With contributions from Vivien Lee, Aaron Ma and Eric Savoie Monthly economic webcast Our monthly economic webcast for August is now available, entitled “ Running the gauntlet: AI concerns / oil shock / tariffs. ” The latest developments Big stock market bounce After a multi-month period of stagnation, the S&P 500 surged again over the past week and a half (see next chart). S&P 500 rallies again As of 08/07/2026. Shaded area represents recession. Sources: S&P Global, Macrobond, RBC GAM The upward catalysts included: Diminishing rate hike fears as the latest U.S. job number arrived weaker than expected (discussed shortly). Reduced Iran oil shock concerns (addressed in the next section). A strong earnings season, with soaring hyperscaler cloud service revenues and Google and Microsoft as particularly prominent beneficiaries. Oil shock update The all-important Strait of Hormuz remains effectively closed (see next chart). But there has been renewed hope of a deal in recent weeks, allowing oil prices to partially unwind their early-July rise (see subsequent chart). Strait of Hormuz is effectively closed again As of 08/10/2026. Sources: Bloomberg, RBC GAM Oil price is off its earlier peak As of 08/10/2026. Sources: Macrobond, RBC GAM Supporting recent optimism, Iran and Oman are reportedly negotiating a deal framework—notably without direct U.S. involvement at this stage. The U.S. had appeared willing to take a patient approach as this unfolded (though recent comments begin to call that into question). The emerging framework – if it survives – is expected to include a toll mechanism. Iran's Foreign Ministry confirmed that the deal includes charges framed as fees for "maritime services." While not economically problematic, this would set a concerning precedent for freedom of navigation and would give Iran additional economic and geopolitical clout. Confidence in any deal's durability remains low. Beyond the toll, outstanding issues include compensation to Iran for damages, sanctions relief and a broader nuclear agreement—all of which face significant hurdles. Or perhaps these matters simply won’t be resolved and the U.S. and Iran will go their separate ways once the oil is flowing. Our working assumption is for a WTI (West Texas Intermediate) oil price of approximately US$80 – near the current level. A credible, lasting deal could push prices below $70, which would be a welcome tailwind. Conversely, if negotiations collapse, prices could spike past $90 — a meaningful headwind and one that arguably understates the consequences since refined products have suffered more than oil prices themselves since this war began. Shutdown averted The U.S. Senate passed a bill that will fund federal agencies until December 11. That means a government shutdown has been avoided in the lead up to fiscal year-end on September 30 and before the midterm elections in early November. This is good news, though we warn that Congress is likely to emerge from the election with power newly split between Republicans and Democrats , At that point the risk of government shutdowns and debt-ceiling close calls increase. The current debt ceiling limit is on track to be breached around next spring. Weak U.S. economic data We remain generally constructive on the U.S. economy. But two big data releases were recently weak: July non-farm payrolls and second-quarter GDP (gross domestic product). Neither were as bad as they first looked. U.S. payrolls for July suffered the loss of 23,000 jobs – a weak, below-consensus outcome. There were also 103,000 fewer jobs created than previously imagined over the prior two months – the 11th time in the last 12 months that has happened (with the implication that the -23,000 number will probably be worse with further revisions). But it wasn’t quite as bad as it looked. There were oddities in the report that may have artificially dragged it downward. For example: 50,000 jobs were purportedly lost in the local government education sector – which doesn’t appear to square with the experience on the ground. Private-sector hiring was still positive: +30,000. Most importantly, the unemployment rate actually fell from 4.2% to 4.1% – hardly evidence of a cracking labour market. While not great news for workers, inflation-watchers cheered the deceleration of hourly earnings from +3.4% year-over-year (YoY) to +3.2% YoY. It was a similar story with U.S. Q2 GDP (see next chart). The aggregate 1.5% annualized gain underwhelmed. But beneath the surface, the composition rendered this much less concerning. Consumer spending and business investment both rose nicely. It was inventories and trade that subtracted. But the inventory decline was likely only temporary and trade subtracted – not because exports fell but because imports rose. This is ultimately a signal of strong demand even if it technically subtracted from GDP. Composition of U.S. Q2 GDP growth signals that underlying demand remains strong As of Q2 2026. Sources: U.S. Bureau of Economic Analysis (BEA), Macrobond, RBC GAM As we touched on in a July note , financial markets no longer actually want the U.S. economy to be particularly strong, for fear that the economy will overheat and prod inflation higher. Our study of the stock market’s reaction to the last seven U.S. payroll reports (see next table) reveals a distinct shift in the market’s attitude. Whereas energy shock concerns had previously persuaded the stock market to celebrate strong job numbers through the spring, the resilience of the economy was such that the May report saw the S&P 500 respond to a strong +129,000 job gain with a sharp retreat that day. Good was suddenly bad. The June data release was ambiguous as it was neither especially strong nor weak – and indeed the stock market was unchanged that day. Now, the July job numbers have arrived weak and the stock market has celebrated. Bad is good. We mustn’t exaggerate this conclusion – the stock market wouldn’t celebrate a disaster and isn’t hoping for a recession. But it is clearly very sensitive to overheating risks right now. Good U.S. macro data is starting to be viewed negatively As of 08-10-2026. Sources: U.S. Bureau of Labor Statistics (BLS), Bloomberg, RBC GAM Fed implications Helping to explain the