February was a month of fear: the fear of displacement.
There were rolling selloffs that crushed baskets of stocks in similar industries on a weekly basis, depending on which type of plugin Anthropic released.
This marks a clear change in how investors are underwriting AI exposure. For much of the last two years, simply being adjacent to AI was enough to expand multiples because the market was primarily rewarding ambition, participation, and headline capex.
Investors are now separating the ecosystem into three buckets: the companies enabling AI buildouts, the companies durably monetizing AI, and the companies whose economics may be pressured by AI-driven disruption.
Each technological wave of enhanced productivity breeds creative destruction. Business models and unit economics have to be reconsidered. This leads me to think that some of these moves are overdone in the short term, but underdone in the long term.
The market is spending more time on cash conversion, payback periods, and evidence that incremental investment is translating into incremental returns. The underlying posture is still constructive on AI as a secular theme, but the bar has moved toward proving that AI will improve your company’s unit economics.
This is especially true for the largest spenders, so let’s start there.
Megacap Return on AI Meets the Capex Tax
The hyperscalers continue to spend at an extraordinary scale, but the tolerance for open-ended spending has narrowed. Investors are less interested in broad statements about strategic necessity and more focused on a simple causal chain: what’s being built, how quickly it becomes productive capacity, and what revenue or efficiency gains it generates.
This is the largest mobilization of capital in the history of mankind. Evercore ISI reports that hyperscalers are collectively forecasting 2026 capex of at least about $630B, with much of that directed toward data centres and AI processors. Expand the universe beyond hyperscalers, and we get north of $800B.
What’s just as eye-popping as the magnitude is the rate of change in forecasts. In less than a year, growth rates have gone from +37% to +85% for 2025 and +9% to +73% in 2026 (from a higher base). Even for companies with elite balance sheets, an investment curve that behaves exponentially while monetization grows more linearly creates a tension that shows up quickly in valuation debates.
Funding mix has also become part of the conversation again. As capex rises, markets are paying closer attention to whether the buildout remains overwhelmingly funded by free cash flow or whether external financing becomes a meaningful marginal contributor. When the story shifts toward capital markets helping fund AI expansion, investors start applying a different sensitivity to duration and downside, because “self-funded” carries a different risk profile than “partially financed.”
Cash flow optics add to the pressure. Data centre capex is an immediate cash outlay, while depreciation spreads the cost across the income statement over time. As a result, earnings can look resilient even when free cash flow compresses. That gap has become more important because the market is trying to distinguish between companies whose AI investments are self-reinforcing and companies whose investments behave like a rising tax on near-term cash generation.
Across Big Tech, the direction of travel matters as much as the absolute number. Even if not every company prints negative free cash flow, the market is increasingly modelling a world where cash conversion deteriorates at the peak of the buildout, and where return on AI (ROAI) debates become a persistent valuation headwind until payoffs are more visible.
NVIDIA: Strong Results, More Skepticism About the Next Step
The shift in sentiment showed up most clearly through the reaction to NVIDIA’s quarterly earnings announcement. On reported fundamentals, the quarter was exceptional: Q4 FY26 revenue of $68.1B, up 20% quarter over quarter and 73% year over year; data centre revenue of $62.3B, up 22% quarter over quarter and 75% year over year; GAAP gross margin at 75.0%; and Q1 FY27 guidance around $78.0B plus or minus 2%. NVIDIA also noted that its outlook did not assume data centre compute revenue from China, which underlines the strength of the guide even with that constraint.
Despite this, the stock response was subdued. That’s rarely about the quarter itself; it typically reflects investor uncertainty about what the next phase looks like. In this case, the question is centred on the durability of the spending that drives NVIDIA’s near-term numbers. The market still believes in AI demand, but it’s increasingly debating whether buildout intensity can keep compounding at the same rate without creating pressure somewhere else in the system, whether through hyperscaler margins, cash flows, or a willingness to keep pushing capex higher.
In other words, NVIDIA’s results can remain strong while investors simultaneously discount a scenario where the marginal dollar of data centre spend becomes harder to justify and where procurement cycles become more variable. The market is trying to determine whether the industry is approaching a phase where optimization and utilization matter more than raw expansion, which would not end the AI cycle but could change the cadence of orders and the slope of capex.
Rotating Bottlenecks: Where Pricing Power Is Showing Up Now
As ROAI debates intensify at the platform level, investors have leaned harder into segments where earnings growth and pricing power are visible today. The market has effectively turned into a rotating bottleneck trade: constraints moved from GPUs to power, then to cooling and networking, and more recently toward memory, storage, and optics.
Memory has been the most obvious beneficiary of the rotation. A global supply crunch has been amplified by AI infrastructure demand, with manufacturers diverting capacity toward high-bandwidth memory (HBM) used in AI servers, tightening supply in other segments. Price moves have been dramatic in parts of the market, including areas that have more than doubled since early 2025. Equity performance has followed the pricing signal, with investor attention shifting heavily toward memory suppliers and South Korea standing out as a market leadership pocket, including strong year-to-date moves in Samsung and South Korean semiconductor firm SK Hynix during the rotation window.
Storage has followed a similar pattern, driven by the data intensity of training, fine-tuning, and inference at scale, and further amplified by agentic workflows that generate and retrieve large volumes of structured and unstructured data. SanDisk became a poster child for the theme after a sharp move higher, accompanied by an outlook that pointed to AI-driven storage demand as a meaningful contributor to revenue expectations.
