With contributions from Vivien Lee, Aaron Ma and Eric Savoie Quick hits Webcast Our monthly economic webcast for November is now available, entitled “ Election season / AI worries ”. Yields reverse course After the rapid and seemingly unstoppable rise of long-term borrowing costs in August and September, the past month witnessed a profound – if incomplete – reversal (see next chart). Figure 1. Bond yields reverse course As of 10/30/2026. Shaded area represents recession. Sources: U.S. Department of Treasury, Macrobond, RBC GAM A variety of minor forces have likely helped this decline, including risk-off attitudes resulting from the tech selloff, a small retreat in oil prices and a reduction in Fed expectations. But the most compelling reason has to do with positioning. It was hard to explain why bond yields were rising so far and so quickly over the prior few months. Momentum was dominating valuations. Once yields started to wobble, there was overwhelming pressure for the reversal that has since transpired. Bond yields will no doubt ebb and flow from here. This is absolutely a higher interest rate environment than what prevailed in the 2010s. We are comfortable with present levels. -EL Tech meltdown The stock market has slumped over the past month (see next chart), with weakness concentrated within tech-related stocks. AI stocks have been a particular concern, especially neocloud companies. These are data-centre operators that lack the scale or balance sheet strength of the hyperscalers. Figure 2. Stock market slump led by tech stocks As of 10/30/2026. Sources: Morgan Stanley, Bloomberg, RBC GAM This weakness has also been visible in the credit space. The lowest-rated CCC and below credits – which include some of the aforementioned neoclouds – have underperformed particularly sharply (see next chart). Figure 3. U.S. lower quality bond spread has widened sharply As of 10/31/2026. Sources: Bloomberg, ICE BofA, Bloomberg, RBC GAM Recent AI concerns have dwelt on the risk that demand for AI might be weaker than previously imagined. We tackle this subject later in this #MacroMemo, arriving at a nuanced but ultimately cautiously constructive conclusion. -EL French fiscal distress French bond yields have soared over the past several months. The country’s 10-year yield has risen from just over 3.50% at the start of 2026 to nearly 5.00% today (see next chart). Figure 4. French bond yields spike beyond Global Financial Crisis levels As of 10/29/2026. Shaded area represents euro area recession. Sources: Macrobond, RBC GAM This has taken France’s financial conditions index to the worst level since the Global Financial Crisis, representing a substantial economic drag (see next chart). Figure 5. France’s financial conditions index approaches peak during Global Financial Crisis As of 10/29/2026. Sources: Goldman Sachs, Bloomberg, Macrobond, RBC GAM Why is this happening? Our fiscal health scorecard identifies France as having the second worst fiscal position among major nations (see next table). Only the U.S. is worse, but it is saved by its reserve currency status, deep financial markets and faster pace of economic growth. Figure 6. Our fiscal health scorecard shows U.S. and France are most challenged 2025 data for all indicators except interest payments (2024) and GDP growth (International Monetary Fund forecast for 2031 used as proxy for "normal"). Fiscal adjustment refers to the necessary reduction in fiscal deficit to stabilize debt-to-GDP ratio. Source: IMF, Macrobond, RBC GAM France’s fiscal deficit has been expanding over the span of decades (see next chart), reflecting serial failures to bring the fiscal account into balance. Figure 7. France’s fiscal balance has deteriorated significantly As of Q2 2026. Shaded area represents euro area recession. Sources: French National Institute of Statistics & Economic Studies (INSEE), Macrobond, RBC GAM Paired with the civil unrest recently experienced in the country – reflecting public desire for more government spending at a time when fiscal consolidation is essential – the immense challenge facing the country is clear. Debt servicing costs are now the single largest line item in the budget. Government spending is now more than 57% of GDP. This is the second highest in the Eurozone and above the euro-area average of 51%. Pension commitments in France are more than 13% of GDP alone (versus a euro-area average of under 11%). It is especially hard to cut pension spending. As the government grapples with these problems, unpopular austerity is likely to further bolster the prospects of France’s far right party, which already leads in the polls before next April’s presidential election. This possibility further concerns the French bond market and simultaneously raises questions about the integrity of the European Union (EU). While we do not expect a full Greek-style meltdown, there is not a lot of room to maneuver. This is likely to remain a problem, with persistently elevated French bond yields and a diminished French economic outlook. -EL Final midterms preview The U.S. midterm election comes on the heels of a spurt of solid U.S. economic data. This includes a robust 3.4% annualized GDP gain in the third quarter and an October Institute for Supply Management (ISM) Manufacturing print that rose to a strong 55.5. At the time of writing, midterm betting markets point to substantial change ahead, including: a 95% chance that the House of Representatives transitions to the Democrats an 88% chance that the party captures the most Governorships a 74% chance that they pick up the Senate. Of course, prediction markets are hardly failsafe. They badly misjudged the Brazilian election just last week. It is notable that Republican support was underestimated in each of the 2016, 2020 and 2024 presidential election cycles. But that isn’t the whole story: the past two midterms – 2018 and 2022 – experienced little bias; and the most recent (off-cycle) elections in 2025 actually underestimated the Democrats. Furthermore, do not forget that pollsters are perpetually working to neutralize biases that surfaced in prior elections. In conclusion, surprises are possible, but it isn’t necessarily the case that Republican support is being undercounted this time. The following graphic summarizes our policy expectations and the medium-term macro and market implications that emerge from the midterms (see next graphic). We don’t believe the midterm elections should be viewed as a central investing theme, but at the margin it could marginally soften GDP and