and the UK have all actually experienced a larger increase in their 10-year yield since the end of June (refer back to an earlier chart in this section)! Of course, two out of the three (Germany and Canada) retain a lower yield than the U.S., and that’s the ultimate arbiter of bond market comfort.
-EL
A history of investment booms
As we contemplate the nature and eventual fate of the current remarkable AI investment boom, it is highly instructive to review the characteristics of past booms.
In so doing, it is seemingly standard practice to examine the railway mania of the 19th century, the fibre build-out of the early 21st century and perhaps a small handful of other examples. This is useful, but insufficient.
We have expanded the customary example set to a substantial eleven episodes (refer to table below). While the list is still only illustrative – the definition of a capital boom being somewhat woolly and our examples focusing disproportionately on the U.S. – it is nevertheless revealing in several ways.
Figure 15: A history of 11 major investment booms
As of 08/31/2026. Source: RBC GAM
What induces an investment boom?
Investment booms tend to be initiated in one of three ways, as depicted via the colour-coding above.
There can be a revolutionary new technology that creates new opportunities and enables new markets, attracting large investments. This is seemingly the most common variety, as demonstrated by the canal mania, the railway mania, the Roaring 20s (which was pushed along in part by a series of new technologies including electricity, the automobile and the radio), the mainframe/PC era, the telecom and fibre buildout, the U.S. shale boom, the cloud build-out, and now the AI boom.
Policymakers can induce a boom out of thin air by creating a policy mix that is highly attractive to investors. As an example, the U.S. housing boom (and indeed others like it around the world) was in significant part the result of structurally low interest rates combined with financial industry deregulation. China’s infrastructure boom was the result of a more complex set of policy forces, including rising globalization, focused state-owned bank lending and local government incentive structures.
There can simply be a severe capital shortfall that naturally attracts a sustained surge of investments. This was the case in post-World War II (WW2) Europe and Japan. Both regions were heavily damaged by the war and so effectively offered a superior return on capital relative to other parts of the world. Admittedly, this glosses over important policy choices that were also helpful in those examples, such as the Marshall Plan. But private investments ultimately constituted most of those investment booms.
How long do investment booms last?
Most capital investment booms persist for decades before resolving. Even the shortest in our examples lasted for five years. There is no natural law that demands this, but there is a certain logic to it: it takes time for capital investments to scale to truly heroic proportions and – at least in some cases – to go too far.
With the caveat that modern technological cycles seem to happen more quickly than ever, this would argue for at least another year for the current AI investment boom, and potentially much longer.
Do all investment booms end in tears for investors?
Many do, though not all. By our tally, eight of the eleven examined investment booms created a financial bubble that then burst.
But this count is not entirely fair. The Japanese investment boom was extremely useful for a long period of time, addressing the country’s post-WW2 capital shortfall and modernizing the economy over the span of decades, before becoming a speculative bubble in the final few years. Similarly, China’s infrastructure boom was broadly positive and wealth-creating for nearly two decades before the housing market became too inflated and manufacturing capacity grew too much.
On the positive side, and receiving too little attention in the public discourse, we identify three important examples of a capital investment boom that ended without tears: the post-WW2 reconstruction of Europe, the mainframe/PC investment boom of the 1960s-1980s, and the recent (pre-AI) cloud build-out. It can be done!
Looking for patterns within all of this, capital booms motivated by the desire to fill a pre-existing capital shortfall tend to go better than the others. After all, the additional capital is clearly useful and it is reasonably clear how much more capital is needed to reach a normal level of capital intensity.
Policy-driven capital booms are probably the least favourable. After all, policy distortions induce the boom and often those distortions are later unwound – to ill effect. Although our policy-boom examples show one bad ending and one mixed ending – seemingly not awful – the bad outcome is just one of many international housing boom-bust cycles that could have been included. As such, we would posit that most policy-driven capital booms eventually go wrong, even if there are exceptions along the lines of what China did – identifying an area of potential competitive advantage and then nurturing an industry to the point that it could outcompete the rest of the world.
