Q1 2026 was the quarter the market stopped paying for stories and started paying for constraints. Gold opened the period as the preferred refuge, oil took over the macro tape by March, and semiconductors lost their monolithic status and split into winners and laggards. Software saw investors draw a much harder line between systems of record and rented labor, while private credit discovered that yesterday’s “stable recurring revenue” can trade very differently once technological duration risk enters the room. Underneath all of that was a more important message: the AI trade remains physical (for now), and the market is still pricing it that way.
The quarter’s first big rotation came through macro commodities. Gold spent the early part of the quarter acting as a classic scarcity asset, then suffered a 12% decline in March, its worst monthly drop since 2008, as higher oil, hotter inflation expectations, and the prospect of tighter financial conditions crowded out the simpler safe-haven trade. Oil moved in the opposite direction with far more force. Brent rose 64% in March, U.S. crude rose 52% over the same period, oil benchmarks were up roughly 60% from the late-February conflict flare-up, and European LNG surged 80%. When molecules and electrons begin to reprice that quickly, the market’s center of gravity shifts away from long-duration stories and toward the companies that own throughput, logistics, fuel, power, and hard capacity. The market has slapped a fun new acronym on these stocks: HALO (Heavy Assets, Low Obsolescence). We saw it in Gold and Oil, but also in the physical buildout of AI. Stay with me as I walk through this...
Technology did not escape rotation, although the internal dispersion was far more interesting than the index-level drawdown. While the NASDAQ finished Q1 down 7%, the Magnificent Seven fell 13%, Software (IGV) fell 24%, but Semiconductors (SOX) finished UP 7%.
As the market de-risked some of the most crowded AI exposures, it began to rotate toward the parts of the stack that touch the physical buildout more directly. But it did so in a VERY discriminating way. Want a fun exercise? Take a look at some of these stocks over the quarter: Sandisk (semi storage, +167% in Q1), Applied Optoelectronics (semi optics, +142%), Lumentum (semi optics, +91%), Micron (semi memory, +18%), Nvidia (semi GPU, -7%), Microsoft (Software, -23%), Salesforce (Software, -29%). I mean the bifurcation is crazy...
So what is this telling us? It looks like the market is in "show-me" mode with the MAG7 as free cash flow and net cash positions move towards zero to fund the greatest mobilization of capital in human history. Concern here is return on AI spending. With software, the market is giving the entire category a big thumbs down. Concern here is that as the marginal cost to develop software moves to zero, incumbents will fail to disrupt themselves, growth falls off a cliff as pricing power erodes when contracts come up for renewal and terminal value goes to zero. Semiconductors are partying like it’s 1999 as many analysts say that this is now a structural buildout era, but historians try to call the cops on the techno-optimists by reminding them that capex cycles are just that...cycles.
Private credit deserves a place in this recap because it may be the cleanest window into how the market is repricing legacy software cash flows. For those keeping score, the alternative asset manager ETF GPZ was down 21% in Q1. Software and services represent about 15% of U.S. CLO collateral, software alone is roughly 12% of CLO holdings, and direct lending exposure to software sits around 19% in some portfolios. I think the numbers are way higher than this because software touches most of these companies and they could very well be mislabelled, but I digress...
The software discussion is usually framed as an equity multiple issue, although the more consequential story may be that a large part of the private credit complex was underwritten to a world where seat-based software cash flows looked close to bond-like. That world is changing. As spreads widened and investors rethought the durability of levered software borrowers, BlackRock’s HLEND fund received $1.2B of withdrawal requests in Q1 and paid out $620M, while Blue Owl restricted withdrawals after a record $5.4B of redemption requests. At the same time, you cannot avoid a headline about one of the big PE guys buying out some random data center company. Amongst the private equity/credit guys, it looks like the software book is going the opposite direction of the data center book. Say it with me again now…HALO.
