AI Super Cycle and Semiconductor Revaluation: How Does TSMC's Pricing Power Reshape the Global Chip Supply Chain?

Mid-2026, the global semiconductor industry is undergoing a profound structural transformation.
Market discussions on the AI cycle are shifting from short-term debates over 'whether AI is a bubble' to long-term judgments on 'whether AI is a structural cycle on a decade-long scale'.
In this transformation, two core variables are being systematically underestimated: first, the persistence of AI demand, and second, TSMC's pricing power in advanced process nodes.
Morgan Stanley clearly stated in its mid-2026 outlook report that the U.S. economy displays stronger-than-expected resilience, with the core driving force being the wave of AI-centered capital expenditures.
The bank significantly revised up its 2026 S&P 500 earnings growth forecast from 17% to 23%, and its U.S. economic growth forecast for the same period from 1.8% to 2.3%.
Behind this round of upgrades is a phenomenon defined by Morgan Stanley as 'inelastic demand' - the ability of businesses and consumers to withstand higher prices, higher financing costs, and even higher geopolitical risks far exceeds market expectations.
At the same time, the core link of AI chip manufacturing - advanced process wafer foundry - is shifting from the old logic of 'cost reduction driven' to a new paradigm of 'advanced process scarcity + AI demand-driven price increases'.
As the world's only foundry capable of mass production at 3nm and below nodes, TSMC is turning this scarcity into systematic pricing power.
From the three dimensions of the persistence of the AI demand super cycle, the shift in semiconductor pricing power, and the restructuring of the industry chain profit structure, systematically analyze the logic and trends of this round of structural changes in the semiconductor industry.

AI Demand Super Cycle: Why 2027–2028 Remain Key Growth Stages

'Inelastic' Expansion of Capital Expenditure

To understand the persistence of the AI demand super cycle, one must first examine the scale and trajectory of capital expenditures.
Morgan Stanley's data shows that the combined capital expenditures of the five major hyperscale cloud providers (Amazon, Alphabet, Meta, Microsoft, Oracle) are expected to reach approximately $800 billion in 2026, and further climb to $1.2 trillion in 2027. This figure represents an almost doubling of the bank's forecast from one year ago (approximately $450 billion each for 2026 and 2027).

Even more noteworthy is the speed and magnitude of the adjustments. Morgan Stanley pointed out that over the past six months, market consensus estimates for 2026 to 2027 capital expenditures have been revised up cumulatively by over $630 billion. U.S. large hyperscale cloud companies' 2026 capital expenditures are expected to reach $805 billion, nearly double the $433 billion forecast a year ago, double the actual spending in 2025, and triple the 2024 level. Looking ahead, the bank expects this figure to exceed $1.1 trillion in 2027 and approach $1.3 trillion in 2028.

From a broader perspective, Morgan Stanley expects that the cash capital expenditure-to-revenue ratio for hyperscalers will reach 34%, 39%, and 37% in 2026, 2027, and 2028 respectively, fully surpassing the historical peak of approximately 32% during the internet bubble in the late 1990s. It is expected that by 2026, AI-related capital expenditures will exceed 50% of the total capital expenditures of all Russell 1000 index components.

Goldman Sachs' updated forecast released in June 2026 indicates that the four major hyperscale data center operators - Alphabet, Amazon, Microsoft, and Meta - will have total capital expenditures of $725 billion in 2026, up 77% year-on-year from $410 billion in 2025. In just the past six months or so, market expectations for 2026 cloud capital expenditures have risen by nearly 80%. Barclays, meanwhile, expects major cloud players' capital expenditures to reach $919 billion in 2027 and further increase to approximately $1.16 trillion in 2028.

Global Scale of AI Infrastructure Investment

Looking at total global AI spending, Gartner forecasts global AI spending will reach $2.59 trillion in 2026, a 47% increase year-on-year, with AI infrastructure spending rising 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 reach $3.49 trillion in 2027.

In a report released on July 1, 2026, Nomura pointed out that global server market growth forecasts have been raised from 43% to 74% (2026) and 65% (2027), while AI server growth forecasts have been raised from 58% to 78% (2026) and 76% (2027).