the stock market, lower inflation, and put downward pressure on yields. The dollar could be slightly weaker as well. Figure 8. Divided U.S. government ahead? Implications (at the margin) 10/30/2026. Source: RBC GAM For more details, refer back to the policy preview we penned in July: “ U.S. midterms preview ”. Of course, many of the biggest policy decisions of the past two years were made via White House edict rather than legislation, including major tariff and foreign policy decisions. That capability won’t be significantly dimmed by the midterms. What about short-term implications in the weeks after the election? From a stock market standpoint, refer to the technical analysis we did in September, highlighting that the stock market often underperforms in the lead-up to the midterms (as it has), and then frequently outperforms over the final two months of the year. There is a considerable likelihood that the conflict with Iran intensifies once the pocketbook concerns dominating the election campaign can be set aside. There are certain policy actions that the government will hope to take during the lame duck session of Congress from November 4 through January 3: A large swath of government funding expires on December 11, so Congress will presumably seek to extend that into the New Year. Likewise, some government funding, including certain healthcare provisions, expires on December 31. The existing government should extend these. The Republican Party will likely seek to pass defence spending legislation for 2027 before the Democrats gain control (if they do). The Republican Party has indicated it hopes to raise the debt ceiling before the end of the year, though pundits assign a relatively low probability to this. If not addressed, it will become a pressing issue next spring. Conversely, a final major Republican fiscal package of tax cuts and the like is unlikely as Republican leadership is doubtful there are the votes necessary at this late juncture. -EL Oil market update Oil prices remain quite elevated (see next chart). Figure 9. Crude oil prices remain high As of 11/02/2026. Sources: Bloomberg, RBC GAM This is not entirely appropriate based on the current physical flow of oil. Middle East oil flows ex-Iran have actually now fully rebounded to pre-war levels. That still leaves the bulk of Iran’s 2 million barrels per day missing. However, this is a relatively small hole in the grand scheme, especially as other parts of the world have increased their production. How have Middle Eastern exports revived so substantially? There are two main contributors. Despite widespread proclamations that it would take years to rebuild damaged energy infrastructure, this has occurred faster than expected. Countries have also managed to increase production at undamaged facilities to the point that they have significantly offset losses at the damaged ones. Crucially, Saudi Arabia’s East-West Pipeline is back to 80%-plus capacity after having been briefly taken out of service by Houthi strikes. U.S. military escorts of ships through the Strait of Hormuz have seemingly proven highly effective. The U.S. military says it has helped over 3,000 commercial vessels through the Strait over the past six months. The data shows a sharp increase in Middle East exports from the likes of Iraq over the past few months in particular. This is not to say that transit has become risk-free. There have been more than 35 ’incidents’ involving ships attacked by Iran in the month of October alone. But Iran’s ability to accurately track and thus attack ships has been significantly degraded. Fascinatingly, the financial calculus is now quite attractive for ships transiting the Strait. Traders stand to earn an additional US$20 risk premium per barrel of oil, which amounts to about $40 million per supertanker. A second-hand tanker only costs $100 million, meaning that shipowners can theoretically remain profitable even if there were a one in three chance of the ship being destroyed per (two-way) transit. In practice, only about 5% of traffic has been damaged by Iranian strikes, meaning this is a good financial bet. Shipowners have even been relocating older vessels that might once have carried Venezuelan or Russian oil to the Middle East. Many of these vessels are making frequent, short runs through the danger zone before transferring their oil to other ships. What about the sailors? They are being offered two to three times their normal monthly salary, and this appears to suffice to keep the ships staffed. Looking ahead, there is the aforementioned risk that the war itself intensifies after the midterm elections. But oil prices already significantly reflect this risk. That’s why they are still elevated even though most of the oil is actually flowing. So, it is far from automatic that the price of oil rises from here, even if the war intensifies. Meanwhile, Iran re-engaged with the U.S. in negotiations starting on November 2. This highlights the possibility that economic sanctions could be starting to wear the country down. In view of this, we would argue that the risks surrounding the current price of oil extend more prominently downward than upward at this juncture. -EL AI demand: if you build it, will they come? The AI CapEx cycle remains central to the U.S. economy’s resilience and the health of domestic and global equity markets. The big five hyperscalers are expected to invest more than $5.5 trillion between 2026-2030. Nearly three-quarters of that will go toward AI data centres. Expected future earnings of those spenders and their major suppliers are contingent on following through with and monetizing that investment. As CapEx expectations have risen, doubling over the past year, (see the chart below), so have the stakes for investors. Figure 10. Estimates for hyperscaler CapEx have been revised significantly higher As at 10/23/2026. Hyperscalers include AMZN, GOOGL, META, MSFT and ORCL. Sources: Bloomberg, RBC GAM There are a few ways shareholders could be disappointed. One scenario is constrained supply: what if hyperscalers aren’t able to add enough compute in a timely manner? For the spenders themselves, less CapEx means slower AI revenue growth, all else equal. For their suppliers, it means reduced chip deliveries, less networking, cooling and electrical equipment, fewer construction contracts and leases, etc. We’ve examined a number of issues that could slow the data centre buildout, including computer-chip shortages, public and political opposition to data centres, and whether the supply