So where do the technology-induced capital booms land? Of the seven such booms examined, five ended badly for investors. The exceptions were:
the mainframe/PC boom (there were undeniably major corporate players that failed to survive, but this was due to fierce competition rather than excess capacity), and
the recent (pre-AI) cloud build-out that occurred as many companies opted to outsource their IT functions.
The inherent challenge during technology-driven investment booms such as the current AI revolution is in trying to project just how extraordinary the new technology is – and in turn how profoundly and quickly demand will rise. There is a genuine and urgent need for capital and real money to be made, but seemingly there is also a tendency to extrapolate too optimistically after the initial period of impressive growth.
Broadening the definition of a successful investment boom
There are multiple ways to define whether an investment boom has been a success.
Clearly the investor perspective is of particular relevance given that this report is written primarily for investors. On that note, as discussed, most of the CapEx booms have eventually ended poorly for investors.
However, some nuance is necessary. To the extent these booms often run for years or even decades, those who exit before the bubble bursts, or even after it has partially burst, can still do very well.
The bubbles also frequently do not fully unwind the prior boom, leaving a net benefit even if it is diminished relative to its peak. An investor who entered the Japanese stock market in December 1984 – decades after the start of the economic boom and a mere five years before the peak – would still have earned a cumulative capital gain of 71% through December 1994 – five years after the crash began. Similarly, even investors who entered China’s housing market just a few years before the 2021 peak would still be marginally up on their investments today.
While of little consolation for investors, even challenging investment boom-bust cycles can still be quite good for society as a whole:
While much of the money invested in fibre-optic cables and related infrastructure was lost during the tech bust of the early 2000s, that infrastructure proved helpful in lowering telecom prices for consumers and ensuring sufficient capacity for the later explosive growth in the internet and mobile computing.
It is beyond dispute that the large-scale construction of canals and railroads was enormously positive in terms of reducing the cost of transportation – previously a key friction for commerce. It also enabled the settling of much of North America.
The U.S. shale/energy CapEx boom was overdone but left in its wake a new American industry that has lowered energy costs, employed Americans, generated profits and reduced the clout of OPEC.
Japan and China both managed to increase their living standards massively during their booms. These gains only minimally unwound in Japan’s case and not at all in China.
Of course, when an investment boom is large enough that the subsequent bust induces a recession or significant economic recoil, its redeeming characteristics often pale in comparison. As an example, not much positive can be said about the U.S. housing boom given the carnage it left in its wake.
Investment boom conclusion
In conclusion, more investment booms end in tears than not, but usually not for many years. Some work out without problems or resolve with asset valuations that, while diminished, are still higher than before the boom started. Technology-driven cycles such as the AI boom presently underway are particularly tricky to anticipate, as demand genuinely rises but it is hard to pinpoint at what point capacity has risen too much.
Finally, of little consolation to investors but welcome nevertheless, even unpleasant boom-bust cycles often leave positive legacies for the economy and society.
-EL
AI circular financing and off-balance sheet liabilities
Tremendous capital spending by a handful of key players is at the heart of the artificial intelligence (AI) infrastructure buildout that is enabling large-language and generative-AI models to run at scale. It is also a material driver of U.S. economic growth.
Financing this buildout is no small feat. Investors are becoming increasingly concerned about arrangements between chipmakers, hyperscalers and other AI companies that are substantially funding each other, creating loops of capital among them that raise questions as to whether seemingly strong demand is real or inflated. At a minimum, these actions constitute large bets that end-user demand will be sufficient to make all the investment worthwhile.
This financing circularity has concerning parallels to the late 1990s technology bubble when companies that helped build out the internet backbone offered financing to their customers so those customers could purchase telecommunications equipment from them.
Today, reflecting its optimism about future AI demand, NVIDIA is committing significant capital to secure memory chips that are increasingly seen as the bottleneck for AI compute capacity. During its latest earnings release, NVIDIA announced US$279 billion in commitments to its suppliers for their memory chips. This is more than double the company’s US$119 billion commitment from a quarter ago (see chart below).
