Overview The way markets price the AI supply chain is quietly changing. For two years, the binding constraint was silicon: GPU allocation, advanced packaging capacity, HBM yields. The question now beiOverview The way markets price the AI supply chain is quietly changing. For two years, the binding constraint was silicon: GPU allocation, advanced packaging capacity, HBM yields. The question now bei
AI Data Center Power Crunch: Which Semiconductor Stocks Face the Biggest Risk?
Overview
The way markets price the AI supply chain is quietly changing. For two years, the binding constraint was silicon: GPU allocation, advanced packaging capacity, HBM yields. The question now being asked on sell-side desks is narrower and more physical. When does the substation come online, and when does the interconnection permit clear?
According to Benzinga, a Morgan Stanley research note published on October 5 estimates that US data centers face a net power shortfall of roughly 32 gigawatts through 2028 even after accounting for mitigation measures, equivalent to a 34% gap against required supply. The more consequential part of the note is its tiering. Nvidia and Broadcom are described as relatively insulated, while suppliers of memory, optical components, power management and analog chips sit closer to the risk, because delayed deployments translate directly into deferred orders and inventory friction.
That is the first framework of this cycle to convert a macro infrastructure constraint into differentiated single-stock exposure, and it deserves to be read carefully rather than as a headline number.
Key Takeaways
The 32 gigawatt figure is a net number, not a gross one. Morgan Stanley's estimate already assumes the industry deploys behind-the-meter generation, gas turbines and nuclear co-location, so it measures the portion that remains genuinely unsourced.
Interconnection is the hardest constraint. The Queued Up 2026 Edition from Lawrence Berkeley National Laboratory reports that more than 2,060 gigawatts of generation and storage capacity were actively seeking grid connection at the end of 2025.
Leading chip vendors hold the allocation decision. When a campus is short of power, operators light up the core accelerators and networking first, and vendors with visibility into deployment sites can redirect supply toward projects that have energy available.
Secondary components absorb the timing risk. Memory, optics, power management and analog parts ship against full-system build schedules, so a rack that cannot be energized is a rack whose bill of materials gets rescheduled.
Deferral is not demand destruction. The power constraint reshapes when revenue is recognized rather than whether AI compute is wanted, but for companies valued on quarterly delivery, timing is itself a risk.
Power Is Replacing Silicon as the Binding Constraint
How the Shortfall Is Calculated
The number only means something in context. Thirty-two gigawatts is a net figure, which implies that after Morgan Stanley assumes the industry leans on on-site generation, peaking gas, fuel cells and nuclear co-location, roughly a third of required power still has no identified source. A separate and grosser framing attributed to the bank has circulated as well, putting incremental data center need at about 68 gigawatts across 2026 to 2028 against roughly 30 gigawatts of built or contracted capacity, leaving a gap near 38 gigawatts with no clear answer. The two estimates use different boundaries but point the same way. Supply-side expansion is not keeping pace with order-side commitments.
Demand data corroborates the picture. The US Energy Information Administration forecasts in its Short-Term Energy Outlook that national electricity consumption reaches 4,135 billion kilowatthours in 2026 and 4,211 billion kilowatthours in 2027, record levels in both years, with commercial sector sales rising 3.3% and 2.7% and accounting for roughly 63% and 56% of total growth. Data centers are the dominant driver of that increment.
Queues and Local Approvals Are the Real Gate
The shortage is less about generating capacity in aggregate and more about transmission and permitting timelines. Interconnection requests commonly take five to seven years to study and complete, while hyperscale buyers want capacity inside two to three years. No purchase order resolves that mismatch.
Texas offers a live case study. As Power Magazine reported, Governor Abbott ordered a pause on new data center connections to the ERCOT grid on August 3, 2026 pending a project-by-project audit, affecting roughly 49.8 gigawatts of pending load, close to a fifth of the national pipeline of about 253 gigawatts. ERCOT's large-load queue stands near 474 gigawatts, with roughly 90% attributed to data centers, and BloombergNEF estimates delays could cost affected projects between $8 billion and $15 billion cumulatively by the first quarter of 2027.
Events of this kind change no company's technology roadmap. They change delivery schedules, which for a supply chain recognizing revenue quarterly is a financial variable in its own right.
