With contributions from Vivien Lee, Aaron Ma and Eric Savoie 

AI concerns

The Philadelphia Semiconductor Index (SOX) has fallen into bear market territory. It’s now down more than 20% from its late-June record high (see next chart). In fairness, the SOX is still up 60% this year and the analyst consensus remains bullish.

Philadelphia Semiconductor Index (SOX) enters bear market territory

As of 07/17/2026. Sources: Bloomberg, RBC GAM

But tech stocks and the AI trade more generally are no longer uni-directional. Dangers include:

• the threat to hyperscalers posed by cheaper open-source models (discussed later in this report)

• souring public attitudes toward data centre construction

• the aftermath of earlier potentially unsustainable gains recorded by memory chipmakers.

For all of that, however, AI continues to make large leaps forward. Whether present valuations are sustained or not, it remains likely to be a transformative technology and a significant driver of productivity gains in the years ahead.

Iran conflict resumes

As we warned in the last #MacroMemo, there was a distinct risk of hostilities resuming between Iran and the U.S. This has happened, with the price of West Texas Intermediate (WTI) oil back up from below $70 to $83 per barrel, and the Strait of Hormuz now seeing fewer crossings (see next chart). While June inflation happily declined and Federal Reserve rate hike expectations accordingly inched lower, the outlook for the coming months will be significantly determined by whether the war fully revives or if these skirmishes can be limited. It increasingly appears that a tidy resolution may be elusive even after the 60-day negotiating period (now just over half done) has expired. As such, the inflation and Fed outlook is no longer as attractive as it had temporarily looked between mid-June and early July.

Tanker crossings in Strait of Hormuz falling anew after brief rise

As of 07/14/2026. Sources: Bloomberg, RBC GAM

U.S. economy no longer surprisingly good, but still good

Our U.S. Focused Economic Surprise Index shows a roughly neutral setting, meaning that economic data has been arriving approximately in line with expectations over the past month (see next chart). But a lot of the heavy lifting has come from a single indicator, and most of the other indicators are running a tad below expectations. We mention this just because economic surprises have been so reliably positive over the prior several months that this represents a notable deviation.

Our U.S. Focused Economic Surprise Index is neutral

As of 07/16/2026. Our U.S. Focused Economic Surprise Index is the equal-weighted average of the normalized data surprises of the underlying components. Sources: Bloomberg, RBC GAM

Still, it is hard to sustain positive surprises once everyone has figured out that the economy is healthy. The data itself is still good. The latest evidence that the economy is still moving forward at a solid pace is our quantification of the most recent Beige Book report from the U.S. Federal Reserve (see next chart). That’s the strongest reading in four years.

Beige Book Sentiment Indicator shows economy is still advancing

As of July 2026. The indicator quantifies the sentiment of local contacts by assigning different weights to a spectrum of positive and negative words used to describe overall economic conditions in the Fed Beige Book. Sources: U.S. Federal Reserve, RBC GAM

More Canadian tariffs?

On July 20, as this report went to print, the White House announced a new 50% tariff on many Canadian products under Section 338 rules. That’s different than the justification used for other tariffs over the past 18 months, and indicates the U.S. is claiming that Canada is discriminating against it. The motivation is reportedly linked to Canada’s treatment of U.S. alcohol, automobiles and dairy products. Wildfire smoke was mentioned as a possible motivation for additional tariffs later. The new tariffs take effect within 30 days, and critically, unlike with prior tariffs, exemptions will not apply for products covered under the U.S.-Mexico-Canada agreement (USMCA). Other à la carte exemptions will exist, however, sparing energy, potash, fish, critical minerals and goods covered by separate sectoral duties such as autos and metals.

It is not clear whether this latest tariff threat will actually be implemented and whether the tariffs would hold up to legal scrutiny. But it is still bad news for Canada, creating elevated uncertainty and potentially slowing the economy. The U.S. is also adversely affected, though at a smaller share of gross domestic product (GDP).

Elsewhere, and arguably of diminishing relevance if a significant tariff shift does take place in the near future, Canada’s June employment report was a healthy +18,000 positions. This allowed the unemployment rate to edge lower to 6.5%. That’s still higher than an optimal 6.0% but represents significant progress from the 7% figures being reported last fall.

Meanwhile, June inflation came in quite soft as lower oil prices began to show up in the numbers. Canada’s core inflation figures are mostly around 2%, which is another win.

Consistent with this, the Bank of Canada on July 15 did not suggest any urgency in raising rates. Instead, it indicated that the current policy rate of 2.25% remains appropriate.

-EL

Space Exploration Technologies (SpaceX) has been trading in the public markets for just over a month now following its record-breaking initial public offering (IPO) in June. The US$75 billion offering shattered the previous record held by Saudi Aramco, which raised US$29.4 billion in 2019. It is one of three mega IPOs expected to take place this year and next.

Two other large companies, OpenAI and Anthropic, which developed the popular AI tools ChatGPT and Claude, respectively, are expected to come to market at some point later this year or next. This piece examines the potential market implications of these IPOs and provides historical context on IPO performance trends.