of electricity can keep up. But recently, investor concern has shifted from supply constraints to worries about excess investment and a potential demand shortfall. What if AI adoption and spending aren’t sufficient to deliver a reasonable return on invested capital (ROIC)? Recent warning signs include: disappointing revenue guidance from Anthropic a widening gap between hyperscaler CapEx and AI revenue (resulting in shrinking ROIC and declining/negative free cash flow) worries about the health of neocloud companies (AI-specific cloud service providers like CoreWeave) declining average token prices. The latest quarterly earnings also raised concerns about some hyperscalers’ non-AI revenue and their ability to fund ongoing AI CapEx. As we discuss below, the latest demand indicators still point to rising AI adoption and a persistent shortage of compute. However, recent estimates of the future revenue required to justify expected CapEx point to a rather high demand hurdle. There is a wide range of such estimates, though, reflecting significant uncertainty and sensitivity to underlying assumptions. We think reasonable estimates set a high but not insurmountable bar for AI adoption and spending. Markets are clearly on edge, though – see the recent negative reaction to Meta’s higher CapEx guidance. More evidence of AI’s rapidly evolving capabilities and deeper integration into business processes is needed to assuage demand worries. Current demand indicators We previously examined falling token prices , concluding that a decline in the market’s weighted average token price likely reflected users shifting toward cheaper, open weight (mostly Chinese) models rather than a broad softening in demand. Since then, the weighted average token price has declined significantly further. It is now below levels seen at the start of the year (see chart below). Figure 11. Weighted average token prices doubled then fell by half As at 10/23/2026. Sources: Silicon Data, Bloomberg, RBC GAM This has indeed been accompanied by further market share gains for open weight models. According to OpenRouter (a platform that allows users to access a variety of AI models), open models accounted for around 60% of token consumption in August and September. That’s twice the share seen earlier this year. That sounds like bad news for closed model providers. But with a rapid increase in overall token consumption – nearly 7x in the past 6 months – both open and closed models are seeing significant demand growth (see chart below). Figure 12. Open models have doubled their share, but both are growing rapidly As at 10/16/2026. Sources: OpenRouter, RBC GAM Data from Ramp, a financial operations platform, similarly shows exponential growth in its clients’ token consumption. Almost all of this consumption comes from closed model providers OpenAI and Anthropic – see the blue line in the chart below. However, growth in token spending (the yellow line) has flattened out in recent months. Not all tokens are created equal. It appears that growing consumption of cheaper cached tokens – which recall past inputs and are thus less computationally intensive – is inflating token volumes. Figure 13. Token spending growth has flattened even as volumes continue to rise As at 10/20/2026. Sources: Ramp AI Index, RBC GAM As the chart below shows, average prices paid by Ramp users for input and output tokens have increased year-to-date (notwithstanding a recent pullback). This suggests frontier developers have retained some pricing power in the face of rising competition from open models. But when cheaper cached tokens are included, the blended price (the light blue line) has fallen sharply in recent months, creating the gap between volumes and spending shown above. The volume data is thus less encouraging than it appears on the surface. Figure 14. Frontier model token prices are generally holding up As of 10/16/2026. Weighted average of OpenAI and Anthropic token prices. Sources: Ramp AI Index, RBC GAM Looking beyond token pricing, the price of compute suggests demand remains firm. Rental prices for NVIDIA’s B200 Blackwell GPUs (released in 2025) have increased by more than 40% year-to-date, while prices for the previous generation H100 Hopper chips (released in 2022) are up by more than 30% (see chart below). This suggests demand for compute is outpacing supply even as average token prices fall. Rising prices for previous generation chips also suggest longer depreciation timelines (as discussed here ) are justified. Figure 15. Rising GPU rental prices point to solid demand As of 10/23/2026. Sources: Silicon Data, Bloomberg, RBC GAM Hyperscalers’ growing order backlogs (remaining performance obligations or RPO) for cloud computing services also suggest demand currently exceeds supply (and increasingly so). Backlogs among four of the big five hyperscalers (Meta doesn’t disclose RPO) now exceed $2.3 trillion. That’s nearly three times year-ago levels (see chart below). However, OpenAI and Anthropic reportedly account for more than half of those orders, which are thus contingent on future demand for those labs’ frontier models. Figure 16. Hyperscaler order backlogs now exceed $2 trillion As at 10/23/2026. Backlogs reflect contracts signed with customers for cloud computing that has not yet been delivered. Sources: Bloomberg, RBC GAM Adoption rates continue to trend higher (see the chart below). According to a U.S. Census Bureau survey, 24% of businesses are currently using AI in at least one business function and 28% plan to do so within the next 6 months. Ramp’s estimated adoption rate is twice as high (56%) but likely reflects some sampling bias. Its clients aren’t necessarily representative of the broader economy. Figure 17. Business surveys point to rising AI adoption As at 10/20/2026. Dotted lines based on a previous question that asked whether AI was used to produce goods or services. Current question asks if AI was used in any business functions. Sources: U.S. Census Bureau, Ramp AI Index, RBC GAM Intensity of use appears to be rising as well. According to Ramp’s data, the median spend on AI is nearly $16 per employee per month, more than triple a year earlier. The top 10% spend more than $800 per employee per month, while the top 1% spend nearly $10,000 per employee per month (see table below). It’s the latter cohort that has seen spending flatten out recently – median and top 10% spending per employee continue to climb. Figure 18. AI spending per employee per month As at 10/20/2026. Sources: Ramp AI Index, RBC GAM