Figure 16: NVIDIA has more than doubled its spending commitments
As of 08/27/2026. NVIDIA fiscal year ends in January. Sources: Wall Street Journal, NVIDIA investor relations, RBC GAM
Of this amount, most of the commitments are due over the next three years. The company has committed US$92 billion of spend for the rest of 2027, US$87 billion in 2028 and US$88 billion in 2029 (see table below).
Figure 17: Most of NVIDIA’s future commitments are due over the next three years
NVIDIA Corporation - Future commitments by fiscal year in billions of USD
As of 07/26/2026. Sources: Global and Mail, NVIDIA investor relations, RBC GAM
In addition to its own committed spending on memory chips, NVIDIA has announced hundreds of billions of dollars in other guarantees helping client companies secure property and energy to build data centres and expand AI compute. Such data centres would then be populated with NVIDIA chips. This has sparked fear NVIDIA is engaging in circular financing akin to the experience seen in the 1990s internet bubble.
Clearly NVIDIA believes its financing of the AI build-out is necessary to support the massive demand for its products. The company is quickly becoming the key financing engine of the broader AI ecosystem. In early August, NVIDIA also announced a US$500 billion financing deal with a variety of asset managers that could be seen as an attempt to transform AI data centres into an investable asset class of their own.
NVIDIA isn’t the only company making significant commitments, though, and many of these material contractual promises don’t get recorded on company balance sheets. The chart below plots the spending commitments of the four major hyperscalers (Amazon, Microsoft, Meta and Alphabet) and it reveals the significant size of obligations that are not officially counted as liabilities. In total, this group has US$248 billion of lease liabilities and US$356 billion of long-term debt that are recorded on their balance sheets. But they also have a combined US$904 billion in leases that have not yet started and US$1.52 trillion in future purchase commitments.
Figure 18: Hyperscalers have made significant spending commitments
As of August 2026. Data are as of each company’s most recent quarterly filing. Sources: Wall Street Journal, RBC
Moreover, these off-balance-sheet commitments have been growing at a staggering pace. The chart below plots the growth over the past 12 months in hyperscaler capital spending and off-balance-sheet commitments. While capital spending has been growing rapidly, rising between 60% to 160% over the past year depending on the company, off-balance-sheet commitments have grown anywhere from 91% to 834% over the same period.
Figure 19: Hyperscalers making massive off-balance-sheet commitments
All data as of June 2026 except NVIDIA (April) and Meta (July). Sources: Wall Street Journal, RBC GAM
These massive commitments present a significant risk for these companies should the demand for AI undershoot expectations. That’s because companies are building massive data centres with private equity and/or private credit investment, which is secured by a commitment that a hyperscaler will lease the data centre for an extended period.
One such example is the Hyperion data-centre project in Louisiana. Here Meta has committed to leasing the data centre for a four-year term starting in 2029, with an option to renew for up to 20 years. Meta has guaranteed to make the bondholders of that project whole even if it doesn’t extend the lease. This binding agreement doesn’t yet show up on Meta’s balance sheet because the lease has yet to begin.
While hyperscalers generate significant cash from their operations, increasing AI-related spending and growing future commitments are leading investors to assign those companies a higher chance of defaulting on their obligations. As per the chart below, the 5-year credit-default-swap (CDS) spread of a basket of AI hyperscalers has widened significantly since late 2025 as overspending concerns have taken centre stage. That said, these spreads remain far below what would be considered distress (i.e. above 1000 basis points). But the trend of widening spreads over the past year suggests there may be a limit to how much spending investors are willing to tolerate.
Figure 20: AI hyperscaler 5-year credit-default-swap spread has widened significantly
As of 08/28/2026. AI hyperscalers include AMZN, GOOGL, META, MSFT and ORCL. GOOGL and META CDS spreads before 11/21/2025 are backfilled using the change in AMZN, MSFT and ORCL spreads. Sources: Bloomberg, RBC GAM
Overall, it appears that AI is a powerful technology on track to have important industrial and commercial applications. The technology’s already-impressive growth from both a usefulness and capital intensity standpoint is accelerating. The opportunities for productivity gains and new capabilities are certainly exciting. But it is worth recognizing that a handful of companies are collectively betting trillions of dollars that ample revenue from end users lies on the other end of this massive investment cycle.