Why Nvidia and Broadcom Look Relatively Protected
Scarce Power Goes to Core Compute First
When available power at a campus falls short of plan, the triage is predictable. Operators energize the accelerators and networking that generate token revenue and push ancillary equipment back. That is the core of Morgan Stanley's reasoning. The bank cites visibility into where chips are deployed, flexibility to shift geographically, and close coordination with data center and power infrastructure partners as reasons the two leaders can redirect scarce supply toward projects with energy available, and it left their 2027 estimates unchanged on power grounds.
Reported numbers still support that view. Nvidia's second-quarter filing with the Securities and Exchange Commission shows fiscal Q2 2027 revenue of $96.2 billion, up 106% year over year, with Data Center revenue of $89.0 billion, up 117%, and third-quarter guidance of $108.0 billion. Broadcom's fiscal third-quarter 8-K reports revenue of $29.6 billion, up 86%, with AI semiconductor revenue of $16.7 billion, up 221% year over year, and guidance for AI semiconductor revenue to accelerate to $21.7 billion in the fourth quarter.
Chip Vendors Are Now Underwriting Power Themselves
The sharper signal is that leading vendors have stopped waiting for customers to solve the energy problem. According to Nvidia's announcement on the PORTS-Pike Technology Campus in Ohio, the company is providing credit guarantees for the land, power and shell buildout to secure an initial 4.25 IT-GW with an option on the remaining 3.75 IT-GW of an 8 IT-GW plan, alongside a $1.5 billion investment in developer SB Energy, with capacity phasing online from 2028 and at least $4.2 billion of grid infrastructure investment. Jensen Huang described land, power and shell as having become vital in the age of AI.
Directing credit toward generation and interconnection is, in effect, buying insurance on your own shipment schedule. Only a handful of balance sheets can do that, which is precisely why the constraint produces structural differentiation inside one supply chain. Demand, meanwhile, is not slowing. Reported infrastructure commitments across the model developers remain large, and the scale involved is covered in this look at Anthropic's AI infrastructure buildout.
Why Memory, Optics and Secondary Components Slip First
Memory Carries the Most Direct Scheduling Risk
Memory is the most elastic link in this cycle and the one most tied to full-system build plans. According to Micron's fiscal fourth-quarter filing, the company reported quarterly revenue of $54.23 billion and fiscal 2026 revenue of $133.19 billion, with its core data center business unit contributing $18.0 billion in the quarter, and guided fiscal first-quarter 2027 revenue to $61.5 billion plus or minus $1.5 billion. Management pointed to strategic customer agreements as a source of confidence in demand durability.
Those figures explain the market's sensitivity. A very high growth base combined with very high margins means any change in schedule gets amplified through both earnings and multiple. Memory suppliers ship against customer bring-up plans, and when racks cannot be energized, the cadence of receipt gets renegotiated. The detail behind the latest quarter is covered in this earnings breakdown.
Optics and Power Management Sit on the Same Build Line
Optical interconnect and power management face the same mechanics. Shipments of 800G and 1.6T modules track switch and rack deployment, so revenue recognition at optical suppliers such as Coherent and Lumentum depends on whether downstream halls come online on schedule. Power management and analog devices are similar, serving the electrical architecture of complete systems rather than individual accelerators, which binds them more tightly to the question of whether a hall has power at all.
This is exactly where Morgan Stanley's caution lands. If deployment slips, these suppliers face more than deferred revenue. They face inventory digestion and order revision, which for a segment that has just expanded capacity quickly is a more immediate risk than any slowdown in end demand.
Server and Rack Builders Sit in the Middle
Original equipment makers occupy the awkward middle ground, enjoying the order surge while carrying delivery-date risk. Dell Technologies' fiscal second-quarter filing shows the company posting revenue of $47.0 billion, up 58% year over year, record AI server orders of $60.9 billion, an AI backlog of $95 billion, and a raised full-year outlook of $74.0 billion in AI-optimized server revenue.
The larger the backlog, the more sensitive conversion becomes to power availability. The order book is real, but the timing of its conversion into revenue is not entirely within the vendor's control. The same logic applies across AI infrastructure suppliers, and it is a useful lens on recent moves such as the rally in HPE shares.
