NVIDIA invests $150 billion a year in Taiwan: the epicenter of the AI revolution

NVIDIA invests 150 billion dollars per year in Taiwan: the epicenter of the AI revolution

🔎 150 billion, a figure that changes everything

On May 27, 2026, Jensen Huang took the stage in Taipei and dropped a figure that shook global tech geopolitics: 150 billion dollars per year invested in Taiwan. Not over ten years. Per year.

This is ten times more than four or five years ago, when NVIDIA was spending between 10 and 15 billion annually there, as recalled by BFMTV. A dizzying leap that transforms Taiwan into the sole center of gravity of the global AI industry.

Why now? Because the demand for compute is exploding, TSMC is doubling its production capacity, and Sino-American tensions make every chip a strategic issue. As we already analyzed in our article on NVIDIA's 40 billion dollars in 2026, Jensen Huang's plan for domination exceeds anything we had anticipated.

The essentials

• NVIDIA is going from 10-15 billion to 150 billion dollars per year in investment in Taiwan, a tenfold increase in four years.
• A new 4-hectare campus in Taipei will open in 2030, with 4,000 employees dedicated to AI R&D.
• Jensen Huang calls Taiwan an « epicenter of excellence » and the center of the AI revolution, according to Republic World.
• The Taiex index reached a historical record (+1.7%) following the announcement, according to LinkedIn News.
• China refuses to buy American chips under the 25% tax imposed by Trump, widening the technological gap, according to Ars Technica.
• AMD also plans 10 billion dollars in Taiwanese investment for its AI chips, according to Frandroid.

Recommended tools

From 10 to 150 billion: the scale of the shift

Jensen Huang doesn't mince his words. At the ceremony at NVIDIA's headquarters in Taipei, he stated that Taiwan was not simply a partner, but the "epicenter of excellence" of the entire AI revolution.

The jump from 10-15 billion to 150 billion per year is not just a budget increase. It's an industrial paradigm shift. According to Chosun Biz (Reuters), this massive investment is accompanied by an unprecedented deepening of the partnership with TSMC.

NVIDIA's net profit for the first quarter of 2026 amounts to $58.3 billion, reports BFMTV. In other words, NVIDIA is reinvesting the equivalent of several quarters of profits into a single territory.

It's an existential bet: the company is going all in on the island that manufactures its chips, at the risk of concentrating its geopolitical vulnerability on a single point.

The financial context making this possible

NVIDIA has never been richer. The demand for GPUs for training models like GPT-5.5, Claude Opus 4.7, or Gemini 3 Pro Deep Think far exceeds the available supply.

Each generation of models requires more compute. Agentic evaluation scores are skyrocketing: GPT-5.5 reaches 98.2, Gemini 3 Pro Deep Think 95.4, Claude Opus 4.7 (Adaptive) 94.3. Behind these numbers lie thousands of GPUs running day and night in Taiwanese datacenters.

Jensen Huang implicitly confirmed this to DigiTimes: massive reinvestments by OpenAI and NVIDIA's other major customers are directly fueling this virtuous cycle. More compute, better models. Better models, more demand. More demand, more investment in Taiwan.

The new Taipei campus: 4 hectares for 4,000 brains

The investment isn't just abstract. NVIDIA is building a physical campus in Taipei, spanning 4 hectares, with an opening scheduled for 2030.

According to Presse-Citron, this site will host 4,000 employees. It won't be just a simple administrative office. It is a leading R&D center, integrated into the heart of the Taiwanese ecosystem.

Why a campus and not just orders to TSMC

Physical proximity between NVIDIA engineers and TSMC teams is a decisive competitive advantage. Advanced chip design requires constant iterations between the fabless company (NVIDIA) and the foundry (TSMC).

When you are etching at 2nm or in angstroms, every nanometer counts. A face-to-face dialogue between engineers can solve in a single day a problem that would take weeks via video conference.

This campus also crystallizes the talent retention strategy. Taiwan trains the best semiconductor engineers in the world. By establishing a physical presence, NVIDIA ensures it doesn't lose them to TSMC, MediaTek, or AMD.

Taiwan, the 36,197 km² island that holds global AI

The headline from Presse-Citron sums it all up: the future of AI is being played out on this island smaller than Switzerland.

TSMC manufactures over 90% of the world's advanced chips. Without Taiwan, no NVIDIA GPUs, no Apple chips, no AMD processors. The entire global tech industry rests on a 36,197 km² territory located 160 kilometers off the Chinese coast.

Why Taiwan remains irreplaceable

Diversification attempts have failed or remain marginal. Intel is struggling with its advanced processes. Samsung is finding it hard to compete on yields. TSMC's factories in Arizona are producing, but at a higher cost and with lower yields.

As Sify points out, the AI boom is now dependent on a single island. It is a single point of failure on a global scale.

The Taiwanese ecosystem is not limited to TSMC. It includes ASE for packaging, MediaTek for SoCs, and a perfectly honed supply chain of chemicals and equipment. Recreating this ecosystem elsewhere would take a decade, at a minimum.

