The global artificial intelligence sector is currently navigating a complex period of price recalibration that has sparked significant debate among market analysts. While bearish observers point to the collapse of AI token prices—which have largely returned to December 2025 levels following a peak in the spring of 2026—technological and economic indicators suggest this decline may be a symptom of rapid expansion rather than a signal of diminishing demand. This phenomenon is increasingly being identified as a modern application of the Jevons Paradox, where technological progress that increases the efficiency of a resource leads to an increase in the rate of consumption of that resource.
The Economic Mechanics of the AI Price Correction
The central argument for a bearish outlook on AI rests on the sharp decline in "retail" pricing for AI tokens. However, market data reveals a stark divergence between these token prices and the "wholesale" costs of the underlying infrastructure. While token prices have trended downward, the cost to rent H100 compute power rose by approximately 27% over a nine-month period, moving from $2.00 to $2.53 per hour. This indicates that while the end-product is becoming more affordable due to engineering efficiencies, the foundational capacity remains scarce and increasingly valuable.
This dynamic aligns with the 1865 observations of economist William Stanley Jevons, who noted that more efficient steam engines led to higher, not lower, coal consumption in Great Britain. In the current context, AI labs are passing efficiency gains to customers to stimulate usage. The result has been a monumental surge in volume. During the Google I/O 2026 conference, Alphabet CEO Sundar Pichai confirmed that AI token usage reached 3.2 quadrillion per month, a sevenfold increase within a single year.
Chronology of the 2025-2026 AI Market Cycle
To understand the current market position, it is necessary to trace the trajectory of the AI trade over the past 18 months:
- Late 2025: Initial surge in AI infrastructure spending as enterprises began integrating Large Language Models (LLMs) into core operations.
- Spring 2026: AI token prices and speculative AI-related assets reached a cyclical peak, driven by "hype-cycle" momentum.
- Mid-2026: Token prices began a significant "roundtrip" correction, returning to late-2025 levels. Simultaneously, demand for GPU-based compute capacity reached record highs, leading to a supply-demand imbalance in the data center market.
- Current Period (Late 2026): The market is entering a "sifting" phase where infrastructure providers maintain pricing power, while software and service providers begin to benefit from lower input costs.
Identifying the Indicator for a Market Peak
Analysts suggest that the most reliable gauge for the "top" of the AI infrastructure boom is not the price of tokens, but rather the profit margins of "neocloud" providers. Neoclouds are specialized data-center operators, such as CoreWeave Inc. (CRWV) and Nebius Group (NBIS), whose business models are predicated on the scarcity of high-performance computing power.
Recent financial disclosures from CoreWeave indicate that the market remains in an expansionary phase. CEO Michael Intrator recently noted that new customer contracts are carrying contribution margins 5 to 10 percentage points higher than those signed in previous quarters. This suggests that despite the broader price corrections in tokens, the providers of the physical hardware and compute environments still possess significant pricing power.
A reversal of this trend—specifically a contraction in neocloud margins—would serve as a primary signal that compute power has transitioned from a scarce resource to a commodity. Such a shift would likely mark a transition in the AI trade, favoring companies that utilize AI to expand internal margins over those that provide the raw processing power.
Strategic Portfolio Adjustments and the Nvidia Litmus Test
As the AI trade matures, investment strategies are shifting toward a more selective approach. Market experts, including Louis Navellier, Luke Lango, and Eric Fry, have recently overhauled their "AI Revolution Portfolio" to account for the changing landscape. This restructuring emphasizes identifying companies with "100X" potential in the current cycle while exiting positions that may be vulnerable to the next industry shakeout.
The immediate technical outlook for the sector hinges on the upcoming earnings report from Nvidia (NVDA). As the dominant supplier of AI chips, Nvidia’s results serve as a barometer for the entire ecosystem. A strong performance is expected to de-risk the technical landscape for the remainder of the year, potentially offsetting the seasonal volatility often seen in the October window.
Political Risks and Social Sentiment
While the economic fundamentals remain robust, a secondary risk factor is emerging in the form of political and social opposition to data center expansion. Although the current administration has provided active support for the domestic AI buildout, sentiment at the local level is souring.
A March 2026 Gallup survey revealed that 71% of Americans would oppose the construction of a data center in their immediate vicinity. This level of opposition is notably higher than the 53% who would oppose a nuclear power plant. Analysts suggest that this "Not In My Backyard" (NIMBY) sentiment could evolve into structural barriers.
The projected timeline for this risk to materialize fully is tied to the 2028 election cycle. A potential shift in federal policy, combined with state-level restrictions hardening from temporary pauses into permanent bans, could eventually bring the AI infrastructure supercycle to a close. However, for the 2026-2027 window, these risks are viewed as secondary to the immediate demand for compute power.
The Macroeconomic Backdrop: Inflation and the Fed
The AI sector’s performance is also inextricably linked to the broader macroeconomic environment, specifically the Federal Reserve’s handling of inflation. Markets are currently awaiting the latest Personal Consumption Expenditures (PCE) report, which will serve as a critical test for the Federal Open Market Committee (FOMC).
The Fed remains divided on the trajectory of interest rates. Fed Chair Kevin Warsh has signaled a preference for "trimmed-mean" PCE data, which filters out volatile price swings. Currently, this metric sits near 2.3%, suggesting inflation is nearing the target. However, several regional Fed presidents remain concerned about core inflation, which has hovered in the low 3% range for several years.
Key PCE Indicators to Watch:
- Headline vs. Core PCE: A discrepancy between a "hot" headline number and a "tame" core number would suggest inflationary pressure is concentrated in energy, rather than being broad-based.
- Trimmed-Mean PCE: If this figure jumps significantly above 2.3%, it increases the likelihood of a rate hike in the coming months.
- Jackson Hole Commentary: Chair Warsh’s upcoming speech at the Jackson Hole symposium will be scrutinized for clues on policy. Terms such as "favorable trend" would signal a dovish stance, while words like "vigilant" or "unfinished" would align with the hawkish members of the committee.
Implications for Investors and Industry Outlook
The current state of the AI market is defined by a transition from a speculative "gold rush" to a more disciplined industrial expansion. The fall in token prices, rather than being a bearish signal, represents a maturation of the technology that is driving unprecedented levels of consumption.
For investors, the primary implications are twofold. First, the infrastructure trade—led by chipmakers and neocloud operators—retains its momentum as long as pricing power remains high and margins continue to expand. Second, the "user" side of the AI trade is becoming increasingly attractive as the cost of AI inputs (tokens) declines, allowing for margin expansion in software and service-oriented firms.
While political headwinds and macroeconomic uncertainty regarding the Federal Reserve’s interest rate path remain persistent threats, the underlying demand for AI compute capacity shows no signs of abatement. The convergence of Jevons Paradox in technology and the scarcity of data center resources suggests that the AI bull market, while evolving, remains fundamentally intact through the current earnings season and into the 2027 fiscal year.
