The New Chip Cold War: How the US-China Dispute Is Redrawing the Global AI Map in 2026

In April 2026, Jensen Huang made a statement that few CEOs have the honesty — or the courage — to make in public.

"In China, we have now dropped to zero. Conceding an entire market the size of China probably doesn't make a lot of strategic sense."

— Jensen Huang, CEO of Nvidia (Special Competitive Studies Project, April/2026)

The Nvidia CEO was not speculating. He was stating a fait accompli. In two years, the company lost 100% of the Chinese market for AI accelerators. It was ~95% in 2024. Zero in 2026 (PC Gamer). A drop of $17 billion in annual revenue, evaporated by decree.

What seemed like a geopolitical chess move by the US — cutting off the supply of advanced chips to China through successive rounds of sanctions — turned into a domino effect that is redrawing the global map of the semiconductor industry. And the side effects are as profound as the initial blow.

On one side, the forced rise of Huawei as an AI chip manufacturer. On the other, TSMC consolidating a dominance that borders on monopoly. In the middle, South Korea profiting billions from a component no one paid attention to until 2024: HBM memory. And turning the crank on everything, the American big techs — which will spend $715 billion in Capex this year, most of it on AI infrastructure.

This is the complete X-ray of the numbers, strategies, and consequences of the new chip cold war.

The Backfired Shot: The Effect of Sanctions on the Numbers

Washington's original calculation was straightforward: deny China access to advanced Nvidia and AMD chips, and thereby slow China's technological advance in AI. The result was the opposite of what was planned.

Sources: PC Gamer, Officechai, The AI Track, Abhs.in

The data is striking. China's self-sufficiency in AI chips jumped from 20% in 2023 to 41% in 2026, according to Morgan Stanley, and is expected to reach 85% by 2030 (Officechai). Huawei, which barely appeared in the AI accelerator market until 2024, is expected to earn $12 billion from its Ascend 950PR chips in 2026 — that's 750,000 units (The AI Track).

The biggest thermometer of this movement came in May: ByteDance, owner of TikTok, placed a $5.6 billion order for Huawei Ascend 950PR chips. It is the largest individual order for domestic AI chips in China's history (Faq.com.tw). For comparison, it is more than Nvidia earned from the entire Chinese market per quarter before the sanctions.

Nvidia's own annual report (10-K, fiscal year 2026) acknowledges the own goal:

"Our effective foreclosure from the China market helped our competitors build larger developer and customer ecosystems to challenge us worldwide."

SMIC, China's main manufacturer, already produces chips at 7nm using DUV lithography — a technology that bypasses sanctions because it does not depend on ASML's EUV machine, which has its export to China prohibited. And it is developing the 5nm process, with an estimated yield of ~20% (Abhs.in). It is not efficient, but it is enough to keep the Chinese innovation machine running while the country races to reduce external dependence.

And it doesn't stop there. An April 2026 report from the American Enterprise Institute (AEI) made an uncomfortable revelation for Washington:

"TSMC's DUVi fleet will almost certainly be capable of producing enough near-frontier dies in 2026 to supply Huawei's domestic AI accelerator demand."

In other words, TSMC itself, with its DUVi lithography fleet, may be enabling the production of Huawei's chips — Nvidia's main competitor that the sanctions aimed to eliminate. The interconnection of the global semiconductor supply chain makes it impossible to isolate one country without affecting all others.

Taiwan at the Center of Everything: TSMC's Absolute Dominance

While China races to become self-sufficient, Taiwan remains the gravitational epicenter of the global chip industry. And the numbers are relentless.

TSMC controls 92% of the most advanced chip market (3 nanometers and below) and 72% of the global foundry market (Closelook.net). There is no Plan B for those who need the most modern semiconductors — TSMC is the only game in town.

The most revealing data point of the American strategy came in May 2026: TSMC approved an additional $20 billion in investments in Arizona, raising the total committed in the United States to $165 billion, spread across 11 fabs (SEC.gov). The US government is literally paying to bring TSMC closer — a tacit acknowledgment that relying on Taiwan for 92% of the world's most advanced chips is an unsustainable strategic risk.

But there is a bottleneck that no amount of money can solve in the short term: CoWoS (Chip-on-Wafer-on-Substrate) packaging, an essential technology for high-performance AI accelerators, is simply sold out. C.C. Wei, chairman and CEO of TSMC, has already warned in an earnings call:

"CoWoS capacity remains sold out through 2025 and into 2026."