Should demand be less than expected, companies that have made massive financial commitments could stand to lose out, with cascading consequences across an interconnected industrial ecosystem.
-ES
CapEx to broaden beyond AI
Spending on AI data centres remains an outsized contributor to U.S. business investment growth. Estimated CapEx by the big 5 hyperscalers continues to be revised higher with nominal spending expected to nearly double this year. The consensus estimate for 2027 is closing in on $1.1 trillion, up from $600 billion estimates at the start of 2026.
Figure 21: Hyperscalers are expected to continue ramping up CapEx
As at 08/27/2026. Sources: Bloomberg, RBC GAM
Meanwhile, there has been scant growth in business investment outside of tech over the past three years (see next chart). In other words, the CapEx impulse driving the economy has been remarkably powerful but narrowly based.
Figure 22: AI-related CapEx is booming while non-AI investment has stagnated
As at 08/13/2026. AI investment includes data centre and power structures, software, and computers and peripherals. Sources: U.S. Bureau of Economic Analysis, RBC GAM
But that story may be changing. Non-AI CapEx showed some signs of life in the first half of 2026 (note the recent uptick in the gold line above). Analysts expect broader growth in investment going forward among the biggest U.S. companies. The chart below shows that 45% of S&P 500 companies (the fraction for which estimates are available) are expected to boost nominal investment by more than 10% over the next 12 months. That’s the best breadth in nearly two decades of history, outside of post-recession rebounds.
Figure 23: Nearly half of S&P 500 firms are expected to increase CapEx by >10%
As at 08/14/2026. Financial companies are generally excluded as no CapEx estimates are available. Sources: Bloomberg, RBC GAM
Some of that broadening reflects spillover from hyperscaler CapEx. Our bottom-up aggregation of analysts’ CapEx estimates for 38 S&P 500 suppliers that are central to the AI buildout points to a 51% increase over the next 12 months. That number builds on a 26% increase over the past year (see the chart below).
Outside of the hyperscalers and their suppliers – and excluding real estate where investment is particularly volatile – analysts expect 10% CapEx growth over the next 12 months, led by consumer staples, tech, utilities and industrials.
Figure 24: AI suppliers are ramping up CapEx, decent growth expected outside AI
As at 08/18/2026. Sources: Bloomberg, RBC GAM
There are good reasons for analysts’ optimism on non-AI CapEx:
Bonus depreciation: Last year’s One Big Beautiful Bill Act (OBBBA) permanently reinstated 100% depreciation on qualifying capital assets like machinery and equipment, computers and data centre infrastructure. It also allows immediate expensing of manufacturing structures built by the end of 2030. According to UBS, these incentives increase the internal rate of return on a new manufacturing facility by 50%. Researchers estimate bonus depreciation under the Tax Cuts and Jobs Act (TCJA), which was in place from 2017 to 2022 and phased out starting in 2023, increased business investment by 2.5-4%.
Deregulation: President Trump kicked off his second term by revoking many of the Biden administration’s executive orders and requiring agencies to repeal 10 existing regulations for every new regulation issued. Changes in energy, environmental, labour, health and financial rules are creating a more business-friendly regulatory environment. Bloomberg’s tracking of S&P 500 earnings calls shows improving sentiment regarding regulation during Trump’s second term, which has coincided with a more optimistic tone on capital spending.
Figure 25: Earnings calls point to improving sentiment regarding CapEx and regulation
As at 08/14/2026. Sentiment score is an average of management discussion and analyst question scores. Sources: Bloomberg, RBC GAM
Reshoring incentives: At the aggregate level, tariffs have not been successful in reshoring manufacturing activity to the U.S., at least thus far (see chart below). But tariffs have contributed to some companies’ decisions to expand manufacturing in the U.S. Pharmaceutical and semiconductor/electronics producers in particular have announced significant new investments in response to tariff threats of 100% or more. The Biden-era CHIPS and Science Act has also supported new construction of computer and electronic manufacturing facilities. Companies need to break ground by the end of this year to take advantage of those tax incentives.