That is why I think the main idea for this quarter is continued investment in the physical economy. The AI buildout has expanded into a power story, a memory story, a networking story, an optical story, a cooling story, and increasingly a regional infrastructure story. Big Tech is expected to spend around $630B to $635B on AI infrastructure in 2026, up from about $383B the year before and roughly $80B in 2019. The largest AI sites are now being designed around more than 1 GW of continuous load, enough to power as many as 850,000 homes, and 46 U.S. data centers are already planning their own power plants with a combined 56 GW of capacity. When an industry starts building like a utility, valuations tend to migrate closer to infrastructure economics than to classic software multiples.
This cycle is still accelerating even after a volatile quarter. During Nvidia’s GTC, the “GPU-Jesus in a leather jacket” (Jensen Huang) told us once again that the industry’s economic center of gravity is shifting from training to inference, with enterprise agentic workloads serving as the main demand driver. Inference computing demand has increased roughly one million times in two years, and the monetization model is becoming much more explicit, with basic inference priced near $3 per million tokens while premium reasoning workflows can command around $150 per million. Read that again and try telling me that LLMs are a commoditized business...
How’s this for ROAI…fewer than 5% of the world’s roughly 12,000 data centers can house next-generation systems such as Blackwell, and we now have preliminary evidence that shows how revenue per gigawatt for the hyperscaler can rise from roughly $30B on Hopper H200 architecture to about $150B on Blackwell NVL72 and toward $300B on Vera Rubin. That is an enormous statement about how much value the market can assign to physical throughput once intelligence becomes metered, persistent, and low latency.
GPUs still sit at the first tollbooth, although the market is becoming more aware that the next constraints sit adjacent to the accelerator rather than inside it. Global semiconductor sales are expected to reach $1T in 2026 after climbing 25.6% to $791.7B in 2025, advanced computing sales grew 39.9% to $301.9B, and memory sales rose 34.8% to $223.1B. Micron is a memory company. They just guided to $33.5B of third-quarter revenue, planning more than $25B of capex this fiscal year, and are already talking about even higher spending in 2027. Samsung is describing the current environment as an unprecedented supercycle with shortages spreading across memory categories. Memory is now estimated to consume 30% of hyperscaler AI data center Capex. Let that sink in. 30% of spend doesn’t go towards a company called Nvidia. The market is expanding.
The bottleneck is rotating toward memory bandwidth, advanced packaging, and wafer intensity at a pace the market is still digesting. Building a frontier AI cluster puts pressure on chip fabs, memory fabs, advanced packaging lines, and the lithography equipment that sits behind all of them. Micron's latest HBM4 is a meaningful generational step in both speed and efficiency, and the machines required to produce it are among the hardest to come by in the entire industry. Bet on the companies that build the machines that make the chips.
CPUs and optical interconnects are next in line, which is a natural consequence of inference becoming more sequential, more agentic, and more distributed. When AI systems move from one-shot prompts toward planning, retrieval, tool use, and API coordination, the orchestration layer becomes just as important as raw tensor throughput. That is why Nvidia’s GTC roadmap put the CPU back at center stage, with concurrent AI sandboxes designed to keep expensive GPU fleets fully utilized. It is also why optics are rapidly moving from background component to core architecture. The whole thing here is the idea of scale-up, scale-out, and scale-across. Nvidia has committed $2B each to Coherent, Lumentum, and Marvell to secure photonics capacity and broaden the ecosystem around optical interconnects, while co-packaged optics can deliver 5x better power efficiency and 10x greater network resiliency as copper becomes a structural constraint. The market is slowly learning that moving data can be every bit as valuable as processing it.
Software fits into this discussion in a more nuanced way than the market has allowed over the last few months. Traditional software continues to face real pressure in categories where revenue scales with seat count and the product’s implicit promise is human labor at a fixed monthly price. A nearly $1T rout in software stocks after a new wave of agent releases makes that clear enough, and investors are finally asking which vendors own something deeper than interface polish and a billing relationship. That pressure is real, especially in workflows where AI can collapse task time, reduce headcount, and shrink the very denominator on which old SaaS models were built. At the same time, software still owns system of record, identity, governance, permissions, and proprietary data, which means the path forward remains very open for platforms that can shift toward consumption, outcomes, and orchestration. The future of software looks increasingly like a transition in pricing and architecture rather than a disappearance of the category.
