LWAYS OPEN
The central question facing technology investors is changing. The issue is no longer simply whether artificial intelligence will generate enormous economic value; the evidence increasingly suggests that it will. The more difficult question is what investors should pay today for profits expected from that transformation years into the future. Recent weakness across the Nasdaq, semiconductors and selected Magnificent Seven stocks indicates that a valuation reset has begun. FRL’s research conclusion, however, is more nuanced: the reset is real, but the evidence does not yet establish that it is complete. Further downside remains plausible without requiring the structural AI investment thesis to fail.
Research Thesis
FRL’s central research question is: Has the market already reset AI and mega-cap technology equities to sustainable valuation levels, or does the combination of elevated expectations, enormous capital requirements and higher long-term interest rates require another stage of repricing?
Our thesis is that the market has moved from the expansion phase of the AI valuation cycle into a more discriminating phase in which earnings delivery, free cash flow and return on invested capital increasingly matter more than simply announcing larger AI expenditure.
This represents a structural change in market behaviour.
During the earlier phase of the AI cycle, increasing capital expenditure was frequently interpreted as evidence of future dominance. Today, the same announcement can produce the opposite reaction if investors conclude that expenditure is growing faster than monetisation.
That distinction explains why a technology company can report strong revenue growth and simultaneously experience multiple compression.
The current adjustment should therefore be understood primarily as a reset of the price investors are prepared to pay for AI growth, rather than evidence that AI growth itself has disappeared.
Supporting evidence remains substantial. Cloud demand is strong, compute capacity remains constrained in important parts of the ecosystem, semiconductor demand remains historically high and the major hyperscalers continue investing aggressively.
Contrary evidence is equally important: valuations remain demanding in parts of the sector, long-duration Treasury yields have risen, AI capital requirements have become enormous, institutional positioning is increasingly divided and investors are beginning to question the eventual return on hundreds of billions of dollars of infrastructure expenditure.
The reset has started. Whether it has finished is a different question.
From AI Expansion to Valuation Discipline
The first stage of the generative AI investment cycle was dominated by scarcity.
There was insufficient advanced compute capacity, insufficient high-bandwidth memory, insufficient data-centre infrastructure and extraordinary demand for the hardware required to train and deploy increasingly sophisticated models.
Scarcity created pricing power.
NVIDIA (NVDA) became the clearest expression of that dynamic, but the economic effects spread through networking, memory, semiconductor equipment, foundries, power infrastructure and cloud computing.
Markets rationally increased earnings expectations.
The Federal Reserve itself noted in its June meeting minutes that higher earnings expectations, particularly in technology, accounted for a large part of the equity-market advance during the period it reviewed.
But equity valuation contains two variables: expected cash flows and the rate used to discount those cash flows.
For much of the AI rally, rapidly increasing earnings expectations overwhelmed the negative effect of relatively high interest rates. That equilibrium has become less comfortable.
Long-term Treasury yields have moved materially higher, increasing the discount rate applied to future earnings. The effect is particularly significant for businesses whose valuations incorporate exceptional growth many years ahead.
At the same time, the market’s expectations have risen.
A company growing from modest expectations can create enormous shareholder value by surprising positively. A company already priced for exceptional execution has a different problem: exceptional performance may merely satisfy the valuation rather than expand it.
This is one reason the AI trade can remain fundamentally successful while individual AI equities trade lower.
The Magnificent Seven Are No Longer One Trade
The following Magnificent Seven chart illustrates a market that has already entered a meaningful repricing phase without yet confirming a structural breakdown. After advancing from approximately 2,050 in March to almost 2,700 by late spring, the index has shifted into a broad and increasingly volatile consolidation range. Recent trading around 2,515 leaves the group roughly 6–7% below its peak, while repeated failures to establish new highs suggest that the earlier phase of almost uninterrupted multiple expansion has weakened. Importantly, however, the longer-term trend structure remains intact around the 2,450 area. A sustained break below that region would materially strengthen the case for a deeper valuation reset, whereas a recovery through approximately 2,580–2,620 would suggest that much of the current repricing has already been absorbed.