AI spending as a share of non-payroll business expenses has also more than doubled over the past year to around 6% (see chart below). But there remains significant dispersion across sectors. AI spending amounts to more than 10% of tech and professional services’ non-payroll spending, but less than 1% for 11 of 20 industries. Figure 19. AI represents a growing share of business spending As at 10/19/2026. Sources: Ramp AI Index, RBC GAM Surveys suggest enterprises plan to further increase AI spending as a share of revenue and IT budgets and integrate the technology more deeply into their business processes. Management teams report better visibility on the financial benefits of AI adoption and fewer are held back by unclear return potential. Overall, recent demand signals are generally positive. Falling token prices likely reflect changes in the composition of demand rather than broad softening in demand. For example, there’s growing use of open models and cached tokens. Rising compute prices and hyperscalers’ growing order backlogs suggest demand exceeds supply. Adoption is expected to continue rising with businesses beginning to see financial benefits from their AI spending. However, exponential growth in token consumption is less encouraging than it first appears, and while median spending per employee continues to rise, overall token spending has flattened out (at least among Ramp users). Future revenue requirements Current AI revenue reportedly exceeds $100 billion (at an annual rate), having more than doubled over the past year. Recent research suggests revenue must continue growing rapidly (50-75% annually) over the next several years to generate a sufficient return on expected CapEx. But there is a wide range of estimates of just how much revenue will ultimately be required. Empirical Research estimates more than $700 billion in annual revenue will be needed by 2030 to match the consensus ROIC forecast for the hyperscalers. That would require more paid subscribers than Microsoft 365’s current half a billion users, with businesses paying between $175-500 per user per month depending on the personal/enterprise split. The lower end of that range would already be 10x the current median spends per employee. But if the recent 3x per year growth rate can be sustained, even the higher $500 per month price tag is feasible. As a reminder, the top 10% of users are already paying more than $800 per employee per month. BloombergNEF sets a somewhat higher bar, based on current data centre CapEx plans. It estimates annual revenue of $2.3 trillion is required by 2032 ($1.8 trillion by 2030) to achieve an economic breakeven (where ROIC equals the company’s weighted average cost of capital). That’s nearly equivalent to current revenue of the entire S&P 500 tech sector (see chart below). Figure 20. BloombergNEF says >$2T AI revenue needed by 2032 to break even on CapEx As at 10/23/2026. Economic breakeven assumes return on invested capital equals weighted average cost of capital. Sources: BloombergNEF, RBC GAM A researcher from Columbia Business School finds that annual revenue will need to rise to $3.7 trillion by 2032 to generate a 10% ROIC. While that would represent a whopping 9% of projected U.S. GDP, the author suggests it’s not implausible based on current frontier compute pricing. The revenue figure would correspond to GPU rental rates of $5.50 per hour at full use, or roughly $7 per hour at 80% usage rates. That is broadly consistent with current market prices for high-end NVIDIA GPUs. However, that requires GPU-hour pricing to hold up even as capacity more than doubles. How do we begin to square all of these estimates? The projections are highly sensitive to a number of assumptions that aren’t always spelled out clearly. Complicating matters, the forecasts don’t all use the same CapEx projections. The Columbia Business School paper converts gigawatts (GW) of planned capacity into a spending profile, and its assumptions look high relative to current consensus estimates. One potentially large source of divergence comes from assumptions around non-AI data centre revenue. Bloomberg’s breakdown of hyperscaler and neocloud CapEx suggests 30% of future spending will go toward traditional cloud data centres or other non-AI investment. And non-AI workloads are still expected to account for 30% or more of data centre capacity by 2030. Empirical’s analysis explicitly assumes hyperscalers’ cloud revenue continues to grow by more than 20% per year. The AI revenues required to generate the assumed ROIC are incremental to that. The Columbia Business School analysis doesn’t seem to account for non-AI cloud revenues that could exceed $1 trillion by 2032 (or its assumptions for AI-specific data centre capacity simply look too high). If we take the BloombergNEF forecast as a reasonable middle ground – $2.3 trillion in annual AI revenue required by 2032 – that would be equivalent to the following: ~4% of current revenue at the world’s 3,500 largest companies 15-20% of expected global IT spending and ~2/3 of enterprise software spending by 2032 ~$2,000 per knowledge worker globally – about 4-6% of the average knowledge worker’s salary Those benchmarks suggest a high but not insurmountable bar for AI revenue to deliver a sufficient return on investment, as long as adoption continues to rise and broadens beyond the few industries that are using AI most intensively. It also requires AI capabilities to continue improving rapidly, and the technology to prove sufficiently transformative to replace some existing software, IT and enterprise services spending, and likely some labour costs as well. The trends we’re tracking suggest that’s feasible but far from a foregone conclusion, and the market’s greater scrutiny of AI CapEx plans seems appropriate. We will be closely watching the demand indicators discussed above in the coming months. -JN Upside risks in focus It is always fashionable to be a nervous bear. After all, many things could go badly in the world. It is not unreasonable to worry about the level of bond yields, the high price of oil, AI uncertainties, France’s fiscal problems, the midterm elections, geopolitics and so on. And yet, most of the time, it is the investment bulls who win. Humans are naturally hard-working, inventive and resistant to entropy, and policymakers do their best to keep the economy on track. And so, while it is unfashionable to acknowledge, there are usually nearly as many upside risks as downside risks. Let us run through six such upside risks. We certainly