Figure 26: Limited evidence of sustained reshoring as U.S. tariff rate has increased
As at 08/18/2026. Manufacturing self-sufficiency = (manufacturing gross output - exports) / (imports – re-exports). Sources: Kearney Reshoring Index, U.S. Census Bureau, RBC GAM
Lending conditions: Demand for business loans has picked up in recent quarters and lending standards are no longer tightening according to a Federal Reserve survey (see chart below). Given the large, lumpy nature of CapEx investments, a friendly lending environment is helpful. A significant increase in hyperscaler debt issuance this year has generated concerns about a supply surge pushing interest rates higher and crowding out other borrowers. But investment grade bond spreads have widened only modestly and remain tight overall.
Figure 27: Business loan demand is rising and lending standards are no longer tightening
As at 08/27/2026. C&I = commercial and industrial. Sources: U.S. Federal Reserve, RBC GAM
Buoyant equity markets: Rising equity prices also represent an easing in financial conditions. Businesses appear to be taking advantage by increasing equity issuance and reducing share buybacks. Net equity issuance turned positive in Q1 and rose to a post-Global Financial Crisis high when M&A-related share retirements are excluded (see chart below). These additional funds are supportive of large business outlays such as on CapEx.
Figure 28: U.S. non-financial corporations are shifting from buybacks to net issuance
As at 08/13/2026. Net issuance ex M&A removes equity retirements related to cash-financed M&A. Sources: U.S. Federal Reserve, RBC GAM
Improving demand: The Institute for Supply Management (ISM) manufacturing index rose to a four-year high in July, helped by robust growth in new orders. Manufacturing capacity utilization is starting to stabilize and breadth is improving. Roughly half of industries are seeing utilization above their 3-year average (see chart below). Nominal capital goods orders excluding aircraft were up 16% year-over-year in Q2, the fastest growth since 2010.
Figure 29: Manufacturing capacity usage is stabilizing and breadth is improving
As at 08/14/2026. Sources: U.S. Federal Reserve, RBC GAM
That said, there are some headwinds to non-AI investment. The AI buildout is driving up material costs and construction wages. Further hyperscaler debt issuance could add to financing costs for other companies. Higher government bond yields are also putting upward pressure on overall borrowing costs. And as we highlighted in a recent midterm election preview, if Democrats win a majority in the House of Representatives (as seems likely), the regulatory environment could become marginally less business friendly.
Overall, we think the ingredients are in place for a broadening out of U.S. CapEx. AI-related spending is likely to remain an essential driver of business investment growth, but our above-trend U.S. GDP growth forecast incorporates some improvement in non-AI CapEx as well.
What about outside the U.S.? When it comes to AI, UBS notes the U.S. remains the epicentre of data centre construction with 45% of global operating capacity and 60% of planned capacity. But the UK, Spain, India and China all have sizeable pipelines of data centre projects.
Countries that are heavily involved in the AI buildout are seeing a pickup in CapEx. South Korea, for instance, home to the top two global memory producers, recorded 12% annualized growth in business investment in the first half of last year. This reverses a multi-year trend of slowing CapEx (see chart below). Reduced political uncertainty and regulatory reforms are also supporting business sentiment and investment.
Figure 30: South Korean CapEx is picking up
As at 08/27/2026. Sources: Bank of Korea, RBC GAM
But beyond AI, there is limited evidence of a broader international CapEx push. Only about one-third of countries are seeing above-average growth in inflation-adjusted fixed investment (see the first chart below). And forward-looking CapEx intentions aren’t particularly inspiring. S&P’s global manufacturing CapEx outlook indicator remains subdued, even as PMI indices have improved and capacity utilization is rising (see the second chart below).
Figure 31: CapEx growth isn’t particularly broad-based globally
As of Q1 2026. Based on a sample of up to 63 countries. Sources: International Monetary Fund, RBC GAM
Figure 32: Global CapEx intentions remain subdued despite rising capacity utilization
As at 08/27/2026. Sources: S&P Global, RBC GAM
There are several factors that might be restraining global CapEx:
Competitiveness challenges and Chinese overcapacity
Higher energy prices and material costs
Rising interest rates
Sluggish housing markets
Political and policy uncertainty
Protectionism and geopolitical tensions
Despite those headwinds, we think there is some underlying impetus to increase investment over time amid population aging, climate change and with governments prioritizing national defense and economic resilience in the face of geopolitical shifts. But outside of some broadening in the AI CapEx super cycle, there is little evidence of a widespread international CapEx upswing at this point.
