Buy and Think

In late March, a longtime client forwarded me an article from The Atlantic—Welcome to a Multidimensional Economic Disaster, by Charlie Warzel and Matteo Wong—that captures the current AI-buildout fears more sharply than most. The article makes several arguments at once. Some concern the geopolitical fragility of an AI supply chain dependent on a handful of chip producers in South Korea and Taiwan and on energy and materials moving through the Strait of Hormuz. Others concern record hyperscaler debt and circular investment chains among hyperscalers, model providers, and chip vendors that some observers have compared to 2008. Those are real operational risks. They become genuinely systemic only if the article’s underlying business-model claim is correct—what Paul Kedrosky, quoted in the piece, calls a “death spiral to zero” for tokens. If a falling inference cost is a symptom of a deflationary collapse, then hyperscalers cannot service their debt, data centers cannot be filled, and the supply chain becomes fatal. If instead it is the catalyst that unlocks more demand than the prior price tier could support, the same token-cost decline is a success story rather than an obituary. Many people still have a problem believing the latter case, also known as the Efficiency Paradox or Jevons Paradox: when the cost of useful work from a resource falls, total consumption tends to rise rather than fall. The market is now rediscovering this 161-year-old idea, by name and by inference, in prepared remarks at every major hyperscaler and at Palantir.

What Jevons Actually Said

William Stanley Jevons published The Coal Question in April 1865. Britain was the world’s largest coal producer, and there was growing public anxiety about how long the country’s reserves would last. James Watt’s improvement on Newcomen’s steam engine had made coal-fired engines dramatically more efficient, and the expectation was that this efficiency would lower coal usage and extend British reserves. The opposite happened. Cheaper coal-powered work made railways, factories, and steamships economically viable on a scale that had never existed, and total British coal consumption rose multiple-fold over the following decades.

Jevons’s own statement is worth reading directly:

The modern generalization, developed by Daniel Khazzoom and Leonard Brookes in the 1980s, makes the macro point: efficiency gains in a critical input are typically net additive to consumption rather than net subtractive. When the cost per unit of useful work falls fast enough, demand becomes use-case-elastic. Cheaper inputs unlock entirely new categories of application faster than they retire old ones.

Semiconductors: Sixty Years of Wrong Forecasts

Anyone who has owned semiconductor stocks for more than a cycle knows this is not a theoretical argument. The semiconductor industry has been the cleanest empirical demonstration of Jevons logic for six decades, and the sell side has been on the wrong side of it almost every cycle.

Moore’s Law is, at its core, a Jevons engine. The price per transistor has fallen roughly by half every two years since the late 1960s. For each semiconductor generation, the sell-side debate has followed the same pattern: ASPs are under pressure, the cycle is maturing, end markets are saturated. And each generation, the use-case explosion has outrun the efficiency gain, often by an order of magnitude.

DRAM is the cleanest example. Through the 1990s and 2000s, falling memory prices were repeatedly read as a leading indicator of broad semiconductor weakness. Veteran sell-side analyst Richard Whittington put the pattern this way in retrospect:

Each missed wave—smartphones, hyperscale cloud, now AI training and high-bandwidth memory—was larger than what consensus was modeling at the time. The cost-per-bit decline was the catalyst, not the constraint.

GPUs are an even sharper example. For most of their first two decades, they were modeled as a gaming-only product with a limited addressable market. Crypto mining added an unanticipated demand layer in the mid-2010s. AI training and inference added another in the late 2010s and 2020s. The miscalibration has been so persistent that the industry’s own CEOs have had to publicly walk their numbers up. At AMD’s Advancing AI event in May 2024, Lisa Su told the audience that AMD’s prior estimate had been wrong:

A 167 percent upward revision in twelve months, from the company that designs the chips. Sell-side has been chasing the number ever since.

The pattern is consistent enough to be a rule. Treating an efficiency gain as a ceiling on demand has been the most reliably wrong analytical move in semiconductors for two generations. Jevons described the misinterpretation in 1865. The industry has been illustrating it ever since.

The AI Hijack Meets Jevons

As I described in Enter the Lame Duck in December 2025, AI infrastructure spending—roughly $423 billion in 2025—is currently doing the bulk of the work of US GDP growth. The Jevons logic justifies the number expanding rather than compressing in 2026.

The first public corporate invocation came from Satya Nadella in the early hours of January 27, 2025, in the middle of the DeepSeek panic, when he posted on X that “Jevons paradox strikes again.” By the May 2026 earnings cycle, the framework had moved from social-media reaction to prepared remarks.

