China's Chip Bottleneck Is Printing Money — But the Capacity Cure Is Already Being Built

China's chip foundry bottleneck is quietly printing money — and the clock on that scarcity is already ticking.

In the second quarter of 2026, a company most U.S. investors have never heard of hit $3 billion in quarterly revenue for the first time. Its net profit tripled year over year to $479 million. It raised prices on customers who had no place else to go. Its factories ran at 93.7% utilization, and management guided that margin would expand further next quarter.

The company is Semiconductor Manufacturing International Corporation, known as SMIC. It is the largest foundry in China and the third-largest globally. And it is sitting at a bottleneck that has nothing to do with whether you believe in artificial intelligence — and everything to do with whether the Chinese economy can build the next generation of AI infrastructure at all.

The reason this matters is simple: SMIC is the physical gate through which China's AI ambitions must pass. Every data center, every AI server, every companion chip supporting those systems needs semiconductors manufactured. U.S. export restrictions have shut Chinese technology companies off from the world's best tools and foundries. That leaves SMIC and a handful of smaller players as the only domestic option. When demand outstrips supply, someone captures the scarcity rent. SMIC just proved who it is.

What happened to the numbers

SMIC's second-quarter results, released in mid-August 2026, showed three things moving in the same direction: volume, price, and utilization.

Revenue reached $3.006 billion, up 36% year over year and 20% from the prior quarter. That growth came from two sources. Wafer shipments — the actual volume of chips produced — rose 14.4% sequentially. Average selling prices per wafer rose 5.7%. Revenue grew faster than volume because SMIC had pricing power.

Gross margin jumped 5.2 percentage points sequentially from 20.1% to 25.3%, well above the company's own guidance range of 20% to 22%. Management attributed the margin expansion to higher prices, a more favorable product mix, and operating at near-full capacity. Operating profit more than doubled quarter over quarter to $534 million.

The company shipped 2.869 million 8-inch equivalent wafers. Total monthly production capacity stood at about 1.097 million 8-inch equivalent wafers. New pricing structures announced in Q2 were already flowing into Q3.

This is what a bottleneck supplier looks like in financial statements. The factory can't run faster. The customers need more than it can produce. Prices go up. Margins expand.

Why Chinese customers have nowhere else to go

To understand why SMIC has pricing power, you need to understand what the U.S. export restrictions actually do.

Since 2022, successive rounds of American controls have blocked China from purchasing the most advanced semiconductor manufacturing equipment from ASML, limited exports of cutting-edge chips, and restricted technology transfer. The goal was to slow China's AI and military computing advancement. The unintended consequence was to create an enormous captive market.

Chinese technology companies — Baidu, Alibaba, Tencent, Huawei, and a wave of AI startups — still want to build AI infrastructure. They can't buy Nvidia GPUs in large numbers. They can't send designs to TSMC for advanced manufacturing. So they buy from Huawei's domestically produced Ascend chips and route manufacturing through SMIC. And those AI systems need supporting chips: power-management ICs, controllers, connectivity devices, optical transceiver components, mixed-signal products. All of them are manufactured on foundry capacity.

SMIC's co-CEO Zhao Haijun told investors that Chinese internet companies had increased hardware investment "beyond expectations," with direct demand and supporting chips remaining "in short supply." Revenue from AI-supporting products grew roughly 40% in absolute dollar terms during the April–June quarter.

Being necessary gets you orders. Being scarce decides who keeps the money.

90.2% of SMIC's revenue comes from Chinese customers. This is the purest possible exposure to the theme — and the largest single risk if the theme turns.

The hidden cost: depreciation is eating the margin gains

The bottleneck story is real. But there's a counterforce building that most earnings headlines don't mention.

SMIC spent $8.1 billion on capital expenditures in 2025, up 10.5% from 2024 and above the company's own guidance. Management expects 2026 capex to remain at the same level. That spending is buying new fabrication equipment to expand capacity by roughly 40,000 12-inch equivalent wafers per month by year-end — on top of 50,000 added in 2025.

Here's the accounting reality: new capacity means new depreciation. SMIC's co-CEO projected that depreciation charges will increase approximately 30% in 2026. Manufacturing-related depreciation and amortization already rose 23.1% year over year in Q2 to about $925 million. Zhao explicitly warned that high capital spending placed considerable depreciation pressure on gross profit margins.