From a macro perspective, the proceeds from these mega IPOs are large. But they remain a relatively small portion of the overall market and do not by themselves create much risk of indigestion. Anthropic and OpenAI are expected to raise around US$60 billion each. Combined with the US$75 billion from SpaceX, this totals US$195 billion – roughly one third of one percent of the U.S. large-cap market.

The chart below provides historical context by plotting annual U.S. IPO proceeds as a share of market capitalization since 1980. It shows that IPO activity was considerably higher during the late-1990s technology bubble, when IPO proceeds exceeded one percent of market cap. Current IPO activity remains modest both by dollar volume and deal count, and does not exhibit the froth typically associated with past market bubbles.

U.S. IPO activity remains modest

As of 07/16/2026. Data excludes banks, S&Ls (Savings and Loans), REITs (Real Estate Investment Trusts), ADRs (American Depositary Receipts), SPACs (Special Purpose Acquisition Companies), close-ended funds and issues with offer price below $5; market cap is the aggregate market cap of the S&P 500 (ex-financial & real estate), potential proceeds from mega-IPOs assume SpaceX raises $75 billion and OpenAI and Anthropic $60 billion each. Sources: Jay Ritter, University of Florida, Empirical Research Partners Analysis, Bloomberg, RBC GAM.

However, a separate more granular concern is that these companies are currently unprofitable and rank among the worst free cash flow in the U.S. large-cap universe. The chart below plots the distribution of free cash flow margins for S&P 500 companies, excluding financials and real estate. Anthropic, OpenAI and SpaceX all fall within the bottom decile of profitability, raising questions about their near-term ability to generate positive cash flows despite substantial revenue growth.

Anthropic, OpenAI and SpaceX rank low in profitability in U.S. large-cap universe

As of July 2026. Large-cap stocks are represented by the S&P 500 Index (ex-Financials and Real Estate). Sources: Empirical Research Partners Analysis, Bloomberg, RBC GAM

Should all three of these companies be added to major indices, they would weigh modestly on the market’s overall profitability metrics. The chart below plots the free cash flow margin of U.S. large-cap stocks over time. Two markers at the end of the chart illustrate the impact:

• The hollow circle represents the expected free cash flow margin for 2026 excluding the three above-mentioned mega IPOs.

• The solid circle shows the estimated margin including them.

The difference is approximately 50 basis points.

While these companies are greatly unprofitable, their impact on the market is modest because they remain relatively small compared to the approximately US$75 trillion U.S. large-cap market.

Anthropic, OpenAI and SpaceX combined would only have modest impact on overall profitability of the large-cap market

As of June 2026. Trailing 12-month free cash flow margin data, averaged over S&P 500 constituents (ex-Financials and Real Estate) and recorded on a quarterly basis. Sources: Empirical Research Partners Analysis, Bloomberg, RBC GAM

One important factor specific to SpaceX is that only a small portion of the company was sold during the offering (see chart below). The remaining shares are held by existing shareholders and the company's founder, Elon Musk, and are subject to a tiered lock-up schedule over the next year.

This is significant because lock-up expirations add supply to the market, expanding the free float and potentially increasing volatility as additional shares become tradable. However, the phased unlock period may allow the market time to absorb this added supply without excessive price pressure.

Only a small portion of SpaceX shares was sold during IPO

As of 07/18/2026. Numbers are approximate. Sources: SpaceX S-1 Filing (2026), RBC GAM.

The anticipated supply overhang and questions around profitability may be contributing to SpaceX's extreme volatility since its debut. The stock started trading on June 12 following its initial offering at US$135 per share, and it quickly surged to an intraday high of US$225.64 per share on June 16, 2026 (67% above the IPO price). But the rapid gains were short-lived. The stock drifted down to US$123.99 by the close on Friday July 17, 2026, trading 8% below its IPO price and down 45% from its peak (see chart below).

SpaceX is now trading below its IPO price

As of 07/17/2026. Candlesticks indicate daily open, high, low and closing values. Sources: Bloomberg, RBC GAM

These mega IPOs offer investors exposure to exciting emerging technologies. However, their elevated valuations may limit upside potential. Historical data shows that IPOs have typically commanded reasonable valuation multiples, with a notable spike during the late-1990s technology bubble when the median IPO priced at 32x price-to-sales.

In comparison, current valuations are substantially higher. SpaceX debuted at 96x price-to-sales and OpenAI is expected to price around 43x. These multiples are extraordinarily demanding by historical standards and suggest limited margin for error, particularly given the companies' current lack of profitability.