stock market’s enthusiasm for macro sluggishness, the soft July job number persuaded the bond market to take out about a fifth of a (25 bps) rate hike by the end of 2026. To the stock market’s way of thinking, that increases the chances of keeping the economic expansion going for longer. The bond market has been less inclined than the stock market to celebrate in recent weeks. The cooler economic numbers have been fine. But the Federal Reserve (Fed) decision itself on July 29 was interpreted as reflecting a Fed that was slightly less committed to inflation-fighting than previously imagined. In turn, even though rate hike expectations retreated slightly, long-term yields have risen on higher inflation expectations and a larger term premium. The U.S. 30-year yield is now north of 5.00% and in the vicinity of its highest level in 18 years (see next chart). U.S. long-term bond yields rise on persistent inflation and mounting budget deficit As of 08/07/2026. Shaded area represents recession. Sources: U.S. Department of Treasury, Macrobond, RBC GAM That, in turn, has slightly tightened financial conditions, representing a modest economic drag (see next chart). Indices tracking global financial conditions have risen lately As of 08/06/2026. Sources: Goldman Sachs, Bloomberg, Macrobond, RBC GAM Canada looking better Finally, even as the U.S. economy has slowed a step, Canada delivered newly strong results. The country’s second-quarter GDP is now tracking a robust 3% annualized gain – a welcome development after two quarters of economic decline. And Canada’s July job print was a stunning +75,100 positions, with the unemployment rate falling from 6.5% to 6.4%. A significant swath of the country has participated in what is now a multi-month improvement in labour market conditions (see next chart). The Bank of Canada is therefore back in focus as a rate hike candidate in the quarters ahead. Unemployment fell across much of Canada in July As of July 2026. Shaded area represents recession. Sources: Statistics Canada, Haver Analytics, Macrobond, RBC GAM -EL Business cycle scorecard says ‘mid cycle’ Our U.S. business cycle scorecard has received its quarterly refresh. This revealed a welcome improvement: the business cycle now appears to be merely at a ‘mid cycle’ stage rather than the prior assessment of ‘late cycle’ (see next chart). To be fair, it is still safest to describe the state of the economy as being somewhere within the mid-to-late cycle range, but the odds have tilted slightly toward the more charitable interpretation of the two. U.S. business cycle scorecard now points to ‘mid cycle’ As of 08/07/2026. Calculated via scorecard technique by RBC GAM. Sources: RBC GAM What prompted the slight improvement in the cycle assessment? This is always hard to summarize, as it represents the cumulative shift in more than 60 input variables broken into 17 buckets. Inevitably, many of the categories reach conflicting conclusions (see next graphic). What we track in our U.S. business cycle scorecard As of 08/07/2026/ Darkness of shading indicates the weight given to each input for each phase of the business cycle. Source: RBC GAM Amid the noise, a few important changes helped to lead the way. Monetary policy shifted from a ‘recession’ reading to ‘mid cycle.’ This is because central banks were previously in rate-cutting mode, and rate cuts are most frequently delivered as a country descends into recession. Conversely, now that rate hikes are expected, these usually commence after an economic recovery is underway, but well before the cycle comes to an end – and so mid cycle. Employment-related variables also delivered a new message. Previously, the unemployment rate had been trending slightly higher. Historically, it is exceptionally rare for such movement to end in anything other than a recession. Hence the sector’s prior association with the ‘end of cycle’ phase. But now the U.S. unemployment rate is low and appears to be trending back downwards. This is a combination that usually occurs in the middle of the business cycle. It must be conceded that business-cycle analysis is less useful than usual for investors right now, for two reasons. The economy is presently less connected to the stock market than normal given the dominance of a handful of technology-oriented companies in the stock market. A large fraction of the toolkit used to assess the state of the business cycle failed utterly a few years ago when recession signals were blinking red and yet a recession was ultimately avoided. That doesn’t invalidate the approach altogether, but it does point to a changing world and argues its value may be diminished relative to prior economic cycles. Alongside the business cycle analysis, we also update our suite of recession models. Unsurprisingly, the models that focus on the current state of the U.S. economy insist with strong conviction that there is no recession right now. More usefully, the forward-looking models make a similar claim about the year ahead. A particularly useful model that combines financial and macro factors delivers a recession probability of just 3.3% (see next chart). The Kalshi probability market gives a similarly low 6.0% likelihood (see subsequent chart). The probability of U.S. recession within a year is low As of July 2026. Based on RBC GAM model which includes financial and macro factors. Shaded area represents recession. Sources: Haver Analytics, RBC GAM Odds of U.S. recession in 2026 have fallen markedly As of 08/07/2026. Sources: Kalshi, Bloomberg, RBC GAM -EL, VL & Dominic Mulraine GDP forecast update Our interim GDP forecasts for the developed world have been updated (see next table). Growth outlook remains above consensus As of 07/30/2026. Sources: Bloomberg, RBC GAM To summarize the updated 2026 outlook with a focus on the numbered ovals that overlay the table: After downgrading most national forecasts a quarter ago due to the onset of the energy shock, the adjustments this quarter are much more varied: The Eurozone outlook dims slightly due in part to a still-struggling auto sector. The U.S. and Canadian outlook remain roughly unchanged. The UK, Japan and South Korea all improve on a stronger handoff and/or the effects of the AI boom. These