The most striking number came from Palantir’s May 4, 2026 earnings call. CTO Shyam Sankar observed that GPT-4-equivalent performance, which cost $20 per million tokens in early 2023, is now approximately one thousand times cheaper three years later. Aggregate token consumption has grown faster than that price decline. Each new price tier has unlocked an application class—agentic workflows, browser tools, real-time voice, autonomous coding, enterprise document processing—that did not exist when the prior tier was the floor.

Sankar’s framing on the same call was the cleanest executive statement of the thesis I have seen:

Sundar Pichai made the same point on Alphabet’s April 29 call. Google has reduced the cost of core AI responses by more than 30 percent since upgrading to Gemini 3, while Google’s first-party models now process more than 16 billion tokens per minute, up from 10 billion the prior quarter. Sixty percent sequential growth in token throughput while unit costs fell thirty percent. Jevons in two sentences of prepared remarks.

Jensen Huang made the bluntest version of the case on Nvidia’s February 25 call. The new GB300 NVL72 system delivers up to 35 times lower cost per token than Hopper, he said, and then the line that should be framed on the wall of every sell-side analyst who has called peak capex over the last three years:

Huang reiterated that cloud customers are on track for approximately $700 billion of data-center capex this year, with a plausible path to $3 trillion to $4 trillion by 2030.

The pattern across these calls is what I have been calling, in correspondence with clients, the Inference Famine. Microsoft’s CFO Amy Hood put it bluntly on the same earnings cycle: “broad and growing customer demand continues to exceed supply.” Total spending is going up not despite the price decline but because of it.

I noted in Enter the Lame Duck that the “circular economy” narratives around AI financing bear scrutiny. They still do. But those concerns are operational rather than structural. The structural question—whether the falling unit cost of compute will collapse aggregate spending—has been answered by 60 years of semiconductor history, the current earnings cycle, and 161 years of Jevons logic. Efficiency is the catalyst here, not the curfew.

Investment Outlook

The portfolio implication follows directly. I continue to focus on the full infrastructure ecosystem that AI workloads require, not on the chip layer alone. As I argued in Brave New World (2025), in FOMC’s Pivotal Moment – Smaller Companies Set for a Surge, and in the “More of Everything” passage of my December 2024 commentary, the AI megatrend requires data centers, electrical generation and grid capacity, ultra-high-speed connectivity inside and between data centers, high-bandwidth memory, advanced cooling, and a long tail of specialized power-management and physical buildout suppliers. Jevons’ logic predicts that demand for each of these inputs will grow faster than its cost falls, and that this growth will continue to surprise the consensus.

The implication is not “own Nvidia.” It is to own a concentrated set of second- and third-derivative suppliers whose products become more valuable, not less, as the per-unit cost of AI work continues to decline. Many of the highest-conviction beneficiaries are not in the Magnificent Seven. They are in small- and SMID-cap categories that the market has continued to underweight on the assumption that efficiency reduces demand.

Three independent tailwinds now point in the same direction for the rest of 2026 and beyond. The 9- to 12-month transmission lag of the rate-cut cycle is finally working through the broader economy. The political mechanics I described in Enter the Lame Duck have largely resolved in the direction I anticipated, and the resolution extends well beyond tariffs alone. Judicially, the Supreme Court’s 6–3 ruling on February 20 in Learning Resources, Inc. v. Trump held that IEEPA does not authorize the President to impose tariffs; when the administration pivoted within hours to a 10 percent universal tariff under Section 122 of the Trade Act of 1974, the U.S. Court of International Trade subsequently struck that down too. Section 232 metals tariffs remain in place, but the broader tariff architecture has been dismantled by the courts. Politically, the midterm calendar is reinforcing the same direction across regulatory posture, fiscal policy, and what remains of the trade agenda, likely compounding the judicial relief. And the Jevons logic of AI demand is moving from research note to consensus, which suggests that the capital-spending leg of the cycle has further to run than the bear case allows.

High-conviction, concentrated, equal-weighted exposure to the best large-, SMID-, and small-cap companies, positioned across all three tailwinds, remains the most direct way to capture what comes next.

← Back to top

As 2025 closes, the investment landscape is defined by two powerful currents: one, the permanent technological acceleration of AI, and the other, the temporary political drama of a “lame duck” era. Both are creating distinct risks and opportunities.