SpaceX and OpenAI IPOs priced at extraordinarily high multiples

As of June 2026. Data excludes banks, S&Ls, REITs, ADRs, SPAC’s closed-ended funds and issues with offer price below $5; offer valuations are assumed to be $1.8 trillion for SpaceX, $852 billion for OpenAI and $965 billion for Anthropic. Sources: Jay Ritter, University of Florida, Empirical Research Partners Analysis, RBC GAM

The appeal for investors and a reason they are willing to pay a high price for these stocks is that these companies are growing at extraordinary rates. All three rank in the top decile of revenue growth among U.S. large-cap companies. Anthropic and OpenAI have posted growth rates so extreme they extend beyond the scale shown in the chart below.

However, sustaining such exceptional growth will be critical to generating favourable returns for shareholders. Any marked deceleration in growth could result in significant multiple compression.

Anthropic, OpenAI and SpaceX are growing at extraordinary rates

As of July 2026. Large-cap stocks are represented by the S&P 500 Index (ex-Financials and Real Estate). Sources: Empirical Research Partners Analysis, Bloomberg, RBC GAM

IPOs have historically underperformed the broader market, although technology-sector IPOs have fared somewhat better -- particularly those with substantial revenue and dual-class voting structures. The chart below plots the average three-year return of IPOs from their issue date using data since 1980. The average IPO underperformed the cap-weighted benchmark by 20.5 percentage points over a three-year holding period. That gap narrowed to 12.7 percentage points for technology-sector IPOs.

For companies with more than US$1 billion in revenue, historical returns improved further, trailing the benchmark by just 2.0 percentage points. This group would include SpaceX, Anthropic and OpenAI. Notably, technology IPOs with dual-class voting structures outperformed the benchmark by 13.8 percentage points on average. This suggests that governance structure may be a significant performance factor, though with a seeming preference for controlling founders as opposed to more classic principles of good governance.

Governance structure may be a major performance factor for U.S. tech mega-IPOs

As of December 2025. Data excludes banks, S&Ls, REITs, ADRs, SPAC’s closed-ended funds and issues with offer price below $5. Data is equal-weighted, three-year return is a buy-and-hold return relative to the cap-weighted market is measured from the first closing price, subsequent return for 2025 IPOs not computed since at least one-year has not elapsed since listing date. Sources: Jay Ritter, University of Florida, Empirical Research Partners Analysis, RBC GAM

The challenge, however, is that valuation matters. When segmenting historical IPO data by price-to-sales multiples, expensive IPOs have consistently delivered inferior returns. For example:

• Those trading above 40x price-to-sales have performed worst (see chart below).

• IPOs priced between 20x and 30x price-to-sales underperformed the benchmark by 16.3 percentage points on average over three years.

• Those priced at more than 40x price-to-sales trailed the benchmark by a staggering 58.5 percentage points on average over three years.

SpaceX (96x price-to-sales), OpenAI (43x price-to-sales) and Anthropic (20x price-to-sales) all fall within these expensive cohorts, suggesting elevated risk of disappointing future returns despite being in a high-growth industries.

Expensive U.S. IPOs have consistently delivered lower returns

As of December 2025. Note: Data excludes banks, S&Ls, REITs, ADRS, SPACs, closed-ended funds and issues with offer price below $5. Data is equal-weighted, three-year return is a buy-and-hold return relative to the cap-weighted marked. Sources: Jay Ritter, University of Florida, Empirical Research Partners Analysis, RBC GAM.

The SpaceX, OpenAI and Anthropic IPOs mark a significant moment for capital markets given their sheer size. But their implications differ at the macro and micro levels.

From a macro perspective, these offerings remain modest relative to the size of the market and have little impact on overall fundamentals. As lockups expire and other AI companies potentially come to the market, the market’s need for capital could become more significant, but is unlikely to prove a major problem.

At the individual stock level, however, elevated valuations mean that these companies will need to deliver sustained exceptional growth to justify their current prices. They are all revolutionary companies in their own way, so this is not impossible. But it does require the space industry (discussed in our last MacroMemo) and the market for AI to truly become world-changing technologies.

In some respects, current conditions echo the late-1990s technology boom. Investors are displaying heightened enthusiasm for transformative new technologies expected to deliver outsized growth and substantial shareholder returns. Like the internet before it, AI has captured imaginations with its potential to reshape entire industries for the better. Whether this period of excitement for space exploration and AI represents another bubble will be clear only in hindsight. While the technologies themselves often do transform the world, investors may want to exercise appropriate caution and manage position sizes given that lofty growth expectations are already embedded in extremely high starting valuations.

-ES

The world’s two largest economies are managing to grow at a reasonable pace even as housing activity remains a drag. This might not seem remarkable given that housing investment only directly accounts for about 5% of U.S. GDP and 8% of Chinese GDP, on average. But related spending on furnishings, appliances and the like, spillover from construction activity, as well as wealth effects from changes in home prices – which are particularly impactful in China, where real estate has been a key savings vehicle – give the sector more reach than official statistics suggest. Adding in rent and homeowners’ imputed rent, which contributes to consumer spending, housing represents more than 20% of GDP.

In turn, it is more than a little unusual that the U.S. economy is growing 0.4ppt above its potential at +2.7% year-over-year, even as housing investment contributed 0.3ppt less than usual – indeed, it subtracted outright from growth in each of the past 4 quarters (see next chart).