The Supply Side Runs Through Grids, Transformers and Cooling
Equipment Lead Times Have Become Common Knowledge
The constraint is not confined to generation. pv magazine USA reported that US power transformer lead times now extend to roughly four years, with prices up about 80% over five years, while demand for generator step-up transformers rose 274% and substation transformers 116% between 2019 and 2025. Limited supply of grain-oriented electrical steel and copper caps how quickly production scales. Hitachi Energy has committed around $1 billion to expansion with a South Boston plant due in 2028, and Siemens is investing $421 million in a Charlotte transformer facility, yet analysts generally expect the imbalance to persist for years.
Order books at electrical equipment makers tell the same story. Eaton's second-quarter filing shows the company generating $8.5 billion in quarterly sales, up 21%, with Electrical Americas rolling twelve-month orders up 41% organically and Electrical Global backlog up 103% year over year.
Thermal Management Has Moved From Ancillary to Core
Rising rack power density has turned cooling from a support system into primary equipment. According to Vertiv's second-quarter release, the company reported net sales of $3,274 million, up 24% year over year, and raised full-year 2026 guidance to between $13.8 billion and $14.2 billion. Management noted that demand for AI and general compute continues to intensify and that each technology advancement makes deployments more complex.
That gives power and thermal suppliers a different exposure profile from secondary chip vendors. Their constraint is how fast they can expand their own capacity. The chip suppliers' constraint is whether the customer's hall gets energized on time.
Gas, Nuclear and Renewables Are Filling the Gap Unevenly
The supply response is forming along three tracks. Gas is the fastest. GE Vernova's second-quarter release shows the company booking $24.2 billion of orders, up 88% organically, with total backlog at $176 billion and gas equipment backlog plus slot reservation agreements rising from 100 gigawatts to 116 gigawatts, management guiding to at least 125 gigawatts by year-end, and year-to-date data center orders above $5 billion, more than double the 2025 total.
Nuclear offers a longer-dated and steadier answer. datacentres.com reported that Constellation Energy announced on April 22, 2026 a 20-year agreement to supply 2 gigawatts of nuclear power to a hyperscale developer at a site beside the Susquehanna station in Pennsylvania, estimated at roughly $35 billion over its life and structured behind the meter to bypass the PJM queue. Small modular reactors move slower still, with the framework between Equinix and Oklo for up to 500 megawatts dependent on reactors actually entering service before power purchase agreements are negotiated.
Renewables continue to lead on installed capacity, with EIA projecting solar reaching 181 gigawatts in 2026 and 222 gigawatts in 2027, but their output profile means they need storage and gas alongside them to serve a constant data center baseload.
Deployment Delay Is Not Demand Destruction
Timing Risk Is Still Risk
The most easily missed element of the Morgan Stanley framework is that the primary effect of a power shortfall is delayed deployment rather than destroyed demand. If chips cannot be installed because power is unavailable, customers are more likely to postpone or restructure orders than to scale back their AI ambitions.
That is not a comfortable conclusion for portfolios. For companies valued on order books and backlog, pushed-out revenue compresses the cushion between near-term delivery and embedded expectations. In a segment priced for high growth, a single quarter of slippage is enough to trigger a sharp repricing, something visible in recent technical developments in AMD shares.
For readers who want to translate a supply chain view into actual positioning, it is worth first understanding how to trade US equities with stablecoins on MEXC before deciding how to weight silicon, power equipment and system builders.
Three Scenarios and the Signals That Separate Them
In a mitigation scenario, gas slot reservations convert on schedule, behind-the-meter projects reach financial close, and state-level approvals normalize, gradually compressing the 32 gigawatt net gap. Deferral risk at secondary suppliers falls, and revenue recognition across the chain resynchronizes.
In a base case, the gap persists without widening. Leading chip vendors retain high delivery visibility, secondary suppliers see wider quarter-to-quarter variance, and power and thermal equipment makers hold visibility on the strength of long-dated backlog.
In a deterioration scenario, more states follow the Texas model with audits and pauses, transformer and turbine lead times stretch further, and slippage spreads from individual campuses to an industry pattern. The market's question would shift from who can secure chips to who can secure electricity, and that shift would reorder relative valuations across the sector.