aren’t promising that they WILL happen. The point is that they COULD happen, and in some cases, they are just as likely or even more likely to transpire than their evil twin (the equivalent negative risk). Faster economic and earnings growth? The U.S. economy – and to an extent, a fair swath of the developed world – have repeatedly surprised to the upside with economic resilience. The country has shrugged off higher yields (the economy is potentially less rate-sensitive these days). It’s survived high oil prices (the economy is definitely less oil-intensive than it once was). And, quite amazingly, it’s growing without any great contribution from the usually pro-cyclical housing sector. To be sure, the AI boom is central to this – and constitutes its own upside risk that will be discussed in a moment. But it is notable that stock market earnings growth is impressive well beyond the tech sector, and also well beyond the supercharged U.S. market (see next two tables). Something special is happening, and it is entirely possible that growth and earnings will continue to surprise to the upside in the year ahead. Figure 21. S&P 500 earnings growth and revisions As of 10/23/2026. Source: Bloomberg, RBC GAM Figure 22. Major stock-market indices – Consensus earnings outlook As of 10/23/2026. Sorted by 2027 EPS growth. Sources: Bloomberg, RBC GAM AI booms even more? The world is currently dwelling on AI risks that span the gamut from inadequate capacity to inadequate demand, with concerns about circular financing thrown in for good measure. But it is important to remember that, at least to this point, AI CapEx forecasts have been revised higher at every opportunity, as have AI earnings forecasts. It’s possible that now comes to an end, but equally possible this upgrading continues based on recent hyperscaler comments about nearly insatiable demand. AI productivity shows up quickly? Looking beyond the AI investments and profits to the real prize – productivity gains – it is possible that AI-driven productivity gains show up more quickly than the market is assuming. Top-down analysis only hints of a slight productivity upgrade so far. But firm-based analysis finds many companies insisting they are already deriving significant benefits. New technologies have a habit of spreading more quickly than past revolutions, and software-based technologies (as AI manifests for users) tend to spread the fastest of all. Further, if the AI productivity gains come early, the deflationary benefits would also arrive more quickly, lowering inflation and interest rates. Energy shock resolves? We have now reached a point of such fatigue around this energy shock that it is natural to assume that it will simply continue forever. Maybe it will. But, as discussed earlier, the supply of oil has actually improved quite nicely despite the constraints imposed by the conflict. And it is still possible that the U.S. and Iran – both of which are feeling pressure to resolve the conflict – will find a way to dial down the conflict and in so doing lower energy prices. This would substantially reduce inflation, take pressure off central banks, and reduce bond yields. All of that could unlock further economic growth. Chinese private savings rate falls? It probably isn’t wise to assume China’s private savings rate starts falling tomorrow, but it represents a real medium-term upside risk. China’s personal savings rate has long been lofty. But it has continued to soar to an unfathomable 32.9% of disposable income as of the latest reading (see next chart). For context, the U.S. inclination to save is just one-tenth of that. Figure 23. China’s personal saving rate continues to soar As of Q3 2026. Sources: China National Bureau of Statistics (NBS), Macrobond, RBC GAM China’s weak housing market has been a central catalyst for this saving, as it has damaged household wealth, hurt consumer confidence (see next chart) and discouraged consumption (see subsequent chart). Figure 24. Chinese consumer confidence fell sharply at the onset of the housing downturn As of September 2026. Sources: China Economic Monitoring & Analysis Center (CEMAC), Macrobond, RBC GAM Figure 25. Chinese retail sales growth has ground to a halt As of September 2026. Sources: China National Bureau of Statistics (NBS), Macrobond, RBC GAM In fairness, there is no particular evidence that Chinese households are preparing to spend their accumulated savings just yet. But they could if they wanted to. Helpfully, there is tentative evidence that the housing market is stabilizing and policymakers have room to deliver consumer-supporting stimulus. Whether this upside risk is realized any time soon, it represents an untapped source of economic potential. Public investment / infrastructure boom? Lastly, there is an upside risk for the (mostly non-U.S.) developed world if the long-discussed public investment / infrastructure boom actually happens. Quite a few countries are committing to more military spending, strengthening infrastructure, achieving greater resource security and re-prioritizing their economies more generally. Some of this is already factored into base-case forecasts, but not all. Conclusion Will all of this happen? Certainly not. But some might. In short, while AI might bust or oil could rocket to US$150 per barrel, AI could instead boom further, or oil could fall to US$70 per barrel. The downside and upside risks are arguably more symmetrical than popularly imagined. -EL Neutral rate check-in The U.S. economy grew at an impressive 2.7% average annualized pace through the first three quarters of 2026. At the same time, the jobless rate has trended gradually lower to levels consistent with full employment. While monetary policy is theoretically restrictive – at least relative to the Fed’s long-run estimate of the neutral rate, which neither stimulates nor slows the economy – it doesn’t appear to be dampening growth. That suggests the neutral rate might be higher than the Fed’s current 3.25% estimate, even as the median long-run fed funds rate projection from the dot plot has increased by 75 basis points (bps) in the past few years (see the chart below). This has implications for the Fed’s current tightening cycle and where the policy rate might end up in the long run. This in turn influences long-term bond yields. Figure 26. The Fed has been revising its long-run neutral rate estimate higher As of 09/16/2026. Sources: Federal Reserve, Macrobond, RBC GAM Taking a step back, the neutral rate is a theoretical, unobservable interest rate that