adjusted forecasts still leave most countries running at or ahead of their potential – this is to say, these remain solid growth forecasts. The Eurozone is the exception, with a slightly below-potential forecast. The UK is merely at its potential growth rate. Our forecasts remain largely ahead of the consensus, with Canada the lone exception with an on-consensus outlook. So we remain slight macroeconomic optimists. What might change for the economy in 2027? We identify a range of shifting macro forces (see next table). As it happens, these net out to a still-supportive economic environment for the year ahead. Growth impulses continue into 2027 As of 08/03/2026. Source: RBC GAM With some variation by geography, monetary policy is expected to become less supportive in 2027. Likewise the stock market wealth effect is assumed to become less helpful – if only because it is hard to fathom stocks continuing to rise at their current frenetic pace. But the energy shock is presumed to become less problematic next year, and AI-related productivity gains may continue to rise in the year ahead. -EL, JN & VL AI CapEx cycle: earnings vs. free cash flow With U.S.-Iran tensions receding slightly, the market has turned its attention back to the AI capital expenditure cycle. Massive investments by hyperscalers continue to raise eyebrows and have investors questioning whether the eventual payoff will be worthwhile. Our May 4th edition of the MacroMemo highlighted how much of that capital spending has flowed into semiconductor companies (semis), which have enjoyed immense earnings growth and stock price appreciation. This analysis will dive deeper into where that spending is showing up and touch on the possible timing mismatch between capital spending and expense recognition that could be blurring the earnings lines for the hyperscalers in particular. In addition to very real and robust revenue growth, part of the reason S&P 500 earnings have been growing so quickly is that hyperscaler investment does not immediately count as an expense from an accounting standpoint. However, the companies receiving that investment are recognizing the associated revenue and profit immediately. Hyperscalers that opt to lease rather than acquire new data centres enjoy a loosely similar accounting advantage. The resulting earnings growth is impressive, but flattered somewhat by this timing mismatch between revenue and expense recognition. Free cash flow (FCF), defined as cash flow from operations less capital expenditures, is a cleaner signal of a company’s financial picture, since it captures actual cash outflows regardless of how the obligations are structured or classified. The chart below highlights a stark divide between how earnings and FFCs are behaving for the semis sector versus the hyperscalers. Earnings are rising for both sets of companies, but the FFCs of these two groups is trending in opposing directions. Hyperscaler profit estimates are expected to rise 17% annualized from now through 2027. The equivalent figure for the semis is 16%. Both estimates are impressive. In contrast, hyperscaler FCF is expected to drop 88% from its 2025 peak to its projected 2027 trough. Semis FCF, by contrast, is forecast to grow in line with earnings. Analysts do expect this dynamic to improve, with hyperscaler FCF projected to recover in 2028-2029 as the return on these investments begins to materialize. That recovery, however, depends on AI monetization coming through as expected. Earnings for semis and hyperscalers are expected to rise, but outlook for free cash flow diverges As of 08/06/2026. 12m forward earnings and free cash flow figures used. Hyperscalers include Alphabet, Amazon, Meta, Microsoft and Oracle. Semis include NVIDIA, Micron, Broadcom and Applied Materials. Sources: Bloomberg, RBC GAM Hyperscaler FCFs have deteriorated to the point that they are now issuing debt to fund further CapEx, marking a meaningful shift from self-funded growth to reliance on external financing. The market has recognized this divergence and rewarded the semis far more than the hyperscalers. Since June 2025, the equal-weighted index of U.S. semiconductor stocks has risen 148.5%. The equal-weighted index of hyperscalers has risen just 25.0% over the same period (see chart below). The semis did experience a sharp drawdown in July, but the broader trend of outperformance remains intact. Semiconductors have outperformed hyperscalers since June 2025 As of 08/05/2026. Indexed to 100 on May 30, 2025. The Bank of America U.S. Hyperscalers Index is equally weighted, its constituents include Amazon, Alphabet, Meta, Microsoft and Oracle. Sources: Bloomberg, RBC GAM Another way to examine the impact of massive capital spending on company fundamentals is by comparing their earnings yield to FCF yield (earnings or FCF divided by price – the inverse of the price-to-earnings (P/E) and P/FCF multiples). The hyperscaler earnings yield has actually improved slightly as earnings have risen. FCF yield for this group, on the other hand, has deteriorated dramatically since early 2025 as CapEx has surged and FCF has accordingly fallen precipitously. From a valuation standpoint, hyperscalers now look far more expensive through a FCF lens, because investors are paying up for low or even negative FCF (see chart below). Hyperscalers now look expensive from free cash flow yield perspective Hyperscaler earnings yield & FCF yield As of 08/06/2026. Forward 12-month EPS and FCF data were used. Sources: Bloomberg, RBC GAM This distinction is important, because FCF is the ultimate measure of a company’s funds available to investors. With less FCF available, the margin for error for these companies is smaller now. Any disappointment in future monetization of the enormous CapEx being directed toward AI compute capacity would pose a risk to stock prices. Moreover, today's capital spending will eventually show up on the income statement as an expense through depreciation once these data centres become operational. As that earnings headwind emerges, hyperscalers will need meaningful AI revenue growth to counteract it. In conclusion, an AI investment boom is underway, and the economy and corporations are reaping the rewards in different ways. At