U.S. GDP growth remains solid despite the drag from housing

As at 07/16/2026. Sources: U.S. Bureau of Economic Analysis, Macrobond, RBC GAM

Existing home sales remain near their lowest l/cevels since the Global Financial Crisis housing collapse. A decade-long recovery in homebuilding activity after the crisis has given way to a multi-year pullback:

U.S. housing starts are softening and home sales remain subdued

As at 07/17/2026. Sources: National Association of Realtors, U.S. Census Bureau, Macrobond, RBC GAM

The outlook isn’t particularly rosy. Homebuilders’ sentiment remains about 1 standard deviation below its long-run average, even as manufacturing sentiment has picked up nicely this year:

Manufacturing confidence has improved but homebuilder sentiment is subdued

As at 07/16/2026. Sources: National Association of Homebuilders, Institute for Supply Management, Macrobond, RBC GAM

Meanwhile, China’s property market has undergone an unprecedented correction over the past half decade. Yet GDP growth – while also decelerating – has been relatively more resilient:

Chinese GDP growth has held up despite a sustained housing downturn

As at 07/16/2026. Sources: China National Bureau of Statistics, Macrobond, RBC GAM

We recently provided an update on the state of China’s housing market here. In short, sales activity is at a multi-decade low, inventories are elevated and affordability remains challenging, although prices are tentatively rising in Tier 1 cities:

Home price growth in major Chinese cities turns positive

As at 07/16/2026. Sources: China National Bureau of Statistics, Macrobond, RBC GAM

Given the importance of housing to these two economies, how is overall growth holding up so well? Different factors are at play in each country.

In the U.S., the AI CapEx boom has been a key driver of recent strength. We think AI CapEx added about ½ percentage point to growth last year and probably more this year based on hyperscalers’ spending plans. That more than offsets the pullback in housing investment. The pickup in tech investment has been so strong that business investment in software, computers and peripherals now accounts for the same share of the U.S. economy as residential investment:

Tech CapEx and housing investment are now equal shares of the U.S. economy

As of 07/16/2026. Sources: U.S. Bureau of Economic Analysis, Macrobond, RBC GAM

That’s not the only way AI is compensating for U.S. housing softness. AI optimism has driven U.S. equity prices higher, generating trillions of dollars in new household wealth. Studies suggest wealth effects (the additional consumer spending associated with an increase in household wealth) are 1.5x stronger for changes in housing wealth compared with financial wealth. The increase in household equity holdings as a share of GDP has significantly outpaced the modest decline in housing wealth, and now outright exceeds it in value.

Equities now exceed real estate on U.S. household balance sheets

As at 07/16/2026. Assumes 60% of mutual fund shares are equities. Sources: Federal Reserve, Macrobond, RBC GAM

China’s pullback in housing wealth has been more pronounced and impactful. Prices are down more than 20% from their peak, and housing historically represents nearly 70% of China’s household wealth. Equities are a smaller share of household balance sheets than in the U.S. and China’s stock market has been less reliably positive. As a result, China’s housing wealth effect is estimated to be nearly 2x that of the U.S.

Perhaps unsurprisingly, Chinese consumer confidence fell sharply at the onset of the housing downturn and has failed to recover.

Chinese consumer confidence fell sharply at the onset of the housing downturn

As at 07/16/2026. Sources: China Economic Monitoring & Analysis Center, Macrobond, RBC GAM

Consumers’ willingness to spend has improved somewhat, perhaps supported by a trade-in program that allows consumers to save on new purchases of cars, appliances and tech products. But spending in China remains subdued. Retail sales grew by a paltry 1.3% over the first half of the year, barely outpacing inflation.

Chinese retail sales growth has ground to a halt

As at 07/16/2026. Sources: China National Bureau of Statistics, Macrobond, RBC GAM

The economy’s saving grace has been strong export growth. Net trade contributed 1.5 percentage points to China’s GDP growth in 2024 and 2025 and remains a tailwind so far in 2026.

Net exports have been a key contributor to Chinese GDP growth

As at 07/16/2026. Sources: China National Bureau of Statistics, Macrobond, RBC GAM

China has done well to diversify its exports in the face of rising trade tensions with the U.S. Exports to other parts of the world – particularly Southeast Asia – have picked up the slack and then some:

Chinese exports to U.S. plummet though overall exports continue upward trend

As at 07/16/2026. Sources: China General Administration of Customs, Macrobond, RBC GAM

Earlier this year, we discussed China’s technological push, including its emergence as the world’s largest auto exporter and its growing proficiency in technologies including batteries, solar, robotics, AI, semiconductors and pharmaceuticals. Its manufacturing and export prowess has been abetted by a competitiveness boost from currency depreciation (when adjusted for inflation differentials).