The expression “Magnificent Seven” increasingly obscures more than it explains.
Apple (AAPL), Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), Meta Platforms (META), NVIDIA (NVDA) and Tesla (TSLA) have different business models, capital requirements, competitive positions and exposures to artificial intelligence.
Their common characteristic is scale, not economic uniformity.
The market is beginning to recognise that distinction.
Institutional positioning provides useful evidence. An analysis of thousands of second-quarter regulatory filings showed an almost evenly divided picture: roughly 44% of reporting investment managers reduced Magnificent Seven exposure while approximately 42% increased it.
That is not evidence of institutional abandonment.
It is evidence of declining consensus.
This matters because one of the most powerful forces behind the previous advance was concentration. Passive investment flows, benchmark pressure, momentum strategies and institutional fear of underperformance repeatedly channelled capital toward the same relatively small group of dominant companies.
Once the group stops moving as a single factor, security selection becomes more important.
Microsoft and Amazon can increasingly be assessed through cloud monetisation and infrastructure economics. Alphabet combines AI investment with exceptionally profitable search and cloud businesses. Meta’s investment case depends partly on whether enormous infrastructure spending strengthens advertising economics and creates additional AI revenue streams. NVIDIA remains primarily exposed to the scale and duration of accelerated-computing demand.
Tesla represents another set of assumptions entirely.
A reset therefore does not require all seven companies to fall together. It can occur through dispersion: some companies continue growing into their valuations while others experience price declines or prolonged periods of sideways trading.
Time itself can reset valuation.
The AI Capex Paradox
Perhaps the most important structural development is the changing interpretation of AI capital expenditure.
Hyperscalers are committing extraordinary amounts of capital to data centres, GPUs, networking, power and related infrastructure.
Initially, larger capex forecasts were generally bullish for the entire AI value chain. More spending implied more semiconductor demand and greater confidence in the ultimate AI opportunity.
That relationship has become more complicated.
Investors now have to ask three questions.
How much capital is being deployed?
How quickly does that capital generate incremental revenue?
And what return will ultimately be earned on it?
Alphabet provides an illustration of the underlying economic strength. Google Cloud has delivered powerful growth, with AI products contributing significantly. This is genuine monetisation rather than theoretical future demand.
Yet the broader market reaction to increasing hyperscaler spending demonstrates that investors are no longer willing to treat every additional dollar of capex as automatically value-creating.
This produces what FRL considers the AI capex paradox.
For NVIDIA and the semiconductor supply chain, increasing hyperscaler capex can be revenue.
For the hyperscaler spending the money, the same capex is an investment that must eventually earn an adequate return.
Those two perspectives can diverge.
The semiconductor supplier can report exceptional demand while the customer’s free cash flow comes under pressure. Both outcomes can occur simultaneously.
This is precisely why the next phase of the AI market is likely to be more selective than the previous one.
Semiconductors Are the High-Beta Test of the Reset
The Philadelphia Semiconductor Index’s approximately 5% decline on August 18 is analytically important because semiconductors have been the highest-conviction expression of the AI infrastructure cycle.
The decline does not establish that semiconductor fundamentals have deteriorated by an equivalent amount.
Instead, it illustrates how rapidly valuation can adjust when several risk factors converge: rising Treasury yields, geopolitical uncertainty, high expectations and crowded positioning.
That distinction matters.
If semiconductor earnings estimates begin falling materially alongside prices, the market would be signalling a fundamental deterioration in the AI investment cycle.
If prices fall while earnings expectations remain comparatively resilient, the process is more consistent with multiple compression.
The latter can actually strengthen the long-term investment structure by reducing the amount of future perfection already embedded in current prices.
But it does not tell us where the reset ends.
Markets rarely move directly from expensive to correctly valued. Momentum, leverage, systematic positioning and investor psychology can carry prices below levels justified by a purely fundamental calculation.