the centre of this technological shift are the hyperscalers and semiconductor companies, which have most directly profited. As we move forward, the crucial question remains whether all this investment will translate into a healthy return on invested capital for the hyperscalers and whether there is a durable pipeline of capital spending to support semiconductor stocks and other companies involved in the build-out of AI compute capacity. We are watching closely to ensure hyperscaler AI revenue is growing fast enough to justify the spending, and for any shift in tone around capital allocation discipline. FCF is a key metric to monitor as the investment cycle matures. -AM AI inflation dynamics – higher today, lower tomorrow We are optimistic that rising AI adoption will result in a sustained lift to productivity growth in the coming years. This could act as a modest disinflationary force. But in the near-term, the AI buildout is putting upward pressure on prices, particularly in the U.S., at a time when tariff and energy shocks have already elevated inflation. At a monetary policy-relevant horizon (i.e., within the next 1-2 years), we don’t think productivity-driven disinflation will fully offset buildout-related cost pressures. The next move in interest rates is more likely to be a hike than a cut. Short-term AI inflation There are several ways in which the AI buildout is already adding to inflation. These price pressures are likely to be sustained for the next year, and perhaps longer: Higher prices for semiconductors , particularly memory chips that remain in short supply (as we discussed here ), are pushing up the cost of consumer electronics. The chart below shows producer prices for semiconductors and computer and peripheral products. Goldman Sachs estimates higher memory chip prices will contribute to a 10% increase in computer and phone prices this year, Some analysts are forecasting even sharper price hikes. Other products like vehicles, TVs and gaming consoles could see prices rise, and the ubiquity of semiconductors could increase the cost to produce a broad range of goods. Prices for semiconductors, computers and peripherals have jumped higher As at 08/07/2026. Sources: U.S. BLS, RBC GAM Software prices are rising at the fastest pace on record as developers charge for new AI features. This quality improvement isn’t adjusted for in PCE (Personal Consumption Expenditures) and CPI (Consumer Price Index) inflation indices, unlike some consumer technology product prices that are quality-adjusted. Software has a significantly greater impact on the core PCE price index, where it has a 1.2% weight, compared with the core CPI index where the weight is just 0.035%. The price trends are the same, however, as shown in the chart below). The weight in PCE is likely overstated and the lack of quality adjustment constitutes another measurement error. Consumer software prices are rising at their fastest pace on record As of 08/07/2026. Sources: U.S. BEA, U.S. BLS, RBC GAM The data centre buildout is contributing to strong demand for electricity and driving up power prices for households (as we discussed here ). Electricity is adding 10-15 bps to headline CPI (see chart below) and a bit less to PCE inflation where it has a lower weight. Rising electricity prices are adding to U.S. inflation As of 08/07/2026. Sources: Energy Information Administration, U.S. BLS, RBC GAM As of May, Goldman Sachs estimated those three factors added 0.3% to PCE inflation and 0.2% to CPI inflation over the past year. These same factors could add 0.3% to both indices over the next 12 months. And there are other indirect effects through which the AI buildout could be contributing to inflation: Portfolio management fees , which are generally tied to investment assets under management, are being boosted by AI-driven equity market gains. This component was up 20% year-over-year in June and contributed 0.3 ppts to PCE inflation. We estimate about half of the S&P 500’s 1-year return has been driven by AI, meaning an approximate 0.15 ppt addition to inflation. (Upcoming revisions to PCE inflation will improve the quantity adjustment of this component and could cut its inflation contribution by nearly half.) Higher equity prices are also creating positive wealth effects for consumers, contributing to stronger spending growth and inflation. And research by the Federal Reserve Bank of St. Louis suggests consumers and businesses might start spending and investing more in anticipation of future AI-related productivity and wage gains. This can add to inflation before those productivity gains materialize. Beyond higher semiconductor prices, data centre CapEx is adding to equipment and materials prices as well as wages for construction workers . These are rising nearly 1 ppt faster than the private sector average (see chart below). Wages for specialty contractors like electricians are rising even faster. Wages for construction workers are rising faster than the private sector average As of 08/07/2026. Sources: U.S. BLS, RBC GAM Broadly speaking, data centre CapEx and wealth-driven spending are contributing to a generally tighter U.S. economy than would otherwise be the case. We think this might be adding 0.1-0.2% to inflation. All told, AI’s gross contribution to annualized inflation – before netting out the productivity effects discussed below – might be around half a percent. That price pressure is likely to persist in the near-term. Increasing memory production and electricity generation capacity will take years, and software prices could continue to rise as new AI features are incorporated. Wealth effects and portfolio management fees are less certain given recent wobbles in AI stocks, but price pressure related to the AI buildout probably aren’t going away anytime soon. Over the next 10 quarters, the big four hyperscalers are expected to invest 2.7x as much as they did over the prior 10 quarters. The deflationary productivity offset Some of this price pressure could be offset by productivity gains from AI adoption. Over the long run the initial inflationary pressures may ease while this deflationary force persists. Stronger productivity growth can accrue to workers via higher wages and to