When explosive AI growth hits systemic supply bottlenecks, Morgan Stanley says that the current market optimism about this AI revolution may still seriously underestimate its true explosive power and depth.
On April 11, according to news from ZhuiFeng Trading Desk, the core judgment of Morgan Stanley's latest research report is: top large language models (LLMs) are experiencing a &#34;non-linear capability leap,&#34; and computing power demand has shown a state of systemically surpassing supply.
From early January to March 2026, global weekly Token usage surged from 6.4 trillion to 22.7 trillion, a jump of approximately 250% in just three months; some LLM service providers have been forced to impose usage caps on users. Morgan Stanley predicts that the growth rate of future computing power demand will be approximately 3 times the CAGR forecast for NVIDIA's computing power supply, and computing power shortages will be long-term and continue to intensify.
Energy is another &#34;ticking time bomb.&#34; Morgan Stanley's model predicts that between 2025 and 2028, U.S. data centers will face a power gap of approximately 55 gigawatts. Previously, $18 billion worth of data center projects were directly canceled, and another $46 billion worth of projects were delayed. Even after combining various &#34;fast power supply&#34; solutions such as natural gas turbines, fuel cells, and Bitcoin site conversions, the net power gap may still be as high as 18% to 30% of the total U.S. data center deployment scale during the same period.
The impact of AI on the labor market has begun to manifest. Morgan Stanley's survey shows that among the five industries most deeply affected by AI, AI has led to 11% of positions being cut in the past 12 months, and 12% of vacancies are no longer being recruited for after positions become empty; new hiring is only 18%, with a comprehensive net layoff rate of approximately 4%. The report estimates that 90% of occupations will be affected to some extent by AI automation or enhancement.
<h2>Large Model Capabilities &#34;Step-Change Mutation&#34;: The Situation is More Extreme Than Market Expectations</h2>
The Morgan Stanley report lists the &#34;non-linear leap in frontier large model capabilities&#34; as one of the most important thematic drivers for 2026 and cites extensive data to support its judgment that the &#34;situation is far more extreme than market expectations.&#34;
Latest analysis from third-party organization METR shows that the best large models can now independently complete continuous complex tasks for over 15 hours—whereas, according to extrapolations from existing technical Scaling Laws, the current level should be around 8 hours. Actual capabilities have significantly outpaced theoretical expectation tracks.
Multiple independent data points confirm this trend:
<ul>
<li>Continuous tracking indicators from Artificial Analysis show that AI capabilities are still advancing rapidly;</li>
<li>OpenAI CEO Sam Altman publicly warned at the India AI Summit: &#34;The world is not ready, extremely capable models are coming&#34;;</li>
<li>Researchers used DNA sequencing and DeepMind's AlphaFold tool to develop a cancer vaccine for their pet dog;</li>
<li>A reader experiment by New York Times tech editor Kevin Roose showed that 54% of readers preferred AI-generated articles over human writing;</li>
<li>Frontier LLMs already possess the capability to solve open problems in the field of physics;</li>
<li>There are reports that an unreleased model represents a &#34;step-change mutation in capability&#34; in the fields of software programming, academic reasoning, and cybersecurity.</li>
</ul>
The report also cited the pre-judgment in Leo Aschenbrenner's paper &#34;Situational Awareness&#34;:
<blockquote>
&#34;There is a surprising possibility of achieving AGI (Artificial General Intelligence) by 2027. In the four years from GPT-2 to GPT-4, we jumped from a preschool level to a smart high school student level... If we go through the same intelligence span again, where will it take us? Likely to models capable of surpassing PhDs and top experts in all professional fields.&#34;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/fb901ba3-970a-49b3-a604-cc03a31912fb.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/fb901ba3-970a-49b3-a604-cc03a31912fb.png"/>
</blockquote>
<h2>Huge Gap in Computing Power Supply and Demand: Behind 250% Token Growth is a 3x Demand Gap</h2>
If the leap in large model capabilities is the &#34;engine on the demand side,&#34; the severe shortage of computing power supply is the &#34;ceiling on the supply side.&#34; Morgan Stanley lists &#34;computing power demand systemically surpassing supply&#34; as the core market theme for 2026.
The report states that the data is very intuitive:
<ul>
<li>According to actual tracking data from the OpenRouter platform, from early January to March 2026, global weekly Token usage increased from 6.4 trillion to 22.7 trillion, an increase of approximately 250% within three months;</li>
<li>The popularity of AI agent tools (represented by OpenClaw) has significantly accelerated the explosion on the demand side;</li>
<li>Multiple LLM service providers have begun to set Token usage limits for users to cope with the uncontrolled growth in demand;</li>
<li>Morgan Stanley predicts that the overall growth rate of computing power demand will be about 3 times the CAGR forecast for NVIDIA's computing power supply;</li>
<li>Three parallel driving forces promoting demand are superimposed: the continuous expansion of AI usage scenarios, the non-linear improvement of AI task complexity, and the accelerated broadening of AI adoption scope.</li>
</ul>
At the specific application level, software programming is currently the single largest use case for Token consumption among all LLM usage scenarios, and this field is dominated by proprietary (closed-source) models.
Morgan Stanley's &#34;Intelligence Factory&#34; model reveals another layer of key logic: as chip generations migrate from Blackwell to Rubin GPUs, the average Token price is expected to fall by more than 70%—the rapid decline in AI usage costs will further stimulate the explosion on the demand side, forming a self-reinforcing positive feedback loop of demand.
Taking a specific case as an example: a data center with a scale of about 250 megawatts, using Blackwell GPUs, a power cost of $100 per megawatt-hour, and running GPT-4o queries, can bring a profit margin of about 60% to top large model developers.
Morgan Stanley expects that actual computing power demand will reach about 3 times previous model predictions. In this context, any company that can break the bottleneck of computing power expansion will welcome significant benefits. This includes not only the chip manufacturing supply chain but also memory, optical networking equipment, and core components of data centers. Morgan Stanley is extremely bullish on this group of &#34;Merchants of Compute,&#34; believing they will directly benefit from this systemic imbalance between supply and demand.
