By mid-2026, the global semiconductor industry is undergoing a profound structural transformation. The market’s discussion around the AI cycle has shifted from the short-term debate over "Is AI a bubble?" to a long-term assessment of whether AI represents a decade-long structural cycle. In this paradigm shift, two core variables are being systematically underestimated: the durability of AI demand and TSMC’s pricing power in advanced process technologies.
Morgan Stanley’s mid-2026 outlook report makes it clear: the US economy is demonstrating resilience far beyond expectations, with the primary driver being a wave of capital expenditure centered on AI. The firm has sharply raised its 2026 S&P 500 earnings growth forecast from 17% to 23%, and its US GDP growth forecast for the same period from 1.8% to 2.3%. Behind this upward revision is a phenomenon Morgan Stanley defines as "inelastic demand"—the ability of businesses and consumers to withstand higher prices, increased financing costs, and even heightened geopolitical risks is far greater than the market anticipated.
At the same time, the core segment of AI chip manufacturing—advanced process wafer foundry—is moving away from the old logic of "cost-driven declines" toward a new paradigm of "advanced process scarcity + AI-driven price increases." As the world’s only foundry capable of mass-producing at 3nm and below, TSMC is converting this scarcity into systematic pricing power. This article systematically analyzes the logic and trends behind this structural change in the semiconductor industry from three perspectives: the sustainability of the AI supercycle, the shift in semiconductor pricing power, and the restructuring of profit distribution along the supply chain.
The AI Demand Supercycle: Why 2027–2028 Remain Pivotal Growth Stages
The "Inelastic" Expansion of Capital Expenditure
To understand the sustainability of the AI demand supercycle, we first need to examine the scale and trajectory of capital expenditure.
According to Morgan Stanley, the combined capital expenditure of the five major hyperscale cloud providers—Amazon, Alphabet, Meta, Microsoft, and Oracle—is expected to reach around $800 billion in 2026, climbing further to $1.2 trillion in 2027. This nearly doubles the firm’s previous forecast from a year ago, which projected about $450 billion for both 2026 and 2027.
Even more noteworthy is the speed and magnitude of these adjustments. Morgan Stanley notes that, over the past six months, consensus forecasts for 2026–2027 capital spending have been revised upward by more than $630 billion. US hyperscale cloud companies are projected to spend $805 billion in 2026, nearly double the $433 billion forecasted a year prior, twice the actual 2025 spend, and triple the 2024 level. Looking ahead, the firm expects this figure to surpass $1.1 trillion in 2027 and approach $1.3 trillion in 2028.
From a broader perspective, Morgan Stanley projects that hyperscale companies’ cash capex-to-revenue ratios will reach 34%, 39%, and 37% in 2026, 2027, and 2028, respectively—surpassing the previous historical peak of about 32% during the late 1990s internet bubble. By 2026, AI-related capital expenditure is expected to account for over 50% of the total capex of all Russell 1000 Index constituents.
Goldman Sachs’ June 2026 update forecasts that the four leading hyperscale data center operators—Alphabet, Amazon, Microsoft, and Meta—will collectively spend $725 billion in 2026, a 77% year-over-year increase from $410 billion in 2025. In just the past six months, market expectations for 2026 cloud capex have risen by nearly 80%. Barclays projects that major cloud providers’ capex will reach $919 billion in 2027 and further rise to around $1.16 trillion in 2028.
Global Scale of AI Infrastructure Investment
On a global scale, Gartner forecasts that total AI spending will reach $2.59 trillion in 2026, up 47% year-over-year. Of this, AI infrastructure investment will rise from $975.6 billion in 2025 to $1.43 trillion in 2026, and further to $1.89 trillion in 2027. Global AI spending is expected to hit $3.49 trillion in 2027.
Nomura’s July 1, 2026 report notes that global server market growth forecasts have been raised from 43% to 74% for 2026 and 65% for 2027, while AI server growth forecasts have been increased from 58% to 78% for 2026 and 76% for 2027.
Structural Drivers of Sustained AI Demand
The "inelastic" nature of AI demand stems from three structural factors.
First, AI investment is both a "necessity" and a "highly coveted asset." Companies are eager to seize the next generation of core technology while fearing being left behind in the competition. Morgan Stanley’s Chief Cross-Asset Strategist Andrew Sheets points out that, for such a strategic priority, whether borrowing costs are 5.50%, 5.75%, or 6.00% has become a secondary concern.
Second, AI inference and AI Agents are emerging as new engines of demand. Morgan Stanley’s research shows that AI inference, Agents, and cloud service growth are driving storage demand to consistently exceed expectations. Global weekly token usage—a key proxy for computing demand—has surged about 350% since early January 2026, rising from roughly 6 trillion tokens to 28 trillion tokens.
Third, the depth and breadth of financing channels have far exceeded expectations. In the first five months of 2026, global AI-related bond issuance reached $236 billion—more than four times the same period in 2025. Morgan Stanley expects full-year issuance to surpass $570 billion. Innovations in financing—from project finance and structured tranches to residual value guarantees—are rapidly adapting to AI-driven capital expenditure needs.