balances the supply of savings and demand for investment. As the chart above shows, estimates of the neutral rate were revised lower during the 2010s, reflecting several factors that boosted supply and weighed on demand. But some of those factors have shifted in favour of a higher neutral rate (see the table below). The extraordinary AI CapEx boom that is now underway is also exerting upward pressure. Figure 27. Are we heading toward a higher neutral rate? As at 11/02/2026. Source: RBC GAM Demand factors AI CapEx: Business fixed investment as a share of U.S. GDP is now higher than the dot-com peak. It should soon eclipse the post-war record seen in the early 1980s (see the first chart below). Over the next several years, Brookings estimates AI CapEx will easily surpass past investment booms as a share of GDP (see the second chart below). While hyperscaler CapEx growth is expected to flatten out after next year, significant sustaining capital is required to regularly replace the servers and IT equipment that account for roughly two-thirds of the upfront cost of a data centre. Figure 28. The U.S. economy is becoming more investment intensive As of Q3 2026. Sources: U.S. Bureau of Economic Analysis (BEA), Macrobond, RBC GAM Figure 29. Expected AI CapEx significantly outpaces past U.S. investment booms As at 10/22/2026. Size is average annual capital expenditure as a percentage of GDP over the indicated period. Sources: Brookings, RBC GAM Productivity: Slow productivity growth after the Global Financial Crisis has discouraged business investment. But as we discussed here , U.S. productivity growth has recently accelerated relative to its previous trend. We think productivity growth can pick up further and remain persistently stronger in the coming years as AI is further integrated into business processes. That should be supportive of investment demand and the neutral rate. Tangible vs. intangible investment: A shift from tangible (e.g., machinery and equipment, structures) to intangible investment (e.g., software, R&D) previously pushed the neutral rate lower, as businesses tend to fund the latter through operating income rather than borrowing. But that is starting to reverse given the more tangible nature of the AI buildout (see the chart below). Figure 30. AI buildout means a shift back toward tangible investment As of Q3 2026. Sources: U.S. Bureau of Economic Analysis (BEA), Macrobond, RBC GAM Public investment: Large deficits and rising public investment are also boosting global demand for capital. Governments are progressively prioritizing infrastructure and defence spending to support economic resilience and national security. CapEx deflator: Falling relative prices of CapEx relative to the broader GDP deflator meant lower costs and less demand for capital per unit of investment. Now, shortages of memory chips, electrical equipment and construction workers are driving investment costs higher, although the CapEx deflator still isn’t outpacing broader price growth. Supply factors Demographics: With more Baby Boomers reaching retirement age, the share of the U.S. population age 65 and older has increased to 19% from 15% a decade ago. As the chart below shows, that means a growing share of the population is past its peak savings years, which is weighing on the domestic supply of savings. However, slower population growth means less demand for capital (to maintain any given capital-to-labour ratio). Longer life expectancy increases the need for retirement savings. On balance, we think demographics are no longer exerting downward pressure on the neutral rate. Figure 31. Baby boomers are past their peak savings years As at 10/30/2026. Savings rate proxy based on average spending and pre-tax income. Sources: U.S. Bureau of Labor Statistics, RBC GAM Global savings glut: The flow of savings from emerging markets to developed economies put downward pressure on interest rates, particularly in the early 2000s. This was a consequence of EM trade surpluses and a desire to build foreign reserves. EM current account surpluses began shrinking after the Global Financial Crisis, but have now been rising once again, although not to previous levels. On net we think this is no longer weighing on the neutral rate. Inequality: Rising inequality puts more income in the hands of households that tend to save relatively more, boosting the supply of savings and putting downward pressure on the neutral rate. Income inequality continues to rise in the U.S., maintaining downward pressure on the neutral rate. Taking all of those factors into account, our modeling suggests the long-run neutral rate is now closer to 3.5-4%. That’s well above the ~2.5% rate during the 2010s and slightly above the Fed’s latest dot plot median of 3.25%. It’s worth noting that there is a wide range of views among policymakers. Seven of 18 Federal Open Market Committee (FOMC) members who submit a long-run dot put it in our higher range. Indeed, minutes of the September FOMC meeting showed several participants viewed the current policy rate (3.75-4%) as not restrictive or only mildly restrictive. While Chair Warsh has declined to give an assessment of the neutral or long-run fed funds rate, he has said, “I would be hard pressed to describe broad financial conditions as restrictive.” Various regional Federal Reserve banks have their own models for the neutral rate. These likely inform the regional presidents’ long-run dot views – although the dots tend to lag the model estimates. Again, there is a wide range of estimates (see the chart below). Two of the six most common models sit currently above 3.5%. Most have been trending higher. Figure 32. Fed long-run neutral rate models have been trending higher As at 10/30/2026. Fed models estimate the real neutral rate: 2% long-run inflation added to the nominal values shown. HLW is Holston-Laubach-Williams. Sources: Federal Reserve Bank of New York, Federal Reserve Bank of Cleveland, Federal Reserve Bank of St. Louis, Federal Reserve Board, RBC GAM The factors discussed above are expected to influence the neutral rate over the medium to long term. But there are also factors that can temporarily push the neutral rate higher or lower than its long-run level – and we think those factors point to an even higher neutral rate at present: Inflation: While we tend to talk about nominal interest rates, it’s real rates that matter. Higher inflation means any given nominal interest rate is less restrictive. Risk premiums: Since funding costs and investment returns reflect more