The Taiwan effect on financial markets

NVIDIA's announcement propelled the Taiex index to a historic record, with a 1.7% rise in a single session, according to LinkedIn News. The market is validating what engineers have known for a long time: Taiwan is the beating heart of AI.

This direct correlation between a single company's announcement and the national stock index shows the total interdependence between NVIDIA and Taiwan. The fate of one is inseparable from the other.

Geopolitics: the dangerous game between Beijing, Washington, and Taipei

The $150 billion investment is not just a corporate decision. It is a major geopolitical act.

The Trumpian contradiction

Ars Technica raises the embarrassing question for the Trump administration: how to reconcile the promise of making the United States the global hub of AI with the NVIDIA announcement that cements Taiwan as the center of the revolution?

Trump's plan to bring chip manufacturing back to American soil is stalling. The CHIPS Act subsidies are slow to produce their effects. TSMC Arizona is failing to match Taiwan's yields. And meanwhile, NVIDIA is announcing $150 billion a year... in Taiwan.

It is a silent but resounding slap in the face for American industrial policy. Jensen Huang has done his math: manufacturing in the United States is not competitive, and he has no intention of sacrificing his margins for a patriotic policy.

China and the refusal of the 25% tax

China responded to the 25% tax imposed by Trump on American chips by simply refusing to buy them, reports Ars Technica. This trade escalation accelerates the technological bifurcation between the two superpowers.

Beijing is pushing its companies to turn to local alternatives. It is in this context that Moonshot AI raised $2 billion and Kimi K2.6 dominates the open-weight, marking the Chinese acceleration in the race for autonomous AI.

The risk of chip smuggling

Firstpost raises an often-overlooked point: the growing concerns around the smuggling of AI chips to China. With $150 billion worth of chips transiting through Taiwan, the volume makes customs control extremely complex.

Latest-generation NVIDIA chips end up in the hands of Chinese companies via diverted circuits (Hong Kong, Singapore, shell companies). The higher the volume, the more the loopholes multiply. It is a security headache that will worsen with the scale of the announced investments.

Competition: BYD, Samsung, Intel — who can compete?

NVIDIA is not the only one betting on Taiwan, but nobody is playing in the same league.

AMD and its 10 billion Taiwanese

Frandroid points out that AMD also plans $10 billion in investment in Taiwan for its AI chips. This is significant, but it is fifteen times less than NVIDIA.

AMD remains the credible challenger, with its Instinct chips equipping some of the world's largest supercomputers. But the investment gap shows that the compute battle is also playing out on the financial and industrial front, not just the technological one.

Samsung and the Korean strategy

Samsung is investing massively in its semiconductor factories in South Korea, seeking to reduce global dependence on TSMC. But the results in terms of yield and performance remain below expectations.

South Korea has the ecosystem, the workforce, and the political will. What it lacks is TSMC's ultra-fine specialization in advanced processes. Samsung tries to do everything (memory, logic, foundry), whereas TSMC focused on a single business and perfected it.

Intel: the wounded giant

Intel embodies the failure of the American attempt to reconquer advanced manufacturing. Despite tens of billions in federal subsidies, the company is struggling to catch up with TSMC.

The balance of power is unequivocal: when NVIDIA announces $150 billion in Taiwan, it is also a message sent to Intel. The American fabless does not believe in Intel's ability to deliver TSMC-level chips within a relevant timeframe.

BYD and the Chinese ecosystem

BYD, the Chinese electric vehicle giant, is also investing massively in chips, but with a logic of Chinese national autonomy. Beijing's goal is clear: replace American chips with local alternatives.

However, without access to ASML's EUV lithography (blocked by Western sanctions), China remains confined to less advanced manufacturing nodes. The technological gap persists, even if models like DeepSeek V4 Pro (Max) with a general score of 88 show that software engineering can partially compensate for the hardware deficit.

What this means for compute and GPUs

The $150 billion investment will concretely transform the compute market in the coming years.

The memory shortage: the hidden bottleneck

Jensen Huang issued warnings about the memory shortage during his speech in Taipei, reports DigiTimes. GPUs are useless without sufficient HBM memory to feed them data.

This is a crucial point that is often underestimated. TSMC can double its manufacturing capacity, but if Samsung and SK Hynix do not follow suit on HBM memory, GPUs remain empty shells. NVIDIA's investment in Taiwan therefore includes strengthened partnerships with the memory supply chain, not just with TSMC.

TSMC doubles its capacity

Jensen Huang's projection includes a doubling of TSMC's capacity, again according to DigiTimes. This means more available wafers, potentially stabilizing prices, and shorter delivery times.

For companies deploying AI, this is good news. Compute costs, which exploded between 2024 and 2026, could finally plateau. The $10 billion enterprise JVs launched by Anthropic and OpenAI to deploy AI in SMBs and large corporations depend directly on this increased availability.