businesses through wider margins. Or, cost savings can be passed along to consumers in the form of lower prices. The extent to which AI boosts productivity, and how the productivity dividend is divided, will dictate how much disinflation we might see from AI. We constructed a three-factor cost-push inflation model with stylized assumptions based on historical pass-through of productivity gains to wages, margins and prices. We assume businesses initially capture about half of the cost savings from a sustained, positive productivity shock via margin expansion. But that competitive pressure causes some of the savings to be passed along to consumers through lower prices (disinflation). Workers capture less of the productivity gain initially as their wages are sticky: some employment contracts are fixed, workers don’t immediately realize their productivity has increased and bargaining for higher wages takes time. But history suggests wages are likely to adjust over time. Labour should ultimately capture much of the productivity dividend. Margin expansion is expected to fade as more productive workers negotiate better compensation and competitive pressures erode pricing power. That said, recent experience shows businesses are capturing a growing share of economic output and workers are getting less (see chart below). Labour’s reduced bargaining power – due to factors like globalization and less unionization – and greater market concentration in some sectors points to workers capturing less of the gains from productivity than has historically been the case. To the extent that AI models may in some cases be capable of replacing rather than simply complementing workers, less of the benefit may accrue to wages than normal. U.S. workers are capturing a shrinking share of economic output As of 08/07/2026. GVA = gross value added. Non-labour cost share not shown. Sources: U.S. BEA, RBC GAM That still leaves some room for lower consumer prices. But history suggests the initial disinflationary impulse fades over time and prices will fall by less than wages rise and margins expand. Using a weighted blend of three different types of productivity shocks, our model suggests a sustained 1 ppt increase in annual productivity growth might reduce annual inflation by about 0.1-0.2%, with the greatest disinflationary effect seen in the early stages of the productivity shock. We found similar results using two large-scale macro models. The good news is that, to the extent productivity growth contributes more to nominal wage gains than lower prices, consumers still benefit via higher real wages. We would put relatively wide ranges around those assumptions for how the productivity gain will be divided among workers, businesses and consumers. There are a number of factors that could result in more or less disinflation than the model predicts. The disinflationary impact could plausibly range from 0-0.4% per year per percentage point of additional productivity growth: Factors that could result in more or less disinflation More disinflation Less disinflation AI is labour displacing , suppressing wages and allowing more cost savings to flow to lower prices (and wider margins). Higher unemployment also reduces aggregate demand and price pressures more broadly. AI is labour augmenting , complementing workers and causing the value of labour to rise. More of the productivity gain flows to higher wages rather than lower prices. Productivity gains may be driven more by total factor productivity than capital deepening – consistent with a pure technology shock. This could result in broader disinflationary pressure, Productivity gains may be driven more by capital deepening , with sustained investment pushing up the cost of capital and leaving less room for disinflation. AI increases competition by lowering barriers to entry for new firms, reducing pricing power, More of the productivity gain flows to lower prices. AI reduces competition due to winner-takes-all dynamics, increasing firms’ pricing power. More of the productivity gain flows to margins. AI services become cheaper/commoditized , driving down costs for businesses and allowing more savings to be passed along to consumers. AI services are more scarce/expensive and less scalable , requiring ongoing investment to increase capacity and sustain productivity growth. Productivity gains are unanticipated , so businesses don’t pre-emptively invest and wages are relatively slow to adjust. Productivity gains are anticipated , with firms investing more in anticipation of future gains, causing demand to overheat. As of 08/06/2026. Source: RBC GAM Most of those factors are not either/or. AI is likely to temporarily displace some workers but complement others. Productivity is likely to be driven by both capital deepening and technology-driven efficiency gains. But we put slightly greater weight on some of the more disinflationary aspects. AI makes it easier to start a business. There is also growing evidence that open weight models are driving token prices lower . This suggests productivity-driven disinflation could be closer to the higher end of our estimates. Of course, this is all dependent on how much productivity growth AI delivers and there is a wide range of views on its prospects. Our forecasts conservatively assume a half percent productivity annual lift, on average, in the coming years. But this is at the low end of the 0.5–2.5% range we cite in our most likely scenario. We’ve flagged scenarios in which productivity growth could be significantly stronger or slightly weaker (see table below). How much productivity growth will AI deliver? As of 08/04/2026. Productivity shock is initial increase in annual productivity growth. Source: RBC GAM As we discussed recently, the U.S. has seen nearly 2 ppts of cumulative excess productivity growth in recent years (relative to forecasts at the start of the decade), up to two-thirds of which can be attributed to AI. We might already be getting some disinflationary impulse from AI, but probably nowhere near enough to offset the aforementioned price pressures related to the AI buildout. Overall, we think AI-driven productivity growth is likely to be a relatively modest disinflationary force. Peak effects will