The variables worth tracking from here are few. The conclusions of the Texas audit and the final SB 6 rule text, whether GE Vernova converts gas backlog toward its 125 gigawatt year-end target, whether transformer lead times improve at the margin, and how memory and optical suppliers describe customer pull-in schedules on their next calls. Those tell more about the gap than any aggregate forecast.
Exclusive View from James Mitchell
For James Mitchell, the value of this research is not the 32 gigawatt figure but the executable framework it supplies. For two years the AI supply chain traded as a single beta, with every link sharing one narrative. Morgan Stanley's note is the first to separate shared demand from unevenly distributed timing risk. That distinction matters far more for portfolio construction than it does for any individual name.
Two misreadings are likely. The first is treating a power shortfall as a demand peak. Nvidia Data Center revenue up 117%, Broadcom AI semiconductor revenue up 221%, and a Dell AI backlog of $95 billion all describe an order book under scarcity pressure, not one in retreat. The second is reading insulated as risk-free. The protection enjoyed by the leaders comes from allocation power and balance sheet capacity, not immunity, and if slippage spreads from individual projects into a systemic pattern, aggregate capital expenditure cadence is affected too.
The signals that matter next are delivery signals in time series rather than installed-base forecasts. Three sets are worth monitoring. The conversion rate from order to recognized revenue at electrical and gas equipment makers, where GE Vernova's slot reservation agreements are the cleanest test. The language in secondary suppliers' filings about customer pull-in cadence and inventory days. And the progression of state-level regulatory milestones, where the final text of the Texas SB 6 rule is the most informative sample of this cycle. Layered together, those three say more about real deployment speed than any single company's guidance.
Viewed across assets, the episode rhymes with classic energy and infrastructure cycles. When the bottleneck migrates from a tradable commodity to a physical asset that cannot be scaled quickly, pricing power moves upstream. Four-year transformer lead times, 2,060 gigawatts sitting in interconnection queues, and a 474 gigawatt large-load queue in Texas are all signatures of asset scarcity rather than component scarcity. Crypto markets know this pattern well, since the contest for compute, power and sites has been the recurring theme of every mining cycle, only this time the participants carry balance sheets several orders of magnitude larger. The risk management implication is correspondingly direct. Facing a variable whose direction is reasonably clear but whose timing is not, positions should be sized to the probability of delay rather than to the slope of demand.
The bottleneck may be shifting from chips to electricity.
FAQ
What does Morgan Stanley's 32 gigawatt power shortfall actually mean?
It is the bank's estimate of the net power shortfall facing US data centers through 2028, equivalent to roughly 34% of required supply. The word net is the important one, because the figure already deducts mitigation measures such as on-site generation, gas turbines and nuclear co-location. It measures the portion of demand that still has no identified source under current plans, and it quantifies the gap between how fast power can be added and how fast compute is being ordered.
Why are Nvidia and Broadcom considered less exposed?
Morgan Stanley points to their visibility into where chips are ultimately deployed, their flexibility to shift supply geographically, and their coordination with data center and power infrastructure partners. When a campus is short of power, operators generally energize core accelerators and networking first, which keeps revenue recognition for the leading vendors comparatively stable. The bank left its 2027 estimates for both companies unchanged on power grounds.
Why do memory and optical suppliers carry more risk?
Their shipments track full-system bring-up schedules. Accelerators can be redirected to projects that have power, whereas memory, optics, power management and analog parts serve the electrical and interconnect architecture of complete racks and cannot be received as planned if a hall is not energized. Morgan Stanley warns that delayed deployments can bring order revisions and inventory digestion, which is especially sensitive for segments that recently expanded capacity.
Does the power shortage mean AI demand has peaked?
The available evidence suggests it affects timing rather than aggregate demand. If chips cannot be installed, customers tend to postpone or restructure orders instead of cutting their AI strategies. That said, deferral is a real financial risk for companies recognizing revenue quarterly, and high-growth segments often react sharply to even a single quarter of slippage.
Why is the interconnection queue so hard to clear?
Because the constraint lies in transmission construction and permitting rather than equipment procurement. Interconnection requests typically take five to seven years to study and complete, while hyperscale buyers want capacity within two to three years. Lawrence Berkeley National Laboratory data shows more than 2,060 gigawatts of generation and storage still queued at the end of 2025, a backlog no individual project can accelerate away.