Optics represents the “pipes” of the buildout, and the market has started paying up for bandwidth-enabling components again as clusters scale. When investors are less willing to award broad AI premiums to platforms still deep in the investment phase, they often gravitate to suppliers that are already receiving checks and demonstrating tangible pricing power. This is why the bottleneck rotation matters. It’s both a bet on scarcity and a way to express AI exposure while sidestepping the ROAI debate at the platform level.
Private Equity Stress and Its Spillover Into Public Tech
February also highlighted a second-order risk that has been underappreciated: private equity stress can transmit into public technology markets through funding conditions, exit dynamics, and multiple repricing.
Private equity accumulated significant exposure to software for familiar reasons: recurring revenue, high gross margins, and operating leverage. Many deals were executed at peak multiples, often above 20x EV to EBITDA in 2021, and structured with heavy leverage, reaching around 10x EBITDA in the most aggressive period. Even after the peak, leverage remained elevated for new deals, averaging roughly 7x EBITDA from 2022 to 2025.
Layer AI disruption on top, and the math becomes uncomfortable. If AI compresses moats and growth durability, multiples compress. With leverage, a modest multiple reset can translate into a severe equity impairment. That is why credit risk scenarios increasingly include AI disruption as a driver of higher defaults, including estimates that default rates in private credit could rise meaningfully under rapid disruption conditions.
The transmission channel is liquidity, especially through semi-liquid structures and the secondaries market. When redemptions rise in vehicles that promise periodic liquidity, managers may be forced to gate or sell assets into weaker bids, which can push marks lower and reinforce risk aversion. In parallel, buyers in the secondaries market are demanding larger discounts, in some cases widening toward 20% versus mid-single digits weeks earlier. This signals a deterioration in the marginal clearing price for tech-heavy private portfolios.
Public markets feel this through multiple compression, tighter financing conditions for growth companies, fewer healthy exits, and a weaker bid for software M&A. Many investors have treated private markets as a separate ecosystem, but the cycle is shared, and February made that linkage harder to ignore.
Looking at the SaaS Model Transition Through Salesforce
The “software is dead” debate has been happening for two years, but the trade happened in two months. The iShares Expanded Tech-Software Sector ETF (IGV), which tracks an index of North American software equities, was down -9.7% in February, following -14.5% in Jan, with the -23% two-month return being the worst since the global financial crisis.
In hindsight, there was a clear catalyst. Breakthrough advancements with Opus 4.5 and GPT 5.3 codex marked a clear delineation where agents could now build, test, and fix entire end-to-end applications. Before, it was sort of happening in the background. In December 2025, it became a reality. As the marginal cost of software development declines rapidly, the entire space is embroiled in Twitter essay wars about the terminal value of software.
The valuation debate is about business model evolution rather than quarterly earnings. Salesforce has begun emphasizing agentic metrics, including 2.4 billion agentic work units delivered and 19 trillion tokens processed all-time, which signals a shift toward usage and consumption framing. That change is rational, but it introduces a difficult transition for classic SaaS economics.
Seat-based SaaS monetization was built around a simple engine: sell users, expand seats, and compound ARR. Agentic AI changes the unit of work. If agents complete tasks that previously required human users, customers may need fewer seats even as total work expands. Salesforce, therefore, needs to grow a new monetization unit that can scale, while managing a migration that protects customer relationships and keeps margins within an acceptable range as AI inference costs become a real component of the cost of goods sold (COGS).
Salesforce needs to cannibalize its own business to do this. I highly doubt that it will.
Markets tend to punish these transitions early because budgeting behaviour shifts from licenses to outcomes, pricing becomes more experimental, and margins can compress before cost structures normalize. That dynamic helps explain why software has tanked, and why investors have been quicker to reward picks-and-shovels exposures during periods when platform ROAI remains uncertain.
The Google Question: Outspend or Out-Execute
Google remains one of the few companies capable of sustained, subsidized competition across models, infrastructure, and distribution, supported by Search, YouTube, Android, Workspace, and internal silicon. In principle, that enables aggressive spending and a long-duration strategy.
In practice, two constraints still apply. First, capital is no longer free in public markets, and the market is actively pricing ROAI uncertainty into the largest spenders. Second, the industry’s endpoint matters more than the arms race, because winning by spending is less attractive if the prize is extended cash flow compression with unclear payback. Will Google play the role of stock market appeaser, or play the long game and ALSO disrupt itself?
The Unifying Lens: LLM Economics and the Path to Equilibrium
The most useful way to connect these themes is the underlying economics of frontier AI. Model-level unit economics can be attractive, with high gross margins possible on an individual generation once trained. At the same time, company-level profitability can remain pressured because the industry is reinvesting into increasingly expensive next-generation models and the infrastructure to serve them.
This framework maps onto the broader market. Megacaps can see free cash flow compress because they are funding successive buildout cycles. NVIDIA can print extraordinary results while investors debate whether the next capex cycle remains as steep. Bottleneck suppliers can outperform because they monetize scarcity in the current phase. Private equity stress accelerates because leverage punishes duration and multiple compression. SaaS businesses face a software analogue, where they can deliver value while absorbing transition costs and cannibalization pressures.
The equilibrium that matters is the point where training scale-up moderates, inference costs fall enough to expand use cases broadly, and monetization grows faster than the compute bill. The market is still optimistic that this equilibrium exists, but it’s becoming more selective about which companies are positioned to reach it with attractive shareholder outcomes.
February’s message was consistent throughout the stack: AI remains a dominant theme, but investors want clearer answers on who captures economic value, how quickly they capture it, and what the cumulative bill looks like until returns become visible.
Strong Convictions. Loosely Held.
— Nick 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.
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.