be felt later this decade when adoption has become more widespread and AI tools are better integrated into business processes. A scenario in which AI causes widespread job losses could result in lower prices due to weaker demand, while more rapid improvement in AI tools could result in greater productivity growth and more disinflationary pressure. In the meantime, our base-case view is that AI may be more inflationary than disinflationary in the next few years. This is most relevant for monetary policy today. -JN Cybersecurity risk – more disruption than destruction There are always important downside risks to consider when evaluating the global economy and markets. Obvious examples today include: a variety of U.S. political and policy risks, such as the potential for more tariffs or continued fiscal excesses broader geopolitical risks, including any intensification of the ongoing energy shock a variety of risks stemming from AI (perhaps it proves a bubble, or maybe AI badly disrupts certain sectors and/or workers) various financial market-specific risks, including the threat of higher yields or a disorderly unwind of the yen carry trade. These are all important risks, and a measure of vigilance is critical for investors. Another risk coming onto the radar screen relates to cybersecurity, which we investigate in this section (see next graphic). Cybersecurity risks through a macro lens As of 08/10/2026. Source: RBC GAM A rising threat There are four main reasons why the threat of cyberattacks is structurally mounting, and now quite serious. The attack surface has expanded significantly. Simply put, interconnected computer systems are now used for nearly every economic purpose, rendering those economic activities inherently vulnerable to outside disruption. There is a significant and mounting concentration risk. Cloud computing has grown to such an extent that large swaths of the economy rely upon a small number of providers and their software and hardware ecosystem. Hackers must only break through one network or find a vulnerability in one piece of software to have the theoretical ability to affect thousands of businesses. Hackers have rapidly improving tools at their disposal. This is primarily a commentary on the remarkable advances of AI in recent years. Large-language models are increasingly capable of not just identifying zero-day exploits in software and hardware but then taking full advantage of them and potentially doing so at a scale that would be impossible for humans to attempt. Quantum computing is still a number of years off but threatens to compromise existing encryption technologies. While the quality of encryption can be augmented, the current information being transmitted using existing encryption technologies can be archived by bad actors and then cracked at a later date once quantum computing catches up. The number of bad actors – hackers – seems to be multiplying. This makes sense. As discussed, there are declining barriers to entry due to AI advances and simultaneously more (and more attractive) targets. The number of phishing attacks have increased 4,151% since the rise of generative AI in 2022 according to U.S. News & World Report. Quantifying the damage Already, the financial damage from cybercrimes is estimated to cost the global economy between US$500 billion and US$1 trillion per year – about 0.5% to 1.0% of world GDP. Note that some public estimates put cybersecurity damage at a level that is somehow 10–20 times greater than this, but this is surely far too large. There is no way that the global economy is losing up to 10% of annual GDP from this. Damage from cyber threats can come in a variety of forms. The most common types are the temporary disruption of businesses, theft and extortion. Secondary forms include the reputational harm to affected businesses, espionage and physical damage to capital and infrastructure. The deadweight cost of preventing cyberattacks is also very real, though rarely included in official estimates and difficult to quantify since it includes not just the cost of firewalls and IT personnel but also frictions for employees and customers as they slog through multi-factor authentication, juggle passwords and take security training. Key businesses As demonstrated by the CrowdStrike outage in July 2024, the concentration risk is very real: a faulty software update knocked out approximately 8.5 million Windows devices worldwide, disrupting airlines, financial institutions, hospitals, broadcasters and retailers in one fell swoop. This means cyber outages can occur at scale, spanning multiple sectors at once. Some industries are particularly vulnerable to cyberattacks. Payment processing and banking are critical as they are increasingly conducted electronically and they are connected to every other sector. Shipping is similarly exposed, not so much because it is primarily conducted online – it isn’t – but because the logistics aspects (ordering a product, organizing its movement and releasing it at the end of the process) are conducted online. Turning to the broader supply chain, a disruption to the operations of the weakest business in a supply chain takes down the entire supply chain. Healthcare is another critical sector due to the low tolerance for disruptions. In 2024, a central clearinghouse connecting U.S. insurers, pharmacies and healthcare providers was hacked, impeding activity and even resulting in a temporary cashflow crisis for some participants. In 2024 in the UK, ransomware disabled a pathology provider serving London hospitals, interrupting blood testing and operations. Increasingly, even motor vehicles have sophisticated computer code inside them. Cyberattacks thus have the theoretical ability to disrupt a significant fraction of the motor vehicles on the road. The ascent of self-driving cars makes this even more concerning. Critical infrastructure Critical infrastructure has already been hacked, to problematic effect. The Ukrainian electrical grid was disrupted via remote access in 2015 and 2016. In 2003, an Ohio-based nuclear plant was penetrated by a computer virus and its safety-monitoring system was disabled. In 2022, the communications sector was affected: Russia hacked a satellite