China’s currency has depreciated on an inflation-adjusted basis

As at 07/16/2026. Sources: Bank for International Settlements (BIS), Macrobond, RBC GAM

Neither country’s housing market looks like it’s set to spring back to life anytime soon. The worst might be over, but the sector could remain a modest drag on both economies in the near term.

Fortunately, this hasn’t been as bad for growth as one would normally have expected despite strong historical GDP-housing growth correlations of +0.52 in the U.S. and +0.76 in China. There appears to be more juice in the AI CapEx and export trends that are supporting the U.S. and Chinese economies, respectively.

Both countries are also seeing strong productivity growth – which rising AI adoption should help to sustain. This provides a solid underpinning for economic expansion even if housing remains on the sidelines.

-JN

In our #MacroMemo two months ago, we examined whether AI might become another winner-take-all technology or instead develop into a diverse, competitive marketplace. We concluded there is a non-trivial chance of the latter. It’s still early days, but recent developments tilt further in that direction.

Over the past two months, the weighted average price of tokens (the basic units of data processed by AI models) has declined by more than 20% (see chart below). This admittedly comes after average prices doubled during the prior 6 months. But nonetheless it suggests the market’s marginal willingness to pay for AI compute is now falling.

Weighted average token prices are down more than 20% from recent highs

As at 07/15/2026. Sources: Bloomberg, Silicon Data, RBC GAM

Lower average prices could indicate a general decline in token prices or a compositional shift toward cheaper models (or some combination of the two). For any given level of intelligence, token prices tend to fall exponentially over time (note the log axis on the chart below) as competing models enter the market. Open-weight models lag their closed-weight counterparts in performance by only a few months, but their token prices are 80% lower on average according to Silicon Data’s expenditure index. Some models are priced at an even steeper discount.

But enterprise users don’t always switch to newer, cheaper models, which might involve changing providers every few months. One survey found only 11% of enterprises switched model vendors in the past year. Two-thirds of enterprises chose to upgrade models within their existing provider and 23% made no change.

This stickiness means the prices paid by users and charged by individual vendors don’t necessarily decline as sharply as the chart below suggests, even if there are cheaper, equally capable alternatives available.

Token prices fall exponentially for any given level of model capability

As at 07/08/2026. Price is a 7:2:1 blend of cache, input and output token prices. Grouped by Artificial Intelligence Index 4.1. Sources: Artificial Analysis, RBC GAM

So, the weighted average price of tokens depends in part on how many users trade up to new frontier models that command premium pricing vs. sticking with legacy models or opting for cheaper versions with similar capabilities. A significant leap in the intelligence of frontier models from Anthropic and OpenAI earlier this year appears to have attracted more users and driven average prices higher. But that is now reversing as cheaper, open-weight Chinese models make significant inroads (see chart below).

So far in July, the top 5 models based on token consumption on OpenRouter (a platform that allows access to multiple AI models) are all Chinese. This suggests compositional effects are a key driver of the recent decline in weighted average token prices.

Chinese models have taken the lead in OpenRouter token consumption

As at 07/14/2026. Based on top 50 models tracked through OpenRouter. Sources: OpenRouter, RBC GAM

Chinese models aren’t just competing at the low end of the price spectrum. Moonshot AI caused a stir in mid-July with the release of its Kimi K3 model that it says rivals frontier models from OpenAI and Anthropic. Benchmarking platform Artificial Analysis concurs, assigning Kimi K3 a score of 57 on its Intelligence Index, just a touch below scores of 59 and 60 for OpenAI and Anthropic’s top models, respectively.

Based on its intelligence, pricing and token efficiency, Artificial Analysis assigns Moonshot’s model a weighted average cost per task of $0.95, slightly undercutting OpenAI’s model ($1.04) and well below Claude’s frontier model ($2.75). But that’s still a significant premium to DeepSeek’s V4 Pro model ($0.04) which has an Intelligence Index score of 44.

Beyond price, supply chain security factors in as well. AI tokens are increasingly viewed as a critical input for businesses. This has put the focus on cybersecurity and geopolitical risk considerations. While some management teams are reportedly concerned about data security and wary of cheaper Chinese models, the Trump administration’s recent export controls on Anthropic’s frontier models also have some international users re-evaluating their dependence on U.S.-based providers.

This all comes as a number of early adopters are cutting back on token spending. Tech companies previously encouraged their employees to experiment freely with new AI tools. The number of tokens an employee consumed was seen as an indicator of productivity and AI savviness. NVIDIA CEO Jensen Huang said he’d be alarmed if a software engineer paid $500,000 per year wasn’t consuming $250,000 in tokens annually.

But rising use of agentic AI models that consume significantly more tokens for complex, chain-of-thought reasoning, combined with the shift toward token-based rather than seat-based enterprise pricing, has caused some companies’ AI spending to balloon unsustainably. For example:

• Microsoft and Uber are among those limiting access or capping spending on frontier models. Uber reportedly blew through its 2026 AI budget in just four months.

• Companies like DoorDash, Airbnb and Siemens are shifting some of their AI workloads to Chinese open-weight models.