<h2>Energy is the Lifeblood of AI: 55 GW Gap and the Race for &#34;Off-Grid&#34; Solutions</h2>
Electricity has become the most critical physical constraint for AI infrastructure expansion. Based on its &#34;AI Power Supply&#34; in-depth analysis model, Morgan Stanley has reached a sobering conclusion.
Between 2025 and 2028, U.S. data center developers will face a power supply gap of approximately 55 gigawatts. At the same time, $18 billion worth of data center projects have been directly canceled due to community opposition and concerns over rising electricity prices, and another $46 billion worth of projects have been delayed. Multiple headwinds restricting data center growth are fermenting simultaneously: competition for grid access resources, shortages of power equipment, labor shortages, and local political resistance.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/62c61ec4-70f5-43c7-97e5-ed1d7401dc06.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/62c61ec4-70f5-43c7-97e5-ed1d7401dc06.png"/>
Facing this gap, Morgan Stanley sorted out four types of &#34;Time to Power&#34; solutions:
<blockquote>
<ul>
<li>Natural gas turbines: Can alleviate a 15–20 GW gap, with a 90% probability of success;</li>
<li>Bloom Energy fuel cells: Can alleviate a 5–8 GW gap, with a 90% probability of success;</li>
<li>Deploying data centers relying on existing nuclear power plants: Can alleviate a 5–15 GW gap, with a 75% probability of success;</li>
<li>Converting Bitcoin mining sites into data centers: Can alleviate a 10–15 GW gap, with a 90% probability of success.</li>
</ul>
</blockquote>
However, even if the probability-weighted contributions of all the above solutions are superimposed, Morgan Stanley's baseline calculation shows that by 2028, the net power gap will still be equivalent to 18% to 30% of the total U.S. data center deployment scale during the same period.
From the perspective of strategic layout, Meta has begun to take active action—providing funds for Terrapower's sodium-cooled fast reactor commercialization project and directly investing in power infrastructure in Louisiana.
Morgan Stanley believes this may be a strategic signal that AI giants are starting to systemically take control of energy infrastructure to ensure their computing power lifeblood.
<h2>Employment Shocks Emerge, Economic Value of AI Adoption Exceeds 25% of S&amp;P 500 Pre-Tax Profit</h2>
Morgan Stanley's survey data and model calculations reveal the early and profound impact of AI on the labor market.
In the five industries most significantly impacted by AI (Consumer Staples &amp; Retail, Real Estate Management &amp; Development, Transportation, Healthcare Equipment &amp; Services, Automobiles &amp; Components), Morgan Stanley's field survey shows:
<blockquote>
<ul>
<li>In the past 12 months, AI has already led to 11% of positions being directly cut;</li>
<li>Another 12% of vacancies are no longer being recruited for after positions become empty;</li>
<li>New hiring is 18%, with a comprehensive calculated net layoff rate of approximately 4%;</li>
<li>Notably, new hiring volume is weakest in smaller enterprises—this may reflect that small businesses are more flexible and rapid in their application of AI.</li>
</ul>
</blockquote>
From a macro perspective, Morgan Stanley estimates that 90% of occupations will be affected to some extent by AI automation or enhancement, and the method of impact is usually not &#34;eliminating job positions entirely,&#34; but &#34;reconfiguring the task structure within positions.&#34;
From the perspective of quantitative economic value, the TAM (Total Addressable Market) for AI adoption calculated by Morgan Stanley is equally astonishing:
<blockquote>
<ol>
<li>
The value TAM corresponding to the cost reduction potential brought by &#34;AI automation&#34; exceeds 25% of the expected adjusted pre-tax profit for the S&amp;P 500 index in 2026;
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4d95df85-8ff5-4671-8c25-ab2ee3532510.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/4d95df85-8ff5-4671-8c25-ab2ee3532510.png"/>
</li>
<li>
This &#34;AI automation&#34; cost reduction is equivalent to more than 40% of total employee compensation expenditure;
</li>
<li>
Among these, agentic AI (software level) and embodied AI (physical level represented by robots) contribute nearly half of the value each;
</li>
<li>
From the perspective of industry distribution, the economic potential of AI adoption is highest relative to pre-tax profits in the fields of Consumer Staples &amp; Retail, Real Estate Management, Transportation, and Healthcare Equipment.
</li>
</ol>
</blockquote>
<h2>The &#34;Moat&#34; of AI Disruption: What Assets Can Truly Retain Value in the AI Era?</h2>
As AI capabilities accelerate their leap, a core question becomes increasingly urgent: In a world where AI can replicate almost everything, what kind of assets possess true defensiveness? The Morgan Stanley report cites investor Michael Bloch's framework, proposing a key distinction:
<blockquote>
&#34;AI compresses the time required to accomplish things, but it cannot compress the time required for things to occur naturally. This distinction is the most important screening criterion in current investment.&#34;
</blockquote>
Based on this, asset types with truly defensive moats include five categories:
<ol>
<li>Continuously accumulating proprietary data—not static datasets, but dynamic data continuously generated through defensible business operations;</li>
<li>Network effects—every additional user makes the product more valuable to other users; network advantages with accumulated liquidity will be further highlighted as AI lowers the threshold for creating competitors;</li>
<li>Regulatory permissions—banking licenses take years, FDA approvals take years; regulatory barriers expand as AI capabilities improve, rather than narrow;</li>
<li>Large-scale capital deployment capability—when the bottleneck shifts from software to physical infrastructure, the ability to mobilize large-scale capital itself becomes a core advantage of the era;</li>
<li>Physical infrastructure—factories, power plants, data centers... the laws of physics set an unbreakable lower limit on time, and the first-mover advantage expands every month.</li>
</ol>
The report further lists eight categories of assets that may appreciate in the &#34;transformative AI&#34; era, covering: real estate with physical scarcity (AI infrastructure land, industrial real estate), 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 critical semiconductor assets (advanced process chip fabs, ASML's EUV lithography machines, rare earth processing capabilities).
<hr/>
<pre><code>
The exciting content above comes from ZhuiFeng Trading Desk.