The AI Hijack

On the technology front, AI has hijacked the current state of the economy. The AI megatrend, a core theme of my January 2025 Brave New World (2025) report, continues to expand its reach. This AI supercycle is dominating the US economy; AI infrastructure spending—estimated to reach $423 billion this year—is driving US economic growth and accounting for most of the US GDP growth in 2025, offsetting weakness in consumer spending and manufacturing. This is a powerful confirmation of the U.S. resilience thesis I articulated in January. While tariffs affected other sectors, the US economy’s unique structure enabled an internal, AI-driven investment boom that offset them, accounting for nearly all of the year’s net economic growth.

This growth is not without consequence. The tariffs, weaponized by the White House, continue to adversely affect most other areas of the economy, from US consumers and manufacturing to housing and the food supply. Separately, the AI data center buildout is driving up energy prices. As I warned in Brave New World (2025), this surge in AI capital spending is now directly pressuring the energy grid and infrastructure. This, in turn, is creating new investor skepticism about the actual cost of financing this buildout, and whether “circular economy” narratives are merely a new form of “financial engineering” to obscure the massive, long-term costs of this infrastructure.

The Pivot and The Pause

This “AI Hijack” has also complicated the market recovery I anticipated last fall. In my September 2024 note, FOMC’s Pivotal Moment, I predicted that the start of interest rate cuts would finally allow smaller companies to surge, ending the singular dominance of the mega-caps.

I observed this play out briefly: markets broadened during the fourth quarter of 2024 as the “rising tide” of lower rates began to lift all boats. However, the renewed volatility in early 2025—driven by tariff shocks and the sheer gravitational pull of AI spending—stunned the recovery, temporarily pushing investors back into the safety of the “Magnificent Seven.” While the timeline has extended, my conviction remains: the long-term effect of rate cuts (which take 9–12 months to fully cycle through the economy) creates a powerful tailwind for broader market growth as we head into 2026.

The Return of Goldilocks DC

Following the shift in political momentum after the solid Democratic wins in the November 4th, 2025, elections, the 2026 U.S. midterm elections are increasingly affecting investors’ outlook as we enter a de facto lame-duck era of political paralysis. This political realignment signals a potential end to the extreme tariff volatility I identified as the primary macroeconomic risk earlier this year.

This unfolding dynamic brings me back to a concept I discussed in Politico in August 2018: the concept of a “Goldilocks DC.” At the time, I argued that a political stalemate—specifically a divided Congress checking a “politically damaged president”—was the ideal scenario for investors. Then, as now, the primary threat to economic growth was, to use my words from that time, “boneheaded tariffs evolving into [a] trade war.” I predicted that a gridlocked government would preserve favorable tax and regulatory structures while limiting the White House’s ability to disrupt global trade.

As we approach 2026, I see a similar “Goldilocks” scenario re-emerging. The approaching midterm election cycle increases the political cost of trade friction. The closer we get to the 2026 midterm elections, the more likely it is that these tariffs will be paused, rolled back, or eliminated to support consumer purchasing power and industrial growth and stability. A potential return to legislative paralysis could effectively end the current era of tariff volatility, leaving the path clear for the AI-driven economy to thrive without government interference.

There is a potential second path that could help dismantle the current tariff regime. The U.S. Supreme Court is currently deliberating on Learning Resources v. Trump, which challenges the administration’s use of the International Emergency Economic Powers Act (IEEPA) to impose broad tariffs. During oral arguments on November 5, 2025, several Justices expressed skepticism about whether the President has the unilateral authority to levy what effectively function as taxes. A ruling against the administration, expected in early 2026, could invalidate tariff architecture and trigger refunds of nearly $200 billion in collected duties. While the White House may attempt to pivot, a Supreme Court rebuke would significantly raise the bar for any future unilateral trade actions.

Investment Outlook

I continue to invest in what I consider to be the most attractive companies in this evolving environment. Digital transformation megatrends like Generative AI, the Fourth Industrial Revolution, robotics, and electrification do not exist in a vacuum. They all require large infrastructure ecosystems to support their growth. As a result, I am looking beyond technology providers to companies building the physical and digital infrastructure necessary to sustain this supercycle. As the dual catalysts of judicial intervention and midterm pressure begin to neutralize tariff volatility, there will be increasing opportunities to capture the resulting margin expansion. Small and SMID-cap companies are likely to benefit most from a rollback in trade friction and lower interest rates.