• Salesforce is closely tracking token consumption to ensure it contributes to better business outcomes.

• Research firm SemiAnalysis suggests many enterprise users are setting spending limits amid the shift to consumption-based pricing, although AI budgets vary widely.

• Some users are also limiting access to premium tier models or adopting new tools that direct tasks to the most appropriate model.

Growing competition and spending restraint are a challenging combination for frontier model developers who spend billions on R&D compute. The good news is that the cost to produce tokens generally declines as GPUs become more powerful. NVIDIA notes that while its newer Blackwell chips cost nearly twice as much per GPU per hour as its earlier generation Hopper chips, they produce 65x as many tokens, resulting in a cost per token that is 35x lower.

Falling token prices (for any given level of intelligence) also encourage AI adoption, which is growing faster than weighted average prices are declining. Weekly usage of models on OpenRouter is up more than 700% year-to-date. Google recently said it is processing 7x more tokens than a year ago. Even during the recent period when average prices fell by 20%, OpenRouter token consumption was up 50%.

This is Jevons paradox at work: falling prices encourage more demand, resulting in an increase in overall consumption/revenue. Goldman Sachs estimates token consumption will increase 24x between 2026 and 2030, driven by growing adoption of agentic AI.

That should leave room for many players with different pricing tiers and capabilities. Enterprise users are becoming more discerning when it comes to their AI spend but are unlikely to turn away entirely from premium frontier models -- particularly for applications where smarter AI tools might confer a competitive advantage.

However, for tasks where cutting-edge performance isn’t necessary and lagging behind by a few months isn’t particularly consequential, users seem to be waking up to the value on offer from open-weight models. If enterprise usership becomes less sticky and more companies are willing to switch model providers to get the lowest price, the profitability of pricier, closed-weight providers could be at risk.

We posit the recent decline in weighted average token prices isn’t cause for panic at this juncture but bears close monitoring as this nascent market continues to evolve.

-JN

The U.S. midterm election is now less than a third of a year away. The purpose of our early start is to get thinking about them before markets have had a chance to fully price in their implications.

The political configuration going into the election is a Republican White House paired with a Republican-held House of Representatives and Senate. The Republicans presently have only a small advantage in the Senate (53 to 47) and the House (218 to 212).

The House of Representatives is expected to flip from the Republicans to the Democrats, with a fairly high level of conviction (see next chart). This is despite the fact that redistricting means that the Democrats will need to win by 3-4 percentage points to capture the House.

Democratic Party poised to take control of the House

As of 07/16/2026. Sources: Polymarket, Bloomberg, RBC GAM

Conversely, the Senate race is still close. For a moment, during the initial phase of the Iran war, the Democrats were even expected to win. But that expectation has since flipped back to the Republicans (see next chart). Why?

• The race structurally favours the Republicans because only one-third of Senators are up for re-election and this year happens to skew toward incumbent Democrats.

• The Republicans also hold a national fundraising advantage.

As such, for the purpose of the discussion that follows, we assume the Republicans retain the Senate. But it is not a certainty, and a messy budget battle leading up to the September 30 end of the fiscal year is more likely to be blamed on the Republicans.

Republicans continue to lead in betting markets for Senate race despite ongoing Iran War

As of 07/16/2026. Sources: Polymarket, Bloomberg, RBC GAM

Implications

We are not convinced that the midterm elections will be a key market driver in the months ahead, especially given powerful AI trends, swirling geopolitics, the gyrating price of oil, the Fed in the hotseat and other critical issues that may divert attention. There are several reasons why the reaction to these specific midterms may underwhelm:

• A flipped House of Representatives is the consensus expectation. Markets may have already priced this in given that they are forward-looking.

• It is completely standard for the party of the president to lose the midterms.

• Legislation has proven less central in the governing process over the last two years – Senators and Representatives have been significantly side-lined by a powerful Oval Office. This election therefore matters less than usual.

Still, at the margin, we figure the midterm elections could result in a slightly weaker stock market, slightly lower yields and a marginally weaker U.S. dollar. It is hardly a slam-dunk affair: there are nearly as many vectors pointing in the opposite direction relative to each of these conclusions. But this is where the balance of risks lies, in our estimation.

Slightly stock market negative

There are a number of ways to evaluate the stock market outlook.

We ultimately expect the midterm election to slightly weigh on stocks. However, a simple reading of how the Dow Jones Industrial Average has historically fared leading up to midterm elections (weak) and following them (strong) makes the opposite case (see next chart). Midterm election years tend to be the weakest portion of the four-year presidential cycle, and the year after tends to be strongest.

U.S. market performance tends to be the weakest in mid-term election years

As of 07/16/2026. Based on daily data back to 01/01/1900. Sources: Ned Davis Research, Bloomberg, RBC GAM

Another approach is to evaluate the specific expected permutation of political parties that should emerge after the election. The current Republican President / Republican Congress is historically associated with a solid but unspectacular 7.3% gain per annum in the equity markets (see next table).