than the fed funds rate and government bond yields, the broader cost of capital also influences the neutral rate. Today’s low equity risk premium and tight corporate spreads suggest that, all else equal, the neutral risk-free rate needs to be higher to balance supply of savings and demand for capital. We think the current short-term neutral rate – that would be consistent with inflation getting back to 2% over time – is currently above 4%. As inflation comes down in 2027, so will the nominal short-term neutral rate. But to the extent that risk premiums remain low – and broader financial conditions remain accommodative – the short-term neutral rate could temporarily remain at or above the higher end of our estimated 3.5-4% long run range. Based on this assessment, we think modest further tightening is required to remove some of the current accommodation in financial conditions. There should be scope to cut rates once inflation gets closer to target. But rates will probably not move all the way to the Fed’s current 3.25% neutral estimate, which could continue to move higher. Even with a long-run neutral rate of 3.5-4%, there is scope for term yields to come down somewhat further with the market pricing in a long-term fed funds rate of 4.25-4.5%. -JN Could the global order snap back to normal? The global order has changed in a major way. The formerly hegemonic world is now a multipolar world with China playing a central role alongside the U.S. Such environments tend to be associated with periods of de-globalization, and, true to form, the U.S. raised its tariffs. The world is also transitioning from a rule-based order to a power-based one in which powerful countries are more willing to throw their weight around. As a result, the world has become a more dangerous place. International attitudes toward the U.S. have also worsened significantly (see next chart), to the point that China and Russia are now more popular than the U.S. (see subsequent chart). Figure 33. Attitudes toward the U.S. are almost universally worsening As at 09/30/2026. Sources: Pew Research Center, RBC GAM Figure 34. The U.S. is now viewed less favourably than China and Russia As of 2026. Sources: Democracy Perceptions Index 2025, RBC GAM But midterm elections are now upon us, with the prospect of Democrats supplanting Republicans in Congress. Further, President Trump will bump up against term limits in just two years. Given all of this, is it possible that the global order snaps back to the prior norm sooner rather than later? For that matter, has the global order really changed that much? Global trade is rising nicely despite U.S. counterefforts (see next chart). The U.S. dollar is still strong despite claims of erosion. And Greenland was resolved relatively benignly despite the revived Monroe Doctrine. Figure 35. Global trade continues to grow As of August 2026. Shaded area represents U.S. recession. Sources: CPB Netherlands Bureau for Economic Policy Analysis, Macrobond, RBC GAM Our short answer is that the global order really is changing despite these counterpoints – see the Strait of Hormuz, the war in Ukraine, Venezuela, China’s economic ascent, China’s diminishing Treasury holdings, rising military spending around the world, lingering tariffs and so on. Furthermore, the changing global order is unlikely to significantly reverse in the next few years, or beyond. In fact, as investors and economic actors gradually awaken to the reality that the global order is not going back to the way it was, that could catalyze some of the anticipated changes that have thus far been slow to arrive, such as a weaker U.S. dollar. The arguments for a persistently altered global order are as follows. There is no reversing the newly multipolar world. China’s economy and military have grown to the point that they now rival the U.S. It is difficult to envision a scenario in which that reverses. U.S. isolationist tendencies present an opportunity for China to further augment its clout. Global trade has proven impressively resilient despite U.S. tariffs – mostly via other countries striking new bilateral deals with one another. But there have been subtle signs of de-globalization dating back to the 2008 Global Financial Crisis. A world with multiple great powers encourages the formation of spheres of influence that don’t trade quite as freely with one another. It is far from automatic that future U.S. presidents will significantly unwind recent tariffs. A Republican president might be ideologically aligned with the current administration or feel compelled to honour President Trump’s legacy. The Democratic Party, meanwhile, has not been especially pro-trade in recent decades. The power-based order has also been coming into view for some time. While U.S. foreign policy is newly emboldened, other great powers have been flexing their muscles for several years. Russia invaded Ukraine in both 2014 and 2022. China is increasingly assertive in the Pacific. As the saying goes, “trust arrives on foot and leaves on horseback.” The rest of the world is now less trusting of the U.S., and this is unlikely to be quickly restored. The rules-based order took generations to construct and would take a long time to reconstruct. Even an objectively friendly future administration would be unlikely to fully restore normal relations as other countries have seen how easy it is for great powers to undermine it. The rest of the world is now arming itself, and such efforts do not halt quickly or lightly. The de-dollarization theme will likely continue. Countries have increased their gold holdings. Some like China are shifting away from Treasury bonds to private assets. Others have even opted to shift where they hold their physical gold to reduce their dependence on the U.S. The world’s reliance on the American SWIFT financial system is also under scrutiny. China has constructed a parallel system, and other countries may yet do the same after the U.S weaponized its network against Russia. Within the U.S., American presidents are claiming ever-more executive power, sidelining Congress. Future presidents will be reluctant to give this up, with the result that the checks and balances built into the U.S. political system have weakened. The scope for forceful action by future presidents is therefore high. The large right-leaning majority in the Supreme Court constitutes the largest political imbalance in the court since the 1960s and could persist for decades to come. More generally, do not underestimate the power of political inertia, or the