Despite the recent volatility, I remain confident that companies with superior earnings growth will likely outperform over the long term. My focus remains on identifying the best large, SMID, and small-cap companies that possess either established high growth or the potential for significant acceleration. By utilizing high-conviction, concentrated portfolios with high active share, I can isolate these specific winners rather than owning the “market.”

This approach allows us to look past the top-heavy benchmarks. Instead of chasing the few dominant names that have already won, our equal-weighted design positions us to capture the upside of the next generation of leaders—companies with the strongest fundamentals and double-digit earnings growth expectations. Whether they are established technology giants or emerging leaders within the broader industrial base, these companies are well-positioned to lead the recovery as the gridlock in Washington clears the way for fundamentals to matter again.

← Back to top

As we approach significant changes in Washington coinciding with the beginning of the new year, we can reflect on the many 2024 milestones. It was an unusually complex and eventful year.

On the economic front, The Federal Open Market Committee (FOMC) initiated a rate cut process at their September meeting—14 months after stopping interest rate increases. The initial 0.5% rate cut set FOMC on target to lower the target rate for the rest of the calendar year. The process continued unabated until FOMC’s Chairman Powell’s commentary on December 18th suggested challenges to the expected path of interest rate adjustments. Powell indicated that some FOMC members chose to include the potential effects of the incoming government policies in their future interest rate expectations. Some of the promised (or threatened, depending on the audience) policies could slow economic growth and revive inflationary pressures. The market reaction was swift, with the largest drop in equity markets in many years, decimating smaller companies’ returns for the calendar year.

Meanwhile, the US economic expansion continued unabated, with continued labor market strength, moderating inflation, and a rebound in corporate earnings.

The threats of the proposed policy negatively affecting economic growth and inflation are real. We believe the US government will use the tariffs to get attention, express the US position, and negotiate various agreements. However, if the government implements blanket tariffs as promised, they will significantly affect the global economy.

Regardless of how realistic one considers the threat of tariffs as a primary economic tool, everyone with potential exposure to their effects must prepare in advance. Supply chain disruptions are not such a distant memory and have become a traumatic lesson for many companies worldwide. We will likely see a pull forward of production and shipment of products, which could be affected by the largest of the threatened tariffs. One recent example is a rumor of Nvidia and AMD accelerating manufacturing and shipping their next generation of video cards—aiming to achieve deliveries to the US before the presidential inauguration. I expect to hear many more such examples.

The trade restrictions and supply chain risks are very real. However, we must also remember that the US economy is very unusual. It is one of the most closed economies among developed nations, as consumer consumption drives the US GDP. Exports of goods and services (as of 2023) are only 11.2% of the US economy, with imports representing 15.3% of the GDP. The US is also energy-independent. We have been a large energy exporter for many years and the world’s largest oil producer since 2017.

While the US economy’s strength is undeniable, one can also argue that traditional measurement of GDP underappreciates the unique strength of the US economy.

The traditional GDP formula adds up all spending done by different economic sectors.

C (Consumer spending): Spending by individuals on goods and services

I (Investment): Spending by businesses on capital goods like machinery and equipment

G (Government spending): Spending by the government on goods and services

(X − M) (Net exports): The difference between the value of exports (X) and the value of imports (M)

The United States has a unique GDP structure. Among Major Advanced Economies (G7), it has the highest percentage of consumer spending (68.9% 2023 est.), the lowest percentage of government spending (17.1% 2023 est.), and by far the lowest percentage of Exports (11.2% 2023 est.), Imports (15.3% est.) and Net Exports (−4.1% 2023 est.).

We believe this unusual economic structure allowed the US to recover faster than any other advanced economy. It also makes the US more resistant to geopolitical turmoil, trade wars, and even to the effects of potential trade tariffs. The tariff threats to Mexico, Canada, China, and the European Union could significantly affect their economies, with a relatively lesser impact on the US economy overall.

Considering the US economy’s structure, the focus on desperately increasing exports is largely misguided. Luckily, the threat of imports being affected by retaliatory tariffs can only have a relatively small effect unless it would involve shutting down entire supply chains like during COVID.

The US economy exceeded most expectations in 2024, driven by consumer spending and rising incomes. There is also a potential for a surge in global trade to frontload both exports to the US and imports by the US companies worried about supply chains, materials, and tariffs-driven inflationary pressures.

We are entering 2025 with a combination of a high degree of uncertainty and business optimism towards a lower regulatory environment and pro-business bias in Washington. The risk of tariffs creates uncertainty around supply chains, imported goods, ingredients, and materials. This uncertainty also extends to the industries, businesses, or customers that might be affected.

