When explosive AI growth meets systemic supply bottlenecks, Morgan Stanley suggests that current market optimism regarding this AI revolution may still be severely underestimating its true explosive potential and depth. According to a report from Morgan Stanley, a core judgment is that leading large language models (LLMs) are undergoing "non-linear capability leaps," while computing power demand has shown a trend of systematically outstripping supply. From early January to March 2026, global weekly token usage surged from 6.4 trillion to 22.7 trillion, an increase of approximately 250% in just three months, forcing some LLM service providers to impose usage caps on users. Morgan Stanley predicts that the future growth rate of computing power demand will be about three times the projected CAGR of NVIDIA's computing power supply, indicating that computing shortages will persist and intensify long-term.
Energy presents another "ticking time bomb." Morgan Stanley's models forecast that U.S. data centers will face an electricity supply gap of approximately 55 gigawatts between 2025 and 2028. Already, $18 billion worth of data center projects have been canceled outright, with an additional $46 billion in projects delayed. Even when considering various "rapid power supply" solutions—such as natural gas turbines, fuel cells, and conversions of Bitcoin mining sites—the net electricity shortfall could still amount to 18% to 30% of the total U.S. data center deployment scale during that period.
The impact of AI on the labor market is already becoming apparent. A Morgan Stanley survey indicates that within the five industries most affected by AI, 11% of positions were eliminated due to AI over the past 12 months, with another 12% of vacancies left unfilled. New hiring accounted for only 18%, resulting in a net job reduction rate of approximately 4%. The report estimates that 90% of occupations will be affected to some degree by AI automation or enhancement.
The "step-change leap" in large model capabilities is more extreme than market expectations. Morgan Stanley identifies the "non-linear leap in frontier model capabilities" as one of the most significant thematic drivers for 2026, citing extensive data to support its view that the situation is far more extreme than anticipated. Third-party analysis from METR shows that the best current large models can independently complete complex tasks lasting over 15 hours continuously—whereas extrapolations from existing scaling laws suggested a capability of only about 8 hours. Actual capabilities have significantly surpassed theoretical projections.
Multiple independent data points corroborate this trend: tracking metrics from Artificial Analysis indicate that AI capabilities continue to advance rapidly; OpenAI CEO Sam Altman publicly warned at an AI summit in India that "the world is not ready, and highly capable models are coming"; researchers have used DNA sequencing and DeepMind's AlphaFold tool to develop a cancer vaccine for a pet dog; an experiment by New York Times technology editor Kevin Roose found that 54% of readers preferred AI-generated articles over human-written ones; frontier LLMs already possess the ability to solve open problems in physics; and reports suggest an unreleased model represents a "step-change leap in capability" in software programming, academic reasoning, and cybersecurity.
The report also references a prediction from Leo Aschenbrenner's paper "Situational Awareness": "There is a surprisingly plausible possibility of achieving AGI (Artificial General Intelligence) by 2027. In the four years from GPT-2 to GPT-4, we leaped from preschool level to smart high schooler level... If we traverse the same intelligence span again, where will it take us? Likely to models capable of surpassing PhDs and top experts across all professional domains."
The massive gap between computing supply and demand: a 250% surge in token usage masks a threefold demand disparity. If the leap in model capabilities is the "engine on the demand side," the severe shortage of computing power is the "ceiling on the supply side." Morgan Stanley identifies "systemic computing demand outstripping supply" as the core market theme for 2026. Data is starkly illustrative: according to tracking data from the OpenRouter platform, global weekly token usage rose from 6.4 trillion in early January to 22.7 trillion by March 2026, a roughly 250% increase in three months. The rapid adoption of agentic AI tools, exemplified by OpenClaw, has significantly accelerated demand-side explosion. Several LLM providers have begun imposing token usage caps on users to manage runaway demand growth. Morgan Stanley forecasts that overall computing demand growth will be approximately three times the projected CAGR of NVIDIA's computing supply.
Three parallel forces are compounding demand: the continuous expansion of AI use cases, the non-linear increase in AI task complexity, and the accelerated broadening of AI adoption. In practical terms, software programming is currently the single largest token-consuming application among all LLM use cases and is dominated by proprietary (closed-source) models.
Morgan Stanley's "Intelligence Factory" model reveals another critical dynamic: as chip generations transition from Blackwell to Rubin GPUs, the average token price is expected to drop by over 70%. This rapid decline in AI usage costs will further stimulate demand-side explosion, creating a self-reinforcing positive feedback loop. For example, a roughly 250-megawatt data center using Blackwell GPUs, with electricity costs at $100 per megawatt-hour and running GPT-4o queries, could yield top-tier LLM developers approximately a 60% profit margin. Morgan Stanley expects actual computing demand to reach about three times previous model forecasts. In this context, any company capable of breaking through computing scalability bottlenecks stands to benefit significantly. This includes not only the chip manufacturing supply chain but also memory, optical networking equipment, and core data center components. Morgan Stanley is highly optimistic about these "merchants of compute," believing they will directly benefit from this systemic supply-demand imbalance.