For more detailed interpretations, including real-time analysis, first-line research, etc., please join [**ZhuiFeng Trading Desk ▪ Annual Membership**]

![Image](https://wpimg-wscn.awtmt.com/3c4a713c-7a38-4582-9850-d0eabaf0e7ad.png)

Risk Warning and Disclaimer

The market is risky, and investment requires caution. This article does not constitute personal investment advice, nor does it take into account the special investment objectives, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are consistent with their specific circumstances. Investment based on this is at your own risk.</code></pre>

NVDL

SOXL

NVDA

NVDU

SOXX

NVDY

NVDX

Morgan Stanley's latest research report points out that top large language models are experiencing non-linear capability leaps, and computing power demand has systemically surpassed supply. It is predicted that the growth rate of computing power demand in the future will be three times NVIDIA's supply forecast, and computing power shortages will persist for a long time. U.S. data centers will face a power gap of approximately 55 gigawatts between 2025 and 2028. The impact of AI on the labor market has already appeared, with AI leading to 11% of positions being cut in the past 12 months. Morgan Stanley believes that the market's optimism about the AI revolution may still be underestimated

- Morgan Stanley reports that the current market may underestimate the impact of the AI revolution amid systemic supply constraints.  
- From January to March 2026, global weekly Token usage surged by 250%, revealing a significant gap between AI demand and computing supply, especially as energy shortages in U.S. data centers could reach 55 gigawatts by 2028.  
- The report emphasizes that AI's rapid advancement is beginning to disrupt the labor market, with 11% of jobs in the most affected sectors eliminated and an estimated 90% of occupations influenced by automation or augmentation.

Wallstreetcn

GraniteShares 2x Long NVDA Daily ETF

Direxion Semicon Bull 3X

VanEck Semiconductor ETF

Spdr Select Tech

NVIDIA

Direxion Daily NVDA Bull 2X Shares

iShares Semiconductor ETF

SPDR S&P Semicon

YieldMax NVDA Option Income Strategy ETF

T-Rex 2X Long NVIDIA Daily Target ETF

Top LLMs' "Capability Mutation," Computing Power Demand "Systemically Surpasses Supply" -- Morgan Stanley: "The Level of Market Optimism May Not Be Enough"