How significant is the Texas pause on data center connections?
Texas ordered a halt to new data center connections to the ERCOT grid on August 3, 2026 pending a project-by-project audit, affecting roughly 49.8 gigawatts of pending load, close to a fifth of the national development pipeline. ERCOT's large-load queue runs to about 474 gigawatts, roughly 90% of it data centers. BloombergNEF estimates the delay could cost affected projects $8 billion to $15 billion cumulatively by the first quarter of 2027.
Which parts of the chain stand to benefit from the power constraint?
The beneficiaries sit on the supply side. In gas generation, GE Vernova's equipment backlog and slot reservation agreements have risen from 100 gigawatts to 116 gigawatts. In electrical equipment, Eaton's Electrical Global backlog is up 103% year over year. In thermal management, Vertiv has raised full-year 2026 guidance to between $13.8 billion and $14.2 billion. Nuclear offers a longer-dated answer, but its delivery timelines are materially slower.
Disclaimer
The information above is provided for general market information and analysis only and does not constitute investment advice, financial advice, legal advice, tax advice or a recommendation to trade. Prices of crypto assets, equities and other related financial assets can fluctuate sharply, and past performance, technical indicators and on-chain data do not guarantee future results. The research estimates, reported results, regulatory developments and market forecasts referenced here may change over time, and third-party research views are judgments rather than established facts, so the latest official disclosures from the relevant companies, regulators and government agencies should be treated as authoritative. Readers should conduct their own research and make decisions based on their own financial circumstances, investment objectives and risk tolerance, consulting a qualified professional where appropriate. The MEXC Crypto Pulse team accepts no liability for any direct or indirect loss arising from the use of this information.
About the Author
James Mitchell specializes in technical analysis, market trends, and trading strategies for both Bitcoin and altcoins. Based in London, he has over 10 years of experience in financial markets. Before joining MEXC Learn, James worked as a senior analyst at a leading European investment firm, where he developed expertise in risk management and quantitative trading. His transition to cryptocurrency markets began in 2017, and he has since become recognized for his data-driven approach. He holds a Master's degree in Financial Economics from the London School of Economics. His analytical approach combines traditional technical analysis with on-chain metrics to provide readers with actionable insights.
Areas of Expertise: Technical Analysis, Market Trends & Cycles, Trading Strategies, Bitcoin & Altcoin Analysis, Risk Management.
Research References
Want the fastest access to MEXC's latest updates? Join our official Telegram group now!
Account Verification: Understand KYC | How to Complete KYC
Market Opportunity
Gensyn Price(AI)
$0.02169
$0.02169$0.02169
USD
Gensyn (AI) Live Price Chart
The articles shared on this page are sourced from public platforms and are provided for reference only. They do not represent the position or views of MEXC. All rights belong to James Mitchell. If you believe any content infringes upon the rights of a third party, please contact service@support.mexc.com for prompt removal. MEXC does not guarantee the accuracy, completeness, or timeliness of any content and is not responsible for any actions taken based on the information provided. The content does not constitute financial, legal, or other professional advice, nor should it be interpreted as a recommendation or endorsement by MEXC. For expert insights and in-depth analysis, visit MEXC Learn.
Latest Updates on Gensyn
View More
DGAI Token Jumps 93%: Can DGrid AI Sustain the Hype?
DGAI made a high-volatility market debut, rising approximately 93% on its first day of trading and reaching a market capitalization of about $110 million. The move immediately put DGrid AI into the spotlight at the intersection of two of crypto’s most active narratives: decentralized infrastructure and artificial intelligence.
2026/08/26
What Is DGAI? Inside DGrid AI’s Tokenized Inference Network
DGAI powers DGrid AI’s decentralized inference network. Learn how payments, staking and node rewards could affect demand, supply and price
2026/09/16
MUSEBOOK Coin Surges as AI Agents Find Their Own Social Network
MUSEBOOK surged after an AI-agent social network went viral. Explore its Robinhood Chain token, attention drivers, contract risks and outlook.
2026/09/16
You May Also Like
HOT
Currently trending cryptocurrencies that are gaining significant market attention
Crypto Prices
The cryptocurrencies with the highest trading volume
Newly Added
Recently listed cryptocurrencies that are available for trading