For that reason, another leg lower in AI and semiconductor equities remains entirely compatible with a constructive three-to-five-year AI infrastructure thesis.
The Bond Market Has Changed the Valuation Equation
The technology reset cannot be analysed independently from the bond market.
The 30-year U.S. Treasury yield recently moved above 5%, while the 10-year yield approached levels around 4.7%.
Those numbers represent an increasingly competitive alternative to equities.
More importantly, they alter valuation mathematics.
Consider a simplified growth company whose investment case depends heavily on cash flows expected five or ten years from now. Increasing the discount rate applied to those cash flows reduces their present value even if the underlying business forecast remains unchanged.
This is why technology equities are frequently described as long-duration assets.
The problem becomes particularly acute when high yields coincide with already elevated valuation multiples.
For much of the AI cycle, accelerating earnings estimates offset this mathematical headwind. The current market is testing whether that remains possible.
If long-term Treasury yields stabilise or decline, technology valuations receive an important source of relief without requiring any improvement in corporate fundamentals.
If yields move materially higher again, the opposite applies.
This is therefore one of the strongest arguments for believing that the reset may not yet be complete.
How Much of the Reset Has Already Happened?
There are three different forms of valuation reset, and investors should distinguish between them.
The first is price correction. Shares decline while earnings expectations remain relatively stable. Valuation multiples consequently contract.
The second is earnings catch-up. Share prices move sideways while corporate profits continue growing, gradually reducing valuation multiples without requiring a major market decline.
The third is fundamental reset. Earnings expectations themselves decline, forcing investors to reassess both the numerator and denominator of the valuation equation.
Current evidence is more consistent with a combination of the first two than the third.
That is significant.
A price-led or time-led correction can restore healthier valuation conditions without ending a secular bull market. A fundamental reset would be substantially more serious.
The Nasdaq 100 chart accompanying this study should therefore not be interpreted simply as a trading signal. Its importance is that it provides a visual representation of how much price adjustment has occurred relative to the extraordinary advance that preceded it.
FRL would not define the current market as fully reset merely because several major technology stocks have declined from their highs.
The more demanding test is whether prices, earnings expectations and interest rates have reached a new equilibrium.
There is insufficient evidence yet to conclude that they have.
Near Term, Medium Term and Structural Horizon
Near Term: 0–12 Months
The near-term environment remains vulnerable to further repricing.
High long-term Treasury yields, elevated geopolitical risk, energy-driven inflation pressure and demanding AI expectations create a combination in which disappointing guidance can produce disproportionate price reactions.
A further decline would therefore not be surprising.
Importantly, such a decline would not automatically constitute evidence that AI fundamentals had broken. Another 5%, 10% or even larger movement in individual high-beta technology stocks can occur through valuation compression alone. FRL is deliberately not assigning a forecast percentage decline because the required adjustment depends on future yields, earnings estimates and market positioning rather than an arbitrary technical target.
Medium Term: 1–3 Years
The market is likely to become increasingly selective.
The key distinction will shift from companies that spend on AI to companies that generate measurable economic returns from AI.
Cloud revenue, productivity improvements, advertising monetisation, enterprise adoption, inference economics and free-cash-flow conversion should progressively replace headline capex as the relevant metrics.
Companies possessing distribution, proprietary data, infrastructure scale and strong balance sheets should have structural advantages.
Structural Horizon: 3–5+ Years
The central AI thesis remains intact unless evidence emerges that the technology fails to generate sufficient economic productivity or commercial returns to justify the infrastructure constructed around it.
The current correction does not provide that evidence.
Indeed, history suggests that transformative technologies and severe equity corrections are entirely compatible. The internet transformed the global economy despite the collapse of numerous internet-era valuations.
Technology can be revolutionary while a stock is overpriced.
That distinction is fundamental to the present research question.
The Counterargument
The strongest argument against expecting materially lower levels is straightforward: earnings may grow into today’s valuations faster than the market expects.