This means that any company wanting to produce AI accelerators at scale — whether Nvidia, AMD, Huawei, or Google — must compete for space in TSMC's queue. And the queue is full.

The Forgotten Pole: South Korea and the HBM Gold Mine

There is a chapter in this story that the American media covers little, but it is crucial to understanding the complete chessboard: South Korea's absolute dominance in the high-bandwidth memory (HBM) market.

These chips are the circulatory system of modern AI. Without HBM, there is no way to feed data fast enough to keep accelerators busy. It is a bottleneck as critical as processing power.

Source: Silicon Analysts

SK Hynix recorded an operating margin of 72% in the first quarter of 2026 (Silicon Analysts) — a number that rivals Nvidia's margins in the best days of the GPU pandemic. The company simply dominates the HBM3 and HBM4 market, and there is no sign it will lose its leadership anytime soon.

Together, Samsung and SK Hynix are expected to generate more than 200 trillion South Korean won (approximately $145 billion) in operating profit in 2026 (Chosun). This is a number that places South Korea as the undisputed third pole on the board — alongside Taiwan and the United States.

The Capex Machine: $715 Billion Fueling the Fire

Behind all these numbers, there is a driving force that explains why demand for AI chips shows no signs of cooling: big techs are spending like never before.

Amazon, Microsoft, Alphabet (Google), and Meta are expected to invest $715 billion in Capex in 2026, according to The Intel Briefing. About 75% of that total goes directly to AI infrastructure — data centers, GPUs, interconnections, cooling systems, and yes: many, many chips.

The most extreme example of this race is Meta's Hyperion in Louisiana. The AI data center campus has a total cost exceeding $200 billion (The Next Web). It is the largest private infrastructure project ever conceived in history, and it exists for one exclusive reason: AI requires computing power in quantities humanity has never needed before.

This tsunami of investment has a direct side effect: any disruption in the global semiconductor supply chain — tougher sanctions, an earthquake in Taiwan, military tensions in the Taiwan Strait — can paralyze not just companies, but entire economies. The world is betting too many chips on too few suppliers concentrated in high geopolitical risk regions.

And Brazil in This Story?

For the Brazilian reader, the inevitable question is: what does this have to do with us? The short answer: everything.

Brazil imports practically 100% of the semiconductors it consumes. It does not produce a single advanced AI chip. It has no foundry, no HBM, nothing. When the sanctions war between the US and China tightens, global chip prices skyrocket — and Brazil, which has no bargaining power or scale, ends up at the bottom of the supplier priority list.

Countries like ours are hostages of a global supply chain that is being reconfigured without our participation. While China races toward 85% self-sufficiency, the US subsidizes TSMC with $165 billion, and South Korea profits $145 billion from HBM, Brazil still does not have a national semiconductor strategy for AI even outlined. The R$23 billion announced by the government in 2026 for industrial innovation, although relevant, barely touches the subject.

The War Map in May 2026

The overall picture is one of a deep and accelerated reconfiguration of the global semiconductor map — with implications that go far beyond technology.

• US sanctions did not stop China's advance. They only redirected it. China went from 20% to 41% self-sufficiency in AI chips in three years, and Huawei became a relevant player where it was once irrelevant.
• TSMC remains the most powerful player at the table. With 92% of advanced chips and 72% of the foundry market, Taiwan concentrates a geopolitical risk that the world still does not know how to mitigate.
• South Korea has emerged as a silent power. HBM has become the oil of AI, and SK Hynix is pumping it with 72% margins.
• Big techs are betting $715 billion. The combined Capex of Amazon, Microsoft, Google, and Meta in 2026 is equivalent to the GDP of countries like Argentina or Sweden.
• The global supply chain is more fragile than ever. Over-reliance on Taiwan and geographic concentration of critical technology are vulnerabilities that governments are beginning to address — too late.

The question that remains for the coming years is simple and frightening: how many players can fit on this board? Because one thing the 2026 data already shows clearly — this is not a war that will have many winners. And the losers may not only be the countries left out, but an entire global innovation ecosystem that depends on an increasingly fragmented supply chain.

Related Articles

Also check out: The New World Order of AI in 2026: Who Regulates, Who Invests, and Who Gets Left Behind Also check out: The Map of AI in the World in May 2026: EU Retreats, Malta Innovates, Google Creates a New Mouse, and the US-China Race Reaches Boiling Point Also check out: South Korea vs. AI Chaos: The Law That Could Change the Game in Asia and Affect Brazilian Companies