The more exciting opportunity, in my view, belongs to the AI-native startup layer that is reimagining the role of software altogether. OpenAI exited February at a $25B annualized revenue run rate. Anthropic is running around $19B, Claude Code alone is above $2.5B, and business subscriptions to that product have quadrupled since the start of the year. What these companies are selling extends well beyond a cheaper version of old SaaS. They are building software priced around usage, outcomes, and work completed, which expands revenue per account while aligning far more tightly with delivered value. Once pricing model becomes business model, AI-native companies have a chance to capture much more of the work itself rather than simply charging rent on a seat.
So when I look back on Q1, I see a market becoming more selective and more physical in the way it expresses the technology theme. Could this flip when real AI-native software and application companies come public? Absolutely. But are there any AI software plays in the public market right now (besides Palantir)? No. If AI-native software startups like Harvey, Sierra, or OpenEvidence were public right now, I bet you they wouldn't be selling off with the rest of software.
On the physical front, Gold and oil reminded investors that scarcity still shapes the macro tape. Semiconductors reminded us that compute still matters, although the value is rotating inside the stack toward memory, CPUs, optics, and power. Private credit reminded us that old software cash flows carry much more technological duration risk than many lenders assumed. The AI-native application layer showed that when software starts selling outcomes instead of seats, growth can accelerate at a pace the old models rarely matched. Further out, the same stack is starting to extend into robotics and autonomy, which is where the physical economy of intelligence begins to show up in factories, warehouses, and fleets. But that’s a ramble for another time…
Continued investment in the physical economy sits at the center of all of this. The future of intelligence will be decided by whoever can move the most data, with the lowest latency, across the most constrained infrastructure, and then wrap it in products that solve real work. That is the path Q1 put in front of us, and I think the market is only beginning to appreciate how long that runway can be.
Strong Convictions. Loosely Held.
– Nicholas Mersch, CFA
The content of this document is for informational purposes only, and is not being provided in the context of an offering of any securities described herein, nor is it a recommendation or solicitation to buy, hold or sell any security. The information is not investment advice, nor is it tailored to the needs or circumstances of any investor. Information contained in this document is not, and under no circumstances is it to be construed as, an offering memorandum, prospectus, advertisement or public offering of securities. No securities commission or similar regulatory authority has reviewed this document and any representation to the contrary is an offence. Information contained in this document is believed to be accurate and reliable, however, we cannot guarantee that it is complete or current at all times. The information provided is subject to change without notice.
Commissions, trailing commissions, management fees and expenses all may be associated with investment funds. Please read the prospectus before investing. If the securities are purchased or sold on a stock exchange, you may pay more or receive less than the current net asset value. Investment funds are not guaranteed, their values change frequently and past performance may not be repeated. Certain statements in this document are forward-looking. Crypto assets can be extremely volatile and there is no guarantee that the amount invested will be returned to you.
Forward-looking statements (“FLS”) are statements that are predictive in nature, depend on or refer to future events or conditions, or that include words such as “may,” “will,” “should,” “could,” “expect,” “anticipate,” “intend,” “plan,” “believe,” “estimate” or other similar expressions. Statements that look forward in time or include anything other than historical information are subject to risks and uncertainties, and actual results, actions or events could differ materially from those set forth in the FLS. FLS are not guarantees of future performance and are by their nature based on numerous assumptions. Although the FLS contained in this document are based upon what Purpose Investments and the portfolio manager believe to be reasonable assumptions, Purpose Investments and the portfolio manager cannot assure that actual results will be consistent with these FLS. The reader is cautioned to consider the FLS carefully and not to place undue reliance on the FLS. Unless required by applicable law, it is not undertaken, and specifically disclaimed, that there is any intention or obligation to update or revise FLS, whether as a result of new information, future events or otherwise.