AI infrastructure demand remains strong. Cloud businesses continue expanding. Capacity constraints persist. NVIDIA and other critical suppliers continue benefiting from substantial compute demand. The hyperscalers possess some of the strongest balance sheets and cash-generating businesses in global corporate history.
If AI monetisation accelerates while Treasury yields stabilise, the valuation reset could occur primarily through earnings growth and time rather than substantially lower prices.
There is also a positioning argument.
The market has already experienced meaningful de-risking. Institutional ownership data show that many investors have reduced exposure. Semiconductor equities have experienced sharp individual correction days, and concerns about AI capex are no longer obscure risks. They are widely discussed.
Markets typically discount known risks before those risks become obvious in reported financial statements.
It is therefore possible that a substantial portion of the reset is already occurring beneath the index surface.
That possibility prevents FRL from interpreting current weakness as evidence of an inevitable major technology bear market.
Risks to the Research Thesis
The principal risk to our thesis is that we underestimate the speed of AI monetisation.
If enterprise adoption, inference demand, advertising improvements, software productivity and cloud utilisation accelerate sufficiently, earnings estimates could rise faster than valuation multiples contract. In that environment, today’s apparent overvaluation could disappear through fundamental growth.
A second risk is interest rates. A material decline in long-term Treasury yields would increase the present value of future technology earnings and could quickly restore investor willingness to pay premium multiples.
The opposite risk is more severe.
If Treasury yields continue higher while AI capex increases and free-cash-flow conversion deteriorates, the valuation reset could become substantially deeper.
The key risk is therefore not simply slower AI growth. It is a simultaneous deterioration in the two variables supporting current valuations: earnings expectations and the discount rate.
The thesis breaker would be clear evidence that the major AI infrastructure investments are failing economically: material hyperscaler capex reductions caused by weak demand, sustained deterioration in cloud AI growth, significant excess compute capacity, falling utilisation and broad downward revisions to semiconductor earnings expectations.
That would transform the present interpretation from valuation reset to fundamental cycle deterioration.
We do not currently see sufficient evidence for that conclusion.
FRL Research Outlook
FRL’s assessment is that a genuine reset is underway across AI, semiconductor and mega-cap technology equities, but it would be premature to declare that reset complete.
The distinction is important.
We are not observing compelling evidence that the structural AI investment cycle has ended. Corporate expenditure, cloud demand and compute requirements continue to support the long-term thesis.
What has changed is the market’s willingness to capitalise those future opportunities at almost any valuation.
That is healthy from a long-term market perspective, even though it can be painful during the adjustment.
Our base interpretation is therefore neither “AI bubble collapse” nor “routine dip that must immediately reverse.”
It is a transition from narrative-driven multiple expansion toward return-driven valuation discipline.
In the near term, the balance of evidence leaves room for lower trading levels, particularly if Treasury yields remain elevated or earnings expectations stop rising. The semiconductor complex remains especially sensitive because it combines strong structural growth with high expectations and substantial positioning.
Over the medium term, however, a reset can create a stronger foundation. Earnings growth can catch up with prices. Weak business models can separate from durable ones. Capital can migrate toward companies demonstrating genuine returns on AI expenditure rather than merely participation in the theme.
The most important signal will therefore not be whether the Nasdaq 100 falls another few percentage points.
It will be whether earnings expectations fall with it.
If prices decline while AI revenue, cloud demand and semiconductor earnings remain structurally strong, the evidence would increasingly support the interpretation of a valuation reset within an intact secular investment cycle.
If both prices and earnings expectations deteriorate together, the research conclusion must change.
For now, FRL’s evidence supports patience and valuation discipline. The reset is real. It has progressed. But the market has not yet provided enough evidence to conclude that the new equilibrium has been established.
Research Notice
This publication represents general financial-market and investment research prepared by Final Resurrection Ltd. for research and informational purposes. It does not constitute individual investment, financial or portfolio advice, a recommendation to buy or sell any security, or a guarantee of future investment performance.