In foundry economics, depreciation is a massive fixed cost. At 93.7% utilization, SMIC can spread that cost across nearly all available wafers. At 75% utilization, every idle wafer of capacity eats into margin. This is why utilization matters more than revenue: the bottleneck protects margin only as long as the factory stays full.

SMIC expects utilization to remain in the mid-90% range in Q3, and guides gross margin to improve further to 26% to 28%. But new capacity is coming online. More depreciation is inevitable. The question is whether demand growth can keep pace with capacity expansion — and whether customers will still accept higher prices once supply tightens less.

The capacity cure is not years away

Foundry capacity has a predictable life cycle. The constraint today is real, but it is not permanent.

SMIC's announced expansion plan is aggressive: 40,000 additional monthly wafers by end-2026, on top of the 50,000 added in 2025. That's nearly 10% new capacity in two years. Hua Hong, China's second-largest foundry, also reported record Q2 revenue of $717.5 million and profit growth of 385.9% year over year, with its own expansion underway.

There's also a structural mismatch to watch. The equipment SMIC has purchased may not translate into full production capacity this year. Zhao acknowledged in February that while critical equipment was pre-purchased, ancillary equipment remains pending. New fabs require validation, qualification, yield ramp, and customer acceptance before they produce revenue. The timeline from equipment installation to qualified output is typically 6 to 18 months.

Beyond the foundries themselves, the broader concern is overcapacity in mature nodes — the 28-nanometer and larger process nodes where most of SMIC's production lives. China has been aggressively expanding mature-node capacity as part of its self-sufficiency drive. The U.S. and EU have already flagged this as a competitive threat. If the government's push to replace imported chips succeeds, the captive market that gives SMIC pricing power could become a supply glut.

The bottleneck half-life here is probably two to three years: enough time for the announced capacity to come online, qualify, and meet demand — but not so long that pricing power is guaranteed through the entire period.

Valuation already prices in a long scarcity story

SMIC trades on the Hong Kong Stock Exchange under ticker 0981.HK and as an ADR on the OTC market under SMI. As of late August 2026, its market capitalization stood around HK$764 billion — roughly $98 billion. Its trailing price-to-earnings ratio was approximately 99 times, while the forward P/E was about 56.

For comparison, TSMC — the world's largest and most profitable foundry, operating at the most advanced nodes — trades at roughly 30 times trailing earnings. TSMC has better technology, better margins (typically 50%+ gross margin), global customers, and access to the best equipment. SMIC is priced at more than three times TSMC's multiple despite lower margins, less technology, and a geographically concentrated customer base.

That multiple tells you what investors believe: they expect SMIC's earnings to grow dramatically for an extended period, and they are willing to pay a premium for monopoly-like positioning in a $1 trillion domestic market. The forward P/E of 56 implies the market expects roughly 2x earnings growth over the next 12 months relative to current levels.

Whether that expectation is justified depends on whether the bottleneck holds. If SMIC maintains high utilization and pricing power through 2027, the story compounds. If capacity expansion and depreciation compress margins faster than expected, the multiple looks very expensive.

What to watch next

This is not a thesis to chase or dismiss. It is a bottleneck with a timer, and the timer is visible in the company's own disclosures.

The confirmation metric is utilization. SMIC guides mid-90% utilization in Q3. If the Q3 report shows utilization holding at 94% or above while new capacity comes online, the bottleneck is still binding. If utilization drops toward 85%, the pricing power cracks.

The margin number carries the depreciation signal. SMIC guides 26% to 28% gross margin for Q3. Achieving the top end of that range would prove that price increases and volume are more than absorbing the 30% depreciation increase. Falling below 25% would suggest the expansion is eating into the rent faster than expected.

The risk that doesn't show up in a single metric is the geopolitical one. Any tightening of U.S. restrictions on equipment shipments to China could delay SMIC's capacity expansion — keeping the bottleneck tighter for longer but also limiting its ability to serve demand. Loosening those restrictions would have the opposite effect: more equipment, faster capacity, lower scarcity. The political variable is not priced in either direction.

SMIC is a real bottleneck supplier with documented pricing power, triple-digit profit growth, and a customer base that has nowhere else to manufacture. It is also a company spending $8 billion a year to eliminate its own scarcity, trading at nearly 100 times current earnings, and deriving 90% of revenue from one country's technology cycle. The bottleneck is real. The clock is visible. The question is whether the market is paying for five years of scarcity when the capacity cure may arrive in two.