The Shift in Semiconductor Pricing Power: From "Moore’s Law" to "Scarcity Premium"
TSMC’s Pricing Power: Data-Driven Evidence
If the AI demand supercycle is the demand-side driver of structural change in semiconductors, then TSMC’s pricing power is the supply-side variable at the core.
TSMC has officially raised its 2026 capex guidance to a record $52–56 billion, with a full-scale push into 2nm and CoWoS advanced packaging capacity. In terms of capacity layout, Hsinchu Baoshan and Kaohsiung plants will be the main 2nm production hubs, with monthly capacity expected to reach 100,000 wafers by 2026. For the in-demand 3nm process, TSMC is expanding its Nanke facility and converting equipment to flexibly shift some 5nm capacity to 3nm, narrowing the supply-demand gap.
On pricing, TSMC plans another price hike for its 3nm process in the second half of 2026, with increases of up to 15%, and a further 5–10% rise possible in 2027. Moreover, TSMC has notified clients that all advanced nodes at 7nm and below will see foundry price hikes of about 5–10%, covering roughly 75% of TSMC’s wafer revenue.
Deutsche Bank warned as early as January 2026 that TSMC’s entire 3nm capacity for 2026 was fully booked, with reservations extending into 2027. Both 2nm fabs are fully booked as well, with monthly capacity at about 35,000 wafers, expected to expand to 140,000 wafers by the end of 2026. In this "seller’s market," price hikes are not the result of negotiation but a reflection of TSMC’s unilateral pricing power—customers either accept or are left out.
The Industrial Logic Behind Pricing Power Shift
TSMC’s pricing power is not simply a result of supply-demand imbalance, but reflects a fundamental shift in the semiconductor industry’s logic.
In the past, consumer electronics cycles drove the industry, and semiconductors followed Moore’s Law—more advanced processes led to lower per-transistor costs and downward price trends. TSMC’s 3nm demand was previously driven mainly by smartphone SoCs, supported by a handful of major clients like Apple. However, with the onset of the AI server upgrade cycle, cloud providers such as Nvidia, AMD, Google, and AWS are rapidly adopting 3nm technology. Demand for AI accelerators and custom ASICs has surged, driving up wafer starts. The demand structure has shifted from a "single engine" to a "multi-engine" model.
Crucially, AI chips’ reliance on advanced process nodes is irreplaceable. TSMC commands 71% of the global foundry market and over 90% of the advanced process market at 7nm and below. In Q4 2025, TSMC held a 70.4% share of the global foundry market. This highly concentrated supply, combined with explosive AI-driven demand, has turned advanced nodes from a "cost center" into a "profit center."
Morgan Stanley’s June 30, 2026 Greater China semiconductor industry report maintains an "attractive" rating for the sector, highlighting robust AI semiconductor demand and long-term drivers such as chip inflation and AI’s cannibalization of non-AI semiconductors.
Profit Distribution Across the Supply Chain: Who Really Benefits?
GPU Supply Chain and Profit Concentration
Profits in the AI supply chain are increasingly concentrated upstream. Morgan Stanley notes that semiconductor vendors remain the most direct beneficiaries, with 2026 sales estimates revised up by about 60%.
In the GPU supply chain, NVIDIA remains the core supplier of compute hardware. As of July 1, 2026 (Beijing time), NVIDIA’s stock price stood at $194.97, with a market cap of about $4.72 trillion. But Morgan Stanley analyst Shane Brett points out that the market is entering a phase where semiconductor equipment suppliers’ returns are beginning to rival those of memory chip stocks. Brett has raised his semiconductor equipment spending forecast for 2026 to $143 billion (up from a previous estimate of $136 billion), a 23% increase year-over-year; his 2027 forecast is now $182 billion (up from $161 billion previously).
Looking more broadly at equipment spending, SEMI’s April 2026 "300mm Fab Outlook Report" projects global 300mm fab equipment spending will grow 18% to $133 billion in 2026 and 14% to $151 billion in 2027. CITIC Securities expects the global wafer fabrication equipment market to grow 26% and 35% year-over-year in 2026 and 2027, reaching $147.8 billion and $199.5 billion, respectively.
HBM and the Tight Memory Chip Cycle
In memory chips, HBM (High Bandwidth Memory) is becoming another key profit driver in the AI chip supply chain.
TrendForce data shows that 2026 HBM demand will mainly come from AI ASICs requiring higher capacity, with HBM per AI chip rising sharply from 96/192GB to 216/288GB. In 2027, with the launch of NVIDIA’s Rubin Ultra platform, HBM per GPU will further increase to 384GB. TrendForce estimates HBM wafer starts as a share of total DRAM will rise from 18% in 2025 to about 30% in 2027; HBM bit supply will climb from 8% to about 13%.
Morgan Stanley forecasts that AI-driven memory shortages will persist for two to three years. Gartner expects supply tightness to last through the end of 2027. At the June 2026 semiconductor industry leadership forum, TrendForce revealed that HBM will remain in short supply through 2027, with price increases inevitable and 2026 shipments expected to jump 60% year-over-year.