Structural Factors Driving AI Demand Persistence

The 'inelastic' nature of AI demand stems from three structural factors.
First, AI investment has dual attributes of both a 'necessity' and a 'highly desired item' - companies are eager to grasp the next generation of core technology while also fearing falling behind in competition. As Morgan Stanley's Chief Cross-Asset Strategist Andrew Sheets noted, for a strategic priority of such importance, whether borrowing costs are 5.50%, 5.75%, or 6.00% has become a secondary consideration.

Second, AI inference and AI agents are becoming new demand engines. Morgan Stanley's surveys show that AI inference, agents, and cloud service growth are driving continuously above-expectation growth in storage demand. The global weekly token usage - a key proxy indicator for computing power demand - has grown by approximately 350% since early January 2026, from about 6 trillion tokens to 28 trillion tokens.

Third, the depth and breadth of financing channels far exceed expectations. In the first five months of 2026, global AI-related bond issuance reached $236 billion, more than quadruple the amount in the same period of 2025. Morgan Stanley expects full-year issuance to exceed $570 billion. Innovation in financing tools - from project finance, layered structures, to residual value guarantees - is adapting at an unprecedented pace to AI-driven capital expenditure needs.

Shift in Semiconductor Pricing Power: From 'Moore's Law' to 'Scarcity Premium'

TSMC's Pricing Power: Data Validation

If the AI demand super cycle is the demand-side logic of the semiconductor industry's structural changes, then TSMC's pricing power is the core variable on the supply side.
TSMC has formally raised its 2026 capital expenditure to a historic high of $52 billion to $56 billion, ramping up 2nm and CoWoS advanced packaging capacity. In terms of capacity layout, the Hsinchu Baoshan and Kaohsiung fabs will be the main production sites for 2nm, with monthly capacity expected to reach 100k wafers by 2026. For the tight 3nm market, TSMC is adopting a dual-track approach of expanding capacity at the Southern Taiwan Science Park and converting equipment, flexibly shifting some 5nm capacity to 3nm to narrow the supply-demand gap.

On the pricing front, TSMC's 3nm process plans another price increase in the second half of 2026, with a maximum hike of 15%, and a further 5% to 10% increase possible in 2027. Moreover, TSMC has been notifying customers that all advanced processes at 7nm and below will see wafer foundry price increases, with an overall adjustment of about 5% to 10%, covering approximately 75% of TSMC's wafer revenue.

Deutsche Bank had already warned in January 2026 that TSMC's 3nm capacity for the entire year of 2026 was fully booked, with orders extending into 2027. Capacity from two 2nm fabs is fully reserved, with monthly capacity of about 35k wafers, expected to expand to 140k wafers by the end of 2026. In this 'seller's market' pattern, price increases are not a negotiation outcome but a reflection of TSMC's unilateral pricing power - clients either accept or exit.

Industrial Logic of the Pricing Power Shift

The formation of TSMC's pricing power is not simply a supply-demand mismatch but a fundamental shift in the logic of the semiconductor industry.
In the past, cycles dominated by consumer electronics followed Moore's Law's cost reduction logic - the more advanced the process, the lower the cost per transistor, leading to a downward price trend. Previously, TSMC's 3nm demand mainly came from smartphone SoCs, supported by a few major clients like Apple. However, with the full-scale launch of the AI server upgrade cycle, multiple cloud service providers, including Nvidia, AMD, Google, and AWS, are accelerating their adoption of 3nm technology. Demand from AI accelerators and custom ASICs is simultaneously flooding in, rapidly increasing wafer starts. The demand structure has shifted from a 'single engine' to a 'multi-engine' model.

More critically, AI chips' reliance on advanced processes is irreplaceable. TSMC holds a 71% share of the global wafer foundry market and over 90% of the 7nm and below advanced process market. In the fourth quarter of 2025, TSMC's global wafer foundry market share reached 70.4%. This highly concentrated supply pattern, combined with the explosive growth of AI demand, is transforming advanced processes from a 'cost center' into a 'profit center'.