The uncertainty is mainly short-term, while the positive expectations are long-term. Looking ahead to 2025, we are optimistic about the opportunities. The US economy has fully recovered from the COVID disruption and is once again leading global growth. We expect the US economy to outperform other major economies in 2025, and the previously discussed accelerating megatrends will offer exciting areas for investment growth for many years.

We also expect market breadth to continue to improve. Some of the Magnificent Seven stocks are finally experiencing challenges to their dominance. At the same time, the anticipated surge in capital spending related to the broad acceptance and implementation of AI tools should benefit a wide range of suppliers, including more traditional technology companies, energy providers, and infrastructure companies of all sizes.

← Back to top

As we approach the end of summer, a crucial moment in the current economic cycle is on the horizon. After increasing and stabilizing interest rates, the Federal Open Market Committee (FOMC) is likely to initiate a rate cut at its meeting on September 18th. This decision could significantly impact the economy.

Despite the volatility in the equity markets, not much has changed in the last 14 months since the FOMC stopped increasing interest rates. As I wrote in July last year, the FOMC’s dual mandate is to promote maximum employment and provide stable prices and moderate interest rates over time. This mandate is incredibly challenging when faced with full employment and high inflation. Increasing interest rates to arrest stubborn inflation will eventually result in a decelerating economy and a slowing labor market. Too much of such a slowdown will result in a recession. An interest rate increase that is too slow might result in stagflation. As we have seen in the 2023 banking crisis, an interest rate increase that is too fast will likely break things along the way. The FOMC believes interest rate changes take nine to 12 months to affect the economy fully. This means that monetary policy needs to try and predict the state of the economy for 9 to 12 months ahead, mainly ignoring the short-term noise emanating from high-frequency economic data, financial media, and innumerable self-proclaimed experts promoting their ideas in social media.

The effect of interest rate change on the economy is slow. However, even the expectation of lower borrowing costs has a powerful and quick impact on the psychology of consumers and business executives.

As a result, US consumer spending rebounded in August, and consumer sentiment rose to a four-month high in early September. Consumers’ perception of their expected financial situation increased to the highest level since April.

It takes longer to discover any change in business executives’ outlook. We will not get any real data until the next earning season. Still, many companies were slowing down spending decisions or delaying large contracts while waiting for more evidence of an economic soft landing and the beginning of the interest rate decline.

The fourteen-month lag between FOMC’s last interest rate increase and the expected interest rate cut next week had a disproportionately deleterious effect on smaller companies. Regardless of their specific circumstances, they are often viewed less favorably during periods of higher interest rates. When the rates finally start to lower, the rising tide is likely to lift all the boats, with smaller companies likely outperforming larger ones. This will likely also spell the end of the reign of the Magnificent Seven mega-cap stocks, which kept reasserting their dominance throughout 2023 and the first half of 2024.

We continue to believe that the economic slowdown will be short-lived. We expect the combination of low unemployment, continued robust consumer spending, and several accelerating Megatrends (within the context of declining interest rates) to result in a soft-landing scenario supporting equity markets.

Digital transformation Megatrends like Generative AI, the Fourth Industrial Revolution (digitizing and automating the manufacturing sector), and increasing vehicle electrification do not exist in a vacuum. They all need large infrastructure ecosystems to support their growth, and they can potentially change every organization’s operation. Generative AI is already putting tremendous strain on data and energy infrastructure. In a recent survey of 1500 IT leaders, 80% of them were worried that their business will be left behind if their infrastructure can’t support AI fast enough, and 81% believed that AI-generated data is likely to outgrow their organization’s current data centers. The AI data center energy demands and ultra-high speed connectivity requirements within data centers and eventually between data centers all create opportunities in addition to memory-hungry specialized AI servers populating these data centers.

We continue to focus on identifying and selecting the best companies to benefit from the expected economic environment and the trends driving global economic expansion. Companies with superior earnings growth are poised to outperform the market over the long term. Our Merlin equity portfolios focus on such companies of all sizes, selecting them based on their solid fundamentals and high long-term earnings growth expectations.

Despite the continued weakness in small and mid-cap stocks, their fundamentals and growth potential remain increasingly attractive. Companies that can maintain earnings growth in this challenging environment are well-positioned to benefit when the current uncertainties are resolved. We expect the declining interest rate environment to accelerate economic growth in the second half of the year and into 2025, creating an attractive environment for investing in growth equities of all sizes.

← Back to top