-JN
Oil market adapting but concerns remain
A sustained peace deal between the U.S. and Iran remains elusive and the flow of energy through the Strait of Hormuz is still restricted. Yet oil prices have been below $100 per barrel for much of the past three months, defying concerns that dwindling inventories would send prices sharply higher, or that damaging demand destruction would wreak havoc on the global economy.
Supply adjustments have done much of the heavy lifting, with a smaller demand adjustment concentrated in a handful of regions. But for all the flexibility shown by the global oil market so far, the current equilibrium is precarious. Refined product markets are showing signs of stress. While not our base case, the risk of higher prices remains in the absence of an agreement to further open the Strait.
The global oil market has proved remarkably resilient and adaptable to the worst supply shock in history. The chart below shows how a disruption equivalent to an incredible one-fifth of global oil supply – the fraction of oil that would normally transit through the Strait of Hormuz – has been redirected and buffered by other sources of supply, resulting in a relatively moderate damage to demand.
Figure 33: Both supply and demand have adjusted to Strait of Hormuz disruptions
As at 08/19/2026. Sources: U.S. Energy Information Administration (EIA), International Energy Agency (IEA), RBC GAM
Hormuz exports and “dark” shipments
Pre-conflict, roughly 21 mb/d of crude oil and petroleum liquids was transported through the Strait of Hormuz. Based on EIA estimates using transponder data and satellite imagery, those exports were down 16 million barrels per day (mb/d) year-over-year in Q2. But that data doesn’t necessarily pick up all ships passing through the Strait.
The IEA estimated “dark activity” exiting Hormuz via the Omani coastline was nearly 2 mb/d in June, leaving total Hormuz shipments at 6-7 mb/d. In August, U.S. Energy Secretary Chris Wright claimed 8-9 mb/d of oil is actually passing through the Strait, although those numbers haven’t been verified by private tracking and seem high.
Bypassing the Strait
Saudi Arabia has diverted some of its oil exports to the Red Sea via its East-West Pipeline. Shipments through the Bab el-Mandeb Strait, Suez Canal and SUMED Pipeline increased by more than 4 mb/d in Q2 compared to year earlier. The United Arab Emirates (UAE) has increased pipeline exports to its Fujairah port on the Gulf of Oman, just outside the Strait. But with few other egress options and limited storage space, the IEA estimates more than 8 mb/d of Gulf production was still shut-in as of July.
Recently, renewed attacks on ships in the Red Sea by Iran-backed Houthi militants temporarily disrupted shipments through alternative routes (see chart below).
Figure 34: Oil shipments through the Red Sea have made up for some of the Strait of Hormuz decline
As at 08/27/2026. Sources: Bloomberg, RBC GAM
Production increases and inventory withdrawals
Oil production outside the Gulf has increased by about 2 mb/d. This was led by higher output in the U.S. and Brazil, and to a lesser extent Canada, Kazakhstan, Guyana and Venezuela.
Inventory withdrawals have supplemented that production increase – observed oil stocks have declined by 2.7 mb/d since the start of the conflict. The U.S., Japan and China accounted for most of the drawdown. The U.S. has withdrawn 122 mb from its Strategic Petroleum Reserve, which is now at its lowest level since the early 1980s. In the year prior to the conflict, China reportedly added 200 mb to its inventory, while Japan maintained the world’s largest stockpile measured in days of demand. Both were well positioned heading into the supply shock.
Demand destruction
The net result is a supply shortfall – and so required demand destruction – of less than 4 mb/d. Developing economies have accounted for most of the decline in consumption, led by Asia where demand-limiting measures have been most widespread (see chart below). China has been the biggest single source of consumption restraint thanks to rising adoption of electric vehicles, growth in renewable power generation, and commuters and travelers opting for public transit. Between lower domestic consumption, a pullback in refined product exports (which shows up as demand destruction in other countries), and a shift from inventory building to withdrawals, China has reduced its oil imports by more than 5 mb/d, acting as a key swing purchaser to help balance the global oil market.