In its Greater China semiconductor industry report released on June 30, 2026, Morgan Stanley maintained an 'attractive' rating for the industry, noting that AI semiconductor demand is strong, with long-term demand drivers including chip inflation and the AI cannibalization effect on non-AI semiconductors.

Industry Chain Profit Distribution: Who Is Really Benefiting?

GPU Supply Chain and Profit Concentration

Profits in the AI industry chain are accelerating toward the upstream. Morgan Stanley notes that semiconductor suppliers remain the most clear beneficiaries, with 2026 sales estimates revised up by about 60%.

In the GPU supply chain, Nvidia remains the core supplier of computing hardware. As of July 1, 2026 (Beijing time), Nvidia's stock price was $194.97, with a market capitalization of about $4.72 trillion. But Morgan Stanley analyst Shane Brett notes that the market is entering a cycle phase where returns for semiconductor equipment suppliers begin to rival those of memory chip stocks. Brett revised up semiconductor equipment spending forecasts to $143 billion in 2026 (up from the previous estimate of $136 billion), up 23% year-on-year, and to $182 billion in 2027 (up from the previous $161 billion).

From a broader equipment spending perspective, SEMI's '300mm Fab Outlook Report' released in April 2026 indicated that global 300mm fab equipment spending is expected to 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-on-year to $147.8 billion and $199.5 billion in 2026 and 2027, respectively.

HBM and Memory Chip Scarcity Cycle

In the memory chip sector, HBM is becoming another key profit distribution link in the AI chip supply chain.
TrendForce data shows that the main demand driver for HBM in 2026 comes from AI ASIC capacity upgrades, significantly raising the HBM capacity configured in AI chips from 96/192GB to 216/288GB. In 2027, with the launch of Nvidia's Rubin Ultra platform, the HBM capacity per GPU will further increase to 384GB. TrendForce estimates that HBM's share of total DRAM wafer starts will rise from 18% in 2025 to about 30% in 2027, and HBM bit supply share will increase from 8% to about 13%.

Morgan Stanley predicts that AI-driven memory shortages will last 2 to 3 years. Gartner forecasts that supply tightness will not ease before the end of 2027. At a semiconductor industry executive forum in June 2026, TrendForce revealed that HBM will remain in short supply until 2027, with price increases unavoidable, and 2026 shipment volume expected to grow 60% year-on-year.

Looking at the overall global semiconductor market size, the World Semiconductor Trade Statistics (WSTS) forecasts that the global semiconductor market in 2026 will grow nearly 90% year-on-year to $1.5 trillion, and is expected to grow 26.6% in 2027 to reach a market size of $1.914 trillion.

Risks and Constraints: The Other Side of the Super Cycle

Financing Leverage and Off-Balance-Sheet Risks

The AI super cycle is not without risks. Morgan Stanley, in a series of reports, systematically analyzed the financial risks behind this cycle.
The combined gross leverage ratio of the five major hyperscale cloud providers surged from 0.9 times in the third quarter of 2025 to 1.8 times in 2026, doubling in just two quarters, surpassing the average leverage level of the entire energy industry, and still climbing at a rate of about 0.3 times per quarter. In terms of free cash flow, by 2026, free cash flow for Amazon and Meta will be near zero or even negative, with only Google and Microsoft still maintaining positive values.

More concerning is the off-balance-sheet risk. Morgan Stanley estimates that in addition to book capital expenditures, there are approximately $1.8 trillion in off-balance-sheet commitments, of which purchase commitments account for about $982 billion. The total value of future procurement contracts signed by hyperscale cloud providers with Nvidia is close to $1 trillion. Nvidia's own inventory and purchase obligations have risen to about 32% of its projected fiscal 2027 revenue, above the historical range of 15% to 20%.