network, disabling tens of thousands of satellite internet connections across Europe. In 2021, a U.S. energy pipeline company had to halt its pipelines due to a cyberattack and there was a temporary gasoline shortage across the country. More generally, there have been many reports of Chinese and Iranian state-linked actors breaking into communications, energy, transportation, water and wastewater systems, with the theoretical capability to act if desired – perhaps as part of some future conflict. State-sponsored cyberattacks Cyberattacks are now a key component of warfare, though not yet a dominant component. The number of theoretical cyber-adversaries is multiplying as the world fractures in a geopolitical context. The UK’s National Cyber Security Centre reported a 50% annual rate of increase in highly significant cyber-incidents as of August 2025, primarily originating from China and Russia. In July and August of this year, suspected Iran-linked hackers have targeted water and wastewater systems across at least 12 U.S. states. Some attacks succeeded in degrading water operations according to the FBI. Russia allegedly attacked the Polish power grid in December 2025. North Korea is also often cited as the source of state-sponsored cyberattacks, with the goal seemingly less about espionage and more about theft and extortion. The cash-starved country now generates several billion dollars of revenue per year from this activity. Of course, the cyberattacks are not uni-directional. An alleged Israeli computer virus managed to damage Iranian nuclear enrichment equipment in 2009-10. The card services of three major Iranian banks were recently interrupted. Final assessment Given the increasingly fertile environment for cyberattacks, it seems likely that the damage associated with them will continue to rise at a brisk pace – perhaps in the range of 6-10% per year. If that sounds low given the extraordinary abilities of new AI models, keep in mind that those same AI models are also capable of defending businesses (indeed, that is their very purpose). A general conclusion is that cybersecurity will become a bigger problem over time as this growth rate outpaces that of the global economy. Cyberattacks are already a substantial operational risk for companies and governments. A superior or inferior strategy for dealing with them can significantly differentiate such entities from one another. Cybersecurity is an evergreen risk as opposed to a regime-specific or cyclical risk (such as the risk of tariffs being implemented or of bond yields rising, which are heavily determined by the public policy environment and the state of the economy, respectively). In other words, cybersecurity problems could pop up at just about any time. Perhaps aside from the special case of being in the middle of a war, there isn’t a time when one should be more or less concerned than normal. Fortunately, the historical record strongly supports the assertion that cyberattacks tend to create disruptions rather than inflict enduring destruction. Businesses and governments are inconvenienced, but manage to work past attacks after a period of hours to months. The goal is rarely to destroy businesses, but rather to extract money from them. Similarly, outside of war, state actors seek information and the potential to disrupt in the event of a future war, rather than to do immediate damage and risk retaliation. In turn, cybersecurity is for the moment more of a constant mild economic friction and a tail risk than a central macroeconomic or financial stability risk. Lastly, this discussion provides a convenient excuse to remind the reader about the dangers of dwelling excessively on downside risks. There are usually many economic and policy actors actively pursuing good rather than bad outcomes. This stacks the deck in a helpful way. It is always fashionable to say that the risks are currently greater than usual – and, boy, does it usually feel that way – but that simply cannot always be true. Many downside risks have a symmetrical if less appreciated upside risk (the current energy shock could end; AI could be even more amazing than the market currently imagines). There is a temptation to assign an overly high probability to some downside risks, and to expect an overly large market recoil when they do actually happen. Yet the historical record shows these events are relatively rare and for the most part quickly recovered from. -EL Tariff tweaks don’t change broad trade policy President Trump has been active on the tariff file once again. This is pushing U.S. trade policy uncertainty to a multi-month high in late July – but still nowhere near levels seen during the first year of his second term (see chart below). Here is a rundown of some of the latest developments, which net out to limited change in broad U.S. trade policy. Trade policy uncertainty has picked up, but well below year-ago levels As of 08/07/2026. Sources: Bloomberg, RBC GAM Goodbye Section 122, Hello Section 301 Section 122 tariffs expired in late July, at the end of their 150-day statutory limit. Recall that those temporary 10% tariffs had partially replaced IEEPA tariffs that were struck down by the Supreme Court in February. The Trump administration has now replaced the expiring Section 122 tariffs with new Section 301 tariffs at a rate of 10% or 12.5% on 60 countries that supply 99% of U.S. imports – though exemptions that were in place under IEEPA and Section 122 generally still apply, so actual coverage is significantly less. The tariffs are ostensibly motivated by trading partners not doing enough to limit imports of goods using forced labour, which theoretically disadvantages U.S. exporters. On net, the effective tariff rate on U.S. imports, based on announced policies, is little changed relative to the prior several months (see the slight uptick in the blue line of the chart below). The effective rate based on actual tariff collection (the yellow line below) has been around 7% since the IEEPA ruling, compared with roughly 10% during the second half of last year. The effective tariff rate on U.S. imports has been fairly steady since February Effective tariff rate as of 07/24/2026, actual tariff rate as of June 2026. Sources: Bloomberg, RBC GAM The U.S. Trade Representative (USTR) continues to conduct a separate Section 301 investigation into excess manufacturing capacity in 16 trading partners. All of these are also covered by the forced labour tariffs. This could eventually result in additional tariffs on imports from countries like China, the European Union and Japan, re-establishing some of the more differentiated country-level tariff rates that were in place under IEEPA. It could also push the U.S.’s effective tariff rate higher, though perhaps not back to IEEPA levels. Canada threats and negotiations Canada is not covered by the excess capacity investigation but now faces its own tariff threats. President Trump said 50% tariffs covering a 5% targeted subset of Canada’s exports to the U.S. could take effect August 19. This is in retaliation for discriminatory trade practices in autos, dairy and alcohol. The tariffs would be imposed under Section 338 of the Tariff Act of 1930. That is the legislation that implemented the infamous Depression-era Smoot-Hawley tariffs, but Section 338 itself has never been used to levy tariffs. If implemented, the tariffs would add 2.5% to the U.S.’s effective tariff rate on imports from Canada, prior to substitution effects (which would reduce the effective tariff rate as importers substitute away from tariffed goods). As of June, the effective tariff rate on imports from Canada (based on actual revenue collected rather than announced tariff rates) was 2.8%. Our models suggest that if the new tariffs were permanently implemented, Canadian GDP would be about 0.2% lower. Retaliation by the Canadian government, which Prime Minister Carney hasn’t ruled out, could increase the economic drag. Despite a relatively modest macro impact, such substantial tariffs would be quite damaging for businesses in targeted sectors, including electronics and electrical equipment, plastics and rubber products, and furniture. But we are not convinced that these tariffs will be put in place on a sustained basis. It seems more likely they are a negotiating tactic to extract concessions in USMCA (U.S.-Mexico-Canada Agreement) talks. USTR Jamieson Greer said he’s aiming to negotiate interim trade deals with Canada and Mexico by year end, with discussions on more challenging issues like content rules and labour regulations (and ultimately a USMCA extension) continuing next year. Based on this threat, we have slightly increased the probability of a “bad” USMCA outcome from 20% to 25% (see chart below). But we still assign higher odds to the status quo (a roughly 7% theoretical tariff rate based on announced policies, but lower in practice) or a “good” outcome that secures some relief from sectoral tariffs. Recent reports suggest Canadian negotiators are weighing potential concessions, including reversing retaliatory tariffs and alcohol import restrictions, in exchange for lower tariffs on steel and aluminum (subject to quotas) and potentially autos and forestry products. Odds still favour a positive or status quo USMCA outcome As of 08/03/2026. Source: RBC GAM Business surveys prior to Trump’s latest threats suggested many Canadian firms were beginning to move on from trade policy uncertainty, and investment intentions were improving. We think Canada’s economy can see a sustained return to growth under the status quo, with risks tilted slightly to the upside if negotiations prove fruitful. Other new tariffs and proposals Of relevance to the solar industry, the Trump administration announced a new 15% tariff on roughly US$10 billion of imports of polysilicon derivatives – including silicon wafers and solar cells – set to take effect in December. The U.S. is also setting a minimum import price for polysilicon and derivative products, although the raw polysilicon price floor is below current market prices and meant to discourage predatory import pricing. Aiming to revive the domestic solar supply chain, the Commerce Department will also establish a program to encourage onshoring, including tariff exemptions for companies setting up manufacturing facilities in the U.S. President Trump threatened a 100% tariff on generic pharmaceutical imports starting in August 2028 (rising to 200% in August 2029) for companies that don’t reshore production to the U.S. Many large pharmaceutical companies negotiated exemptions from 100% tariffs on patented pharmaceuticals, and generic drugmakers could offer to reduce prices or increase U.S. manufacturing to secure similar exemptions. Countries could also negotiate exemptions as part of bilateral trade deals with the U.S. Reshoring generic drug manufacturing to the U.S. could be challenging given thin margins and higher domestic manufacturing costs. If the tariffs are implemented, they could significantly increase drug prices for Americans, with generics accounting for more than 90% of U.S. prescriptions. Section 301 tariffs on Brazil took effect on July 22. The new 25% levy covers slightly less than one-quarter of U.S. imports from the country. Brazil was previously subject to tariffs of up to 50% under IEEPA. The tariff hike has a modest (<0.1%) impact on the U.S.’s overall effective tariff rate. Overall, despite recent developments, U.S. trade policy is currently a bit less restrictive than a year ago. To reiterate, the new Section 301 forced labour tariffs mostly represent continuity from the Section 122 tariffs that were in place between March and July, which in turn incompletely replaced the pre-existing IEEPA tariffs. Tariff refunds continue U.S. Customs and Border Protection has now paid out $100 billion in IEEPA tariff refunds (including interest). A total of $129 billion in claims have been accepted, out of $166 billion in tariffs collected under IEEPA. Some companies that received refunds have said they will reimburse customers and others are pledging to lower prices, but we think the bulk of the refunds will help repair margins for importing firms. We assume the refunds, which amount to 0.3% of U.S. GDP, will have a relatively low economic multiplier. Importing firms are unlikely to substantially increase hiring or CapEx in response to the one-time payments. But the refunds could provide a marginal boost to the economy this year. -JN