extent to which the beneficiaries of new policies will fight to retain them – even if they are net negative for the economy or society. Policies can also be sticky – it can be hard for new administrations to disentangle from pre-existing military engagements, for example (see Vietnam, Afghanistan). Even short-lived policies can have lasting consequences. Cutting a program results in the loss of institutional knowledge and expertise, and this can only be restored gradually after the program is restarted. In the context of the global order, the elimination of the many programs associated with USAID greatly reduced U.S. soft power around the world. It will be a lengthy process to revive the program later. Finally, the rise of populism is not limited to the U.S. It is also visible in Europe and a variety of other markets. This means that a return to multilateralism may be viewed skeptically by more than just the U.S. In conclusion, the world might become somewhat more predictable under a future U.S. administration. But the genie is out of the bottle when it comes to the transition toward a multipolar world, elements of the de-globalization narrative, the shift toward a power-based order, and the loss of trust in the U.S. We do not budget for a major reversal in 2029 or beyond. In turn, the new global order will likely remain in place. At the highest level, we believe this is consistent with slightly reduced economic growth, slightly higher inflation and slightly larger investment risk premiums, all else equal. -EL How long can the bull run? The cyclical picture Since 1929, there have been 27 cyclical bull markets in the S&P 500 and an equal number of cyclical bears (see chart below). The average cyclical bull market lasted two and a half years and produced a cumulative return of 91%. Bear markets, by contrast, tended to be shorter and shallower, lasting just under a year and declining 31% on average. Figure 36. U.S. has seen an equal number of bull and bear markets since 1929 As of 10/26/2026. 54 cycles (27 bull and 27 bear markets) since 1929. Sources: Bloomberg, RBC GAM The current cyclical bull market, which began after the inflation-driven bear market of 2022, is now four years old. It has generated a return of approximately 119%. As such, it has moderately exceeded the average cyclical bull in both duration and magnitude. But it is hardly unprecedented. The run-up to the dot-com bubble, for example, lasted nearly eight years and saw the market quadruple in value. Importantly, the duration of a bull market is determined not just by its chronological age, but also by: Whether the fundamental drivers that initially drove the bull market remain in place Any evidence that the economy and/or markets are building up excesses The emergence of large external shocks that interfere with the bull market. The long-term, secular context matters A further nuance that must be considered when evaluating bull and bear markets is the long-term or secular backdrop. The long-term market trend has an impact on shorter-term fluctuations. Cyclical bull markets that occur within secular bull markets tend to be more powerful and last longer than those in secular bears. As shown in the graphic below, the average cyclical bull within a secular bull market produces a 93% return over almost 3 years, compared to a lesser 67% over just under 2 years for those within a secular bear. Cyclical bears are also much milder in secular bulls, averaging a 21% decline over 9 months versus a 38% decline over 2 years in secular bears. We classify the current market phase as a secular bull that began in 2009 in the aftermath of the Global Financial Crisis. This secular cycle has been characterized by extraordinary earnings growth, accommodative monetary policy, earnings multiple expansion, technology leadership, robust capital return programs and a pattern of short-lived drawdowns followed by quick recoveries. This secular bull market has just passed the average secular lifespan of 17 years, but several secular bulls have lasted considerably longer. The 1942-to-1968 secular bull phase lasted 27 years, The 1978-to-2000 cycle ran for 22 years. For context, the average annualized return for the S&P 500 during secular bull markets has been approximately 14%. Figure 37. The S&P 500 has spent far more time in bull markets – with cumulative gains that have significantly exceeded losses during bear markets Uses Robert Shiller’s historical U.S. stock market data since January 1870. Data based on monthly closing prices before 1929. Sources: Robert J. Shiller, RBC Capital Markets, RBC GAM Can this cyclical bull persist? We think the current cyclical bull market may have further room to run. Despite recent choppiness, AI-driven investment and productivity gains have the potential to sustain elevated earnings growth and margins. Corporate balance sheets remain in healthy shape. These are supportive conditions for continued equity market appreciation. That said, there are risks. Valuations remain elevated by historical standards in the U.S., even after the recent pullback. Monetary and fiscal policy are no longer expected to be the tailwinds they once were. Market concentration has increased, making the index more vulnerable to a selloff in a small number of stocks. Geopolitical shocks may be more frequent going forward as the world continues to fragment along economic and political lines. And there are renewed concerns about the AI boom, though we believe present fears may be overblown. Were we to enter a long-term, secular bear market, history suggests an average duration of roughly 12 years, though more recent ones have been shorter. The last secular bear, which spanned the height of the tech bubble in 2000 to the lows of the Global Financial Crisis in 2009, was the shortest on record at nine years. During secular bear markets, the S&P 500 has produced an average annualized loss of approximately 6%. What this means for investors Equities tend to rise over time. Since 1929, the S&P 500 has spent far more time in bull markets than bear markets, and the cumulative gains during bulls have significantly exceeded the losses during bears. While bear markets are an inevitable part of the cycle, they have historically been relatively short and shallow compared to the expansions that follow. The current setup – a mature but not exhausted cyclical bull within a secular bull that could prove durable – is consistent with maintaining equity exposure. However, investors should remain attentive to the risks outlined above. -AM