Physical Constraints and Return Validation

Beyond financial risks, physical constraints are also worth noting. Power grid access, power generation equipment, skilled labor, and permit approval delays are becoming significant bottlenecks for AI infrastructure construction.
Goldman Sachs, in a research report released in June 2026, noted that U.S. technology investment as a share of GDP has risen to about 4.9%, surpassing the peak of the internet bubble era around 2000. Since November 2022, the market capitalization of AI-related companies has surged by $27 trillion, far exceeding the $9 trillion estimated by macro benchmarks. The market's pricing of AI's future earnings is significantly outpacing the actual realization of productivity dividends.

Conclusion: Structural Revaluation, Not Cyclical Bubble

In summary, the current AI-driven changes in the semiconductor industry are closer to a structural revaluation than a simple cyclical bubble.

On the demand side, AI capital expenditure exhibits clear 'inelastic' characteristics - demonstrating above-expected resilience to prices, financing costs, and geopolitical risks. The combined capital expenditures of the five major hyperscale cloud providers are approximately $800 billion in 2026 and $1.2 trillion in 2027, coupled with global AI spending of $2.59 trillion in 2026 and $3.49 trillion in 2027, pointing to a long-term investment cycle that extends at least to 2028.

On the supply side, TSMC's irreplaceable monopoly in advanced process nodes - with a 71% global wafer foundry market share and over 90% in advanced processes - is translating into systematic pricing power. The 15% price increase for 3nm and the full booking of 2nm capacity are not short-term phenomena but inevitable results of the new paradigm of 'advanced process scarcity + AI demand-driven'.

In terms of industry chain profit distribution, profits are concentrating from downstream application layers to upstream manufacturing and equipment segments. GPU suppliers, wafer foundries, semiconductor equipment makers, and HBM manufacturers are the core beneficiaries of this cycle.

Of course, risks cannot be ignored - rapidly rising leverage ratios, large off-balance-sheet commitments, physical infrastructure constraints, and the tension between market valuation and fundamentals are all potential limiting factors for this super cycle.

But in terms of core logic, the combination of the AI demand super cycle and semiconductor pricing power is reshaping the industry logic that has been driven by cost reduction for the past three decades in the global semiconductor industry. The duration and depth of this transformation may far exceed current mainstream market expectations.

FAQ

Q1: What does the AI super cycle mean?
The AI super cycle refers to a long-term structural growth cycle driven by the large-scale commercial application of artificial intelligence technology, distinguishing it from short-term tech hype or cyclical fluctuations. Its characteristics include: sustained multi-year large-scale capital expenditures, profit concentration upstream in the industry chain, and demand exhibiting 'inelastic' features in response to prices and financing costs. Morgan Stanley expects this cycle to last at least until 2028.

Q2: Why does TSMC have pricing power?
The core source of TSMC's pricing power is the irreplaceability of advanced process nodes. TSMC holds a 71% share of the global wafer foundry market and over 90% of the 7nm and below advanced process market, making it the only manufacturer in the world capable of mass production at 3nm and below nodes. Against the backdrop of explosive AI chip demand, 3nm capacity for the entire year of 2026 is fully booked and extends into 2027. This supply monopoly combined with demand explosion gives TSMC the unilateral pricing ability to raise prices.

Q3: How long can AI infrastructure investment last?
According to forecasts from Morgan Stanley, Goldman Sachs, Barclays, and others, the high-growth phase of AI infrastructure investment will extend at least through 2028. The combined capital expenditures of the five major hyperscale cloud providers are expected to grow from about $800 billion in 2026 to $1.2 trillion in 2027 and nearly $1.3 trillion in 2028. Global AI spending is expected to grow from $2.59 trillion in 2026 to $3.49 trillion in 2027.

Q4: Has the AI chip cycle already peaked?
Current data and institutional forecasts do not support the view that the AI chip cycle has peaked. The global semiconductor market size is expected to reach $1.5 trillion in 2026 and $1.914 trillion in 2027. HBM demand shipment volume is expected to grow 60% year-on-year in 2026, with supply remaining tight until 2027. AI server growth forecasts have been raised from 58% to 78% (2026). However, risk factors such as rising leverage ratios, valuation premiums, and physical constraints warrant continued attention.