Looking at the global semiconductor market as a whole, the World Semiconductor Trade Statistics (WSTS) projects the market will grow nearly 90% from 2025 to reach $1.5 trillion in 2026, and a further 26.6% in 2027 to $1.914 trillion.
Risks and Constraints: The Other Side of the Supercycle
Leverage and Off-Balance-Sheet Risks
The AI supercycle is not without risks. Morgan Stanley’s series of reports systematically dissect the financial vulnerabilities behind this cycle.
The combined gross leverage ratio of the five major hyperscale cloud providers jumped from 0.9x in Q3 2025 to 1.8x in 2026—doubling in just two quarters and surpassing the average leverage of the entire energy sector. Free cash flow is also under pressure: by 2026, Amazon and Meta are expected to see free cash flow approach zero or turn negative, with only Google and Microsoft maintaining positive figures.
Off-balance-sheet risks are even more concerning. Morgan Stanley estimates that, in addition to reported capex, there are about $1.8 trillion in off-balance-sheet commitments, including roughly $982 billion in purchase obligations. The total value of future purchase contracts between hyperscale cloud providers and NVIDIA is close to $1 trillion. NVIDIA’s own inventory and purchase obligations have risen to about 32% of its expected FY2027 revenue, up from a historical range of 15–20%.
Physical Constraints and Return Validation
Beyond financial risks, physical constraints are also coming to the fore. Grid access, power generation equipment, skilled labor shortages, and permitting delays are becoming significant bottlenecks for AI infrastructure buildout.
Goldman Sachs’ June 2026 report notes that US tech investment as a share of GDP has risen to about 4.9%, exceeding the highs of the dot-com bubble era. Since November 2022, the market cap of AI-related companies has soared by $27 trillion, far outpacing the $9 trillion macro benchmark. The market’s pricing of future AI returns is clearly running ahead of the actual realization of productivity gains.
Conclusion: Structural Repricing, Not a Cyclical Bubble
To sum up, the current AI-driven changes in the semiconductor industry are much closer to a structural repricing than a simple cyclical bubble.
On the demand side, AI capital expenditure shows clear "inelasticity"—demonstrating far greater resilience to price, financing costs, and geopolitical risk than expected. The five major hyperscale cloud providers are projected to spend about $800 billion in 2026 and $1.2 trillion in 2027, while global AI spending is set to reach $2.59 trillion in 2026 and $3.49 trillion in 2027, pointing to a long-term investment cycle that will last at least through 2028.
On the supply side, TSMC’s irreplaceable monopoly in advanced process nodes—71% global foundry share and over 90% in advanced nodes—is translating into systematic pricing power. The 15% price hike for 3nm and fully booked 2nm capacity are not short-term phenomena but the inevitable result of the "advanced process scarcity + AI-driven demand" paradigm.
In terms of profit distribution, value is concentrating upstream from downstream applications to manufacturing and equipment. GPU vendors, foundries, semiconductor equipment makers, and HBM manufacturers are the core beneficiaries of this cycle.
Of course, risks remain—rapidly rising leverage, massive off-balance-sheet commitments, physical infrastructure constraints, and the tension between market valuations and fundamentals all pose potential constraints on this supercycle.
But at its core, the combination of the AI demand supercycle and semiconductor pricing power is reshaping the industry logic that has driven the global semiconductor sector for nearly three decades—shifting from cost reduction to value creation. The duration and depth of this transformation may far exceed current mainstream market expectations.
FAQ
Q1: What is the AI supercycle?
The AI supercycle refers to a long-term structural growth cycle driven by the large-scale commercialization of artificial intelligence, as opposed to short-term tech hype or cyclical fluctuations. Its features include years of massive capital investment, profit concentration upstream in the supply chain, and demand that is "inelastic" to price and financing costs. Morgan Stanley expects this cycle to last at least through 2028.
Q2: Why does TSMC have pricing power?
TSMC’s pricing power stems from the irreplaceability of its advanced process technologies. TSMC holds a 71% share of the global foundry market and over 90% in advanced nodes at 7nm and below, and is the only company worldwide capable of mass-producing at 3nm and below. With AI chip demand surging, TSMC’s 3nm capacity for 2026 is fully booked, with reservations extending into 2027. This supply monopoly, coupled with explosive demand, gives TSMC unilateral pricing power.
Q3: How long can AI infrastructure investment continue?
According to forecasts from Morgan Stanley, Goldman Sachs, Barclays, and others, the high-growth phase of AI infrastructure investment will last at least through 2028. The combined capex of the five major hyperscale cloud providers is expected to rise from about $800 billion in 2026 to $1.2 trillion in 2027 and nearly $1.3 trillion in 2028. Global AI spending is projected to increase from $2.59 trillion in 2026 to $3.49 trillion in 2027.
Q4: Has the AI chip cycle peaked?
Current data and institutional forecasts do not support the view that the AI chip cycle has peaked. The global semiconductor market is expected to reach $1.5 trillion in 2026 and $1.914 trillion in 2027. HBM demand is forecast to grow 60% year-over-year in 2026, with shortages persisting through 2027. AI server growth forecasts have been raised from 58% to 78% for 2026. However, rising leverage, valuation premiums, and physical constraints remain risks to watch.