Among developed economies, it’s the big oil importers – Japan and Europe – that have adjusted consumption most significantly (by 9% and 3.5%, respectively). Demand in the U.S. and Canada was little changed despite higher gasoline prices.
Figure 35: Developing Asia is taking the most measures to limit energy consumption
As at 08/18/2026. Sources: International Energy Agency, RBC GAM
The key question is whether the energy market has reached a sustainable equilibrium of moderately elevated prices, or instead a shaky balance that could eventually send prices higher and necessitate further demand destruction. There are risks to some of the current supply mitigants:
Inventory withdrawals can’t continue perpetually. The International Energy Agency (IEA) estimated global observed oil inventories were below 7.9 billion barrels in July. Compare that with recent lows close to 7.6 billion barrels in March 2022 (shortly after Russia’s invasion of Ukraine). At the recent pace of withdrawals, inventories would fall to that level – which JP Morgan analysts characterize as operational stress levels – in about 4 months. If inventories are tapped out, global supply could fall by 2-3 mb/d.
The U.S. claims significantly more oil is being transported through the Strait of Hormuz than official tracking suggests. However, “dark” exports might not be sustainable in the absence of a peace deal. That could threaten 2-4 mb/d of supply.
Renewed attacks on ports and vessels in the Red Sea only temporarily disrupted alternative export routes through the Bab el-Mandeb Strait and Suez Canal. But if an intensifying regional conflict causes sustained declines in Red Sea exports, another 2-4 mb/d of supply could be at risk.
The longer production in the Gulf remains shut in, the greater the risk of damage to wells that then delays or precludes a return to pre-conflict output once transportation constraints on the Strait normalize.
Supply of crude oil isn’t the only issue. Gulf states are also key producers of refined products. The IEA estimates refinery throughput was down nearly 5 mb/d year-over-year in July. Ukraine’s attacks on Russian refineries added to the shortfall. Diesel exports from the Middle East, Russia and Asia were down 1.3 mb/d in July (equivalent to 20% of global seaborne trade). Jet fuel exports were nearly 0.7 mb/d lower (one-third of global trade).
Relative to crude oil, prices for those products are significantly higher compared with pre-conflict levels (see the first chart below). Refined product shortages have pushed refining spreads to record highs (see the second chart below).
Figure 36: Refined product prices remain elevated relative to crude oil
As of 08/27/2026. Sources: Bloomberg, RBC GAM
Figure 37: U.S. refining spreads are at a record high
As at 08/27/2026. The 3:2:1 crack spread approximates the product yield at a typical U.S. refinery: for every three barrels of crude oil the refinery processes, it makes two barrels of gasoline and one barrel of distillate fuel. Sources: Bloomberg, RBC GAM
That has helped ease fears of a supply shortage during the summer travel season, and European jet fuel inventories appear to have stabilized, albeit at a seasonally low level (see chart below). But those shifts have put pressure on supply of other products. U.S. diesel inventories are at a three-decade seasonal low and gasoline stocks are the lowest since 2012 for this time of year.
Figure 38: European jet fuel inventory has stabilized but remains below seasonal norms
As of the week ending 08/20/2026. Amsterdam-Rotterdam-Antwerp (ARA) jet inventory levels. Sources: Bloomberg, RBC GAM
Supply losses are still outpacing the decline in demand. If inventories continue to be drawn down, refining spreads could widen further and refined product prices might rise even if crude oil prices remain stable. Again, that could necessitate further demand destruction to help balance the market. Developing economies would likely continue to shoulder a disproportionate burden of any demand adjustment, but advanced economies might not be immune – including in North America where consumption has been relatively steady thus far.
Overall, there is probably more cause for concern than crude oil prices suggest. Refined product markets still look fragile, even if the worst fears of summer supply shortages haven’t been realized. Our base case is still for modest easing in energy price pressure in the coming months. But it’s hard to dismiss the risk that prices could rise further into year-end if a deal to further reopen Hormuz isn’t reached.
-JN