Historically, equity markets performed poorly under a Republican president and a split or Democratic Congress

U.S. government composition and performance of Dow Jones Industrial Average

Based on daily data from1900 to 2024. Source: Ned Davis Research

If Congress flips to a split, as is expected, that results in a much less favourable permutation. With a Republican president and split Congress, the average per annum market return has historically been an unappetizing –2.0% per year.

However, we hesitate to endorse such conclusions because they represent the historical return over a small handful of time periods and some historical oddities interfered. The Republican President / Split Congress combination most recently occurred between 2019—2021 (the pandemic) and 2001-2003 (9/11 and a recession). These market traumas were not the specific result of a particular Congressional configuration.

In our view, it is far more useful to evaluate the stock market outlook in the context of specific policy changes that may occur over the next two years due to the election. Let us do that.

An important if broad observation is that today’s Democrat Party is less business-friendly than the Republicans. This is negative for most companies.

Favourable legislation may not be passed (though outright unfavourable legislation is unlikely on any scale given that the Republican Party will still wield considerable power). If the Democrats retake the House, they gain subpoena power, creating a risk for corporate executives across a range of sectors. A variety of Congressional investigations are likely to follow. The composition of key committees including Budget, Finance, Ways & Means and Banking will also be affected.

Republican deregulation efforts – of greatest benefit to the financial and energy sectors – may be slowed by the addition of Democrat influence.

The budget itself will be more difficult to pass given that it will require cooperation across parties. It wasn’t exactly easy even when the Republicans had the majority in both chambers. The U.S. is next expected to hit its debt limit somewhere between late 2026 and mid-2027 – more likely the latter, in our view – and raising it could be quite contentious.

To the extent the Democrats are currently styling themselves as the party of fiscal responsibility and given that a divided Congress usually limits fiscal excesses, the public debt may not grow quite as quickly in the years after the midterm election. This would be a short-term GDP negative, though a long-term fiscal and GDP positive.

Republican deregulation efforts – of greatest benefit to the financial and energy sectors – may be slowed by the addition of Democrat influence.

Some experts argue there is no major partisan divide between the two parties over AI. Both are largely “pro.” But the Democrats may express stronger reservations about potential labour disruptions and could also more forcefully express their concern about increasingly unpopular data centres and their electricity consumption. Notably, New York State just imposed a moratorium on new data centres. As such, the AI sector could be slightly more encumbered than before.

Both parties care about pocketbook economics and the consumer, and so any policy action in this realm is more likely to be supportive than antagonistic.

The U.S. pharmaceutical sector might experience negative repercussions if the Democrats push for more drugs to be subjected to price negotiation. Though it is possible that more generous Medicaid benefits would increase health spending for the broader health care sector.

The oil and gas sector might fare worse with greater Democratic Party influence, while the green sector might enjoy a small revival.

The Democratic Party is thought to be less supportive of defence and aerospace spending. However, the subject is nuanced given very different opinions about the war with Iran versus the Ukraine-Russia war between the two parties.

Conversely, concerns about the risk of tax hikes are likely overblown. The Democrats would not have enough clout to enact such changes via legislation. And – unlike with several prior temporary tax initiatives – last year’s One Big Beautiful Bill Act permanently enacted several key provisions including bonus depreciation, R&D expensing and lower individual tax rates. Therefore, these are not at serious risk.

Both parties care about pocketbook economics and the consumer, and so any policy action in this realm is more likely to be supportive than antagonistic.

Tariff pressures are unlikely to substantially change given that the ones implemented so far have been enacted using executive orders rather than new legislation. However, the Democrats could attempt to undermine the process somewhat and could complicate any effort by the White House to significantly curtail the USMCA agreement with Mexico and Canada. Changing the “fundamental architecture” of the deal would require a Congressional vote and encounter less receptive Democrats (though Democrats are not themselves classically pro-free trade).

Slightly lower yields

Historically, Republican governments are associated with slightly higher yields and Democrat governments with the reverse. Thus, a shift in the direction of the Democrats makes a (weak) case for lower yields.

More fundamentally, monetary policy should not be greatly altered in one scenario versus the other, though it might be slightly less restrictive if there were less fiscal stimulus and a slightly less business-friendly economic environment.

From a fiscal standpoint, a divided Congress is likely to be a bit less stimulative, which means marginally less bond issuance – bond yield negative.

Marginally weaker dollar

We have a longstanding secular view that the U.S. dollar should gradually decline over time. The midterm election perhaps marginally adds to that claim. Dollar-negative forces may include slightly less fiscal expansion, a slightly lower tariff risk and a slightly slower rate of economic growth.

In conclusion, the midterm elections are unlikely to be the main market driver but are worth watching. They may have a slightly negative effect on equities, yields and the dollar. More substantial effects may be visible at the sectoral level.

-EL

We all talk about the U.S. stock market an awful lot, and that makes sense. The U.S. economy is very large and also a global bellwether.