Energy is the lifeblood of AI: a 55-gigawatt shortfall and the race for "off-grid" solutions. Electricity has become the most critical physical constraint on AI infrastructure expansion. Based on its in-depth "AI Power" analysis model, Morgan Stanley has reached an alarming conclusion. Between 2025 and 2028, U.S. data center developers will face an electricity supply gap of approximately 55 gigawatts. Concurrently, $18 billion in data center projects have been canceled due to community opposition and concerns over rising electricity prices, with another $46 billion in projects delayed. Multiple constraints on data center growth are simultaneously intensifying: competition for grid interconnection resources, shortages of electrical equipment, labor shortages, and local political resistance.
To address this gap, Morgan Stanley analyzed four categories of "time to power" solutions: natural gas turbines (could address 15-20 GW, 90% probability of success); Bloom Energy fuel cells (could address 5-8 GW, 90% probability); deploying data centers at existing nuclear plant sites (could address 5-15 GW, 75% probability); and converting Bitcoin mining sites to data centers (could address 10-15 GW, 90% probability). However, even when combining the probability-weighted contributions of all these solutions, Morgan Stanley's baseline calculation indicates that the net electricity shortfall by 2028 will still equate to 18% to 30% of the total U.S. data center deployment scale during that period.
Strategically, Meta has begun taking proactive steps—providing funding for Terrapower's commercial sodium-cooled fast reactor project and directly investing in Louisiana's power infrastructure. Morgan Stanley views this as a potential strategic signal that AI giants are beginning to systematically control energy infrastructure to secure their computing lifeline.
Early signs of employment impact: AI adoption economic value exceeds 25% of S&P 500 pre-tax profits. Morgan Stanley's survey data and model calculations reveal the early yet profound impact of AI on the labor market. In the five industries most significantly impacted by AI (consumer goods retail, real estate management and development, transportation, medical equipment and services, and automobiles and components), field surveys show that over the past 12 months, AI has led to the direct elimination of 11% of positions; an additional 12% of vacancies were left unfilled; new hiring accounted for 18%, resulting in a net job reduction rate of approximately 4%. Notably, new hiring was weakest among smaller firms—possibly reflecting their more agile and rapid adoption of AI.
From a macro perspective, Morgan Stanley estimates that 90% of occupations will be affected to some degree by AI automation or enhancement, typically not through the "complete elimination of jobs" but through the "reconfiguration of task structures within roles." Quantifying the economic value, Morgan Stanley's calculated TAM (Total Addressable Market) for AI adoption is equally staggering: the cost-reduction potential from "AI automation" corresponds to a value TAM exceeding 25% of the S&P 500 index's expected adjusted pre-tax profits for 2026. This "AI automation" cost reduction is equivalent to over 40% of total employee compensation expenses. Agentic AI (software layer) and embodied AI (physical layer, e.g., robotics) contribute nearly equally to this value. By sector, the economic potential of AI adoption is highest relative to pre-tax profits in consumer goods retail, real estate management, transportation, and medical equipment.
The "moat" in an AI-disrupted world: what assets truly hold value? As AI capabilities accelerate, a core question becomes increasingly urgent: in a world where AI can replicate almost anything, what assets possess genuine defensiveness? The Morgan Stanley report cites investor Michael Bloch's framework, proposing a key distinction: "AI compresses the time it takes to do things, but it cannot compress the time it takes for things to happen naturally. This distinction is the most important screening criterion in investing today." Accordingly, asset types with genuinely defensive moats include five categories: continuously accumulated proprietary data—not static datasets, but dynamic data generated through defensible business operations; network effects—where each new user makes the product more valuable for others, with accumulated liquidity advantages becoming more pronounced as AI lowers barriers for competitors; regulatory licenses—bank charters take years, FDA approvals take years, regulatory barriers widen rather than narrow as AI capabilities advance; large-scale capital deployment capability—as bottlenecks shift from software to physical infrastructure, the ability to mobilize large-scale capital becomes a core advantage; and physical infrastructure—factories, power plants, data centers... physical laws set an unbreachable lower limit on time, and the lead of first-movers expands with each passing month.
The report further lists eight major asset classes likely to appreciate in the "transformative AI" era, encompassing: real estate with physical scarcity (AI infrastructure sites, industrial properties); AI application adopters with pricing power; luxury goods and unique services; platforms with network effects; authentic and unique human experiences; regulatory franchises; proprietary data and brands; and key semiconductor assets (advanced-node chip fabs, ASML's EUV lithography machines, rare earth processing capabilities).
Comments