But quite often the U.S. market almost seems like the only story in town -- and it isn’t just due to the current AI frenzy. Longstanding traditional asset mixes often assign a substantially larger equity allocation to the U.S. than the country’s 15-26% share of global economic output (depending on the definition).

Why is this? There are several solid reasons.

The U.S. is a huge part of global equities

The key insight is a well-trodden one: “the economy is not the stock market.” Much as that pains economists to admit, it is true. In the case of the U.S., the value of all U.S. publicly traded equities is a startling 51% of the world’s listed stocks. That’s a single country representing more than half the total and it is 2x to 3x higher than the country’s economic weighting (see next chart).

U.S. equities outmuscle the U.S. economy

As at 2025-2026. U.S. share of global GDP based on International Monetary Fund (IMF) forecast. Sources: IMF, Bloomberg, RBC GAM

A reasonable rebuttal to this is that U.S. stocks are overvalued according to conventional valuation metrics, and so evaluating the market during this particular moment is exaggerating just how central the U.S. should be for equity investors. But the counterpoint is that once you factor in sector composition (the tech sector share of the U.S. market is rising and it traditionally has higher valuations) and the remarkable upward trajectory of margins and earnings, the overvaluation argument becomes, at a minimum, debatable.

Embedded international exposure

Even though the U.S. is a huge share of the global stock market, one might still feel trepidation at such a concentrated exposure to a single country. Fortunately, it is a more geographically diversified holding than it first looks (see next chart). In the S&P 500, an impressive 42% of company revenues are actually generated outside the U.S. (The fraction of S&P 500 foreign earnings is lower – 25% – but this may be sandbagged by transfer pricing and other tax maneuvers).

U.S. stock market has high international exposure

As at 2025-2026. Sources: Bloomberg, Mizuho, RBC GAM

The S&P 500 tech sector leads with approximately 54% of its revenues derived from foreign markets. Prominent examples include Broadcom at 74%, Apple and Meta both at 57% and Alphabet at 52% (see next table). But it isn’t just tech companies with large foreign orientations: Visa has a 61% foreign revenue share, Caterpillar has 46% and Johnson & Johnson has 43%.

Many S&P 500 companies derive a significant share of their revenues from foreign markets

As at 2026. Sources: Bloomberg, RBC GAM

As an aside, this foreign revenue exposure significantly explains why the U.S. stock market is so sensitive to currency movements: a large fraction of the revenue grows or shrinks on that basis.

The world comes to the U.S.

Constituting a related but distinctly tertiary driver, more foreign companies are listing in the U.S. and issuing shares there. SK hynix, one of the prominent Korean memory chip companies, just raised US$26.5 billion in the U.S. this month. The companies receive a durable valuation uplift that averages about 20%, and the U.S. market promises access to an enormous amount of capital – domestic and foreign – and offers a deep, liquid market. In turn, the U.S. market becomes ever more international (though SK hynix will not appear in the major U.S. indices).

U.S. stock market special ingredients

It hasn’t hurt that the U.S. stock market has resoundingly outperformed its peers over the past 15 years – both in terms of convincing investors to up their exposure and passively pulling it higher via capital gains. But that isn’t the source of the structural overweight. The U.S. market stagnated across the 2000s and yet such positioning existed during that period and long before.

Instead, part of the logic is that the U.S. has some special ingredients that render it structurally attractive to investors. These include:

• its large economy that benefits from economies of scale

• deep, liquid and sophisticated markets with a strong shareholder-return culture

• strong rule of law and property rights

• a capitalistic, entrepreneur-oriented economic system.

Not coincidentally, several of these features have enabled U.S. companies to benefit from a first-mover advantage in a range of technology-oriented industries.

Unique sector exposure

Related to the above built-in advantages in the U.S., international investors also favour the U.S. because it offers certain unique or at least unusual exposures that are difficult to replicate elsewhere. These include exposure to:

• The AI revolution – the hyperscalers and chipmakers, significantly.

• Global leaders in entertainment and media – Netflix, Disney, Alphabet/YouTube and so on.

• Defence & aerospace – even as the rest of the world seeks to build up its own such industries, the U.S. complex is by far the largest.

• Biotech and pharmaceuticals – a mix of large, established players and hundreds of clinical-stage companies.

Bottom line

In conclusion, it makes sense that traditional investment portfolios allocate a larger share to the U.S. than the country’s economic clout would strictly suggest.

• The U.S. stock market is comparatively enormous.

• It contains quite a large exposure to the rest of the world.

• The investible universe is itself tilting toward the U.S. as foreign companies increasingly present themselves to the U.S. market.

• The U.S. economic system has some special ingredients that are particularly attractive to investors.

• The U.S. sector mix contains some unique/unusual components that international investors desire.

Most of these are not new arguments, nor are we arguing to overweight U.S. equities at this juncture. But this analysis does help to explain why the U.S. continues to figure so centrally in many investment portfolios, even as there are concerns about valuations and questions about whether U.S. exceptionalism might be starting to ebb.

-EL