Impact on cloud and API pricing

More available compute means downward pressure on cloud service pricing. AWS, Azure, and Google Cloud all have an interest in seeing GPU prices drop to accelerate enterprise AI adoption.

The most compute-hungry models — GPT-5.5, Claude Opus 4.7 (Adaptive), Gemini 3 Pro Deep Think — could become accessible to a greater number of businesses. The virtuous circle expands: more compute, better models, more adoption, more revenue to reinvest.

TSMC at the center: the partner you can't bypass

NVIDIA's investment consolidates TSMC's position as the natural monopoly of advanced chip manufacturing.

A deepening partnership

Chosun Biz (Reuters) emphasizes the deepening of the NVIDIA-TSMC partnership. This is not a classic client-supplier relationship. It is a co-development where both companies share strategic industrial secrets.

NVIDIA designs the GPU architecture. TSMC provides the process innovations that make it feasible. This symbiosis is impossible to quickly replicate with another foundry.

The monopoly risk and regulators

A $150 billion investment that reinforces a de facto monopoly will attract the attention of regulators. In the US, Europe, and China, antitrust authorities are closely monitoring the concentration of the AI supply chain.

But what can they do? Banning NVIDIA from buying chips from TSMC would amount to destroying the American AI industry. Forcing TSMC to diversify its client base when NVIDIA is its largest client makes no economic sense. The monopoly is structural, not contractual.

The impact on the Chinese AI race

150 billion dollars a year in Taiwan is also a clear message sent to Beijing: the West is not going to ease the technological pressure.

The Chinese acceleration in response

Faced with this display of industrial might, China is accelerating. Moonshot AI and its 2 billion dollars show that Chinese investors are betting on software solutions to circumvent the hardware handicap.

Kimi K2.6 reaches 88.1 in agentic and 84 in general, remarkable scores for an open-weight model that does not benefit from NVIDIA infrastructure. The Chinese strategy is clear: do more with less compute, thanks to more efficient architectures.

The risk of a two-speed AI

The massive investment in Taiwan is widening the gap between the NVIDIA ecosystem and the rest of the world. On one side, models like GPT-5.5 (98.2 in agentic) trained on thousands of latest-generation GPUs. On the other, Chinese models optimized to run on previous-generation hardware.

In the short term, the performance gap persists. In the medium term, Chinese algorithmic efficiency could make up for some of the hardware lag. In the long term, everything depends on the geopolitical evolution around Taiwan.

❌ Common mistakes

Mistake 1: Confusing total investment with annual investment

150 billion is not a ten-year commitment. It is per year. Many commentators reported the figure, presenting it as a total amount, thereby minimizing the true scale of the announcement. Always check the unit of time in NVIDIA's press releases.

Mistake 2: Thinking this investment makes Taiwan safer

It's the opposite. The more global AI depends on Taiwan, the more the island becomes a strategic target. The theory of economic deterrence (China would not attack Taiwan because it would destroy the global economy) is reassuring but unproven. The 150 billion increases the stakes, not the security.

Mistake 3: Believing that TSMC's American factories will replace Taiwan

TSMC Arizona exists, but its yields remain lower and its costs higher. The 20-30% gap in cost per chip makes American factories complementary, not substitutable. NVIDIA's investment implicitly confirms this: if Arizona were competitive, the 150 billion would go there.

Mistake 4: Ignoring the memory risk

Focusing on TSMC and GPUs without considering the HBM memory chain is a mistake. Without SK Hynix and Samsung for memory, NVIDIA GPUs are bricks without cement. The memory shortage mentioned by Jensen Huang is the real hidden bottleneck.

❓ Frequently Asked Questions

Why doesn't NVIDIA build its own factories?

NVIDIA is a fabless company (without factories). Its model is based on design, not manufacturing. Building an advanced semiconductor plant costs 20-30 billion and requires 5-7 years of development. Betting on TSMC remains infinitely more rational.

Is 150 billion per year realistic?

It is a multi-year commitment, not a budget allocated all at once. NVIDIA smooths this investment over several years, depending on actual demand. The figure is ambitious but consistent with the company's growth trajectory and TSMC's capacity doubling projections.

What happens if China invades Taiwan?

Catastrophic scenario for the global industry. 90%+ of advanced chip production would be interrupted. The consequences would affect all sectors: AI, smartphones, automotive, defense. This is precisely why the 150 billion investment makes the situation more tense, not less.

Will GPU prices drop thanks to this investment?

Ultimately, yes. More production capacity means more supply, which should exert downward pressure on prices. But in the short term, demand is growing faster than supply. Don't expect a significant drop before 2028-2029.

How can French companies take advantage of this?

By sourcing through cloud providers (AWS, Azure, Google Cloud, OVH) that buy the NVIDIA GPUs produced in Taiwan. The investment does not change the distribution chain, but it ensures that GPUs will be available in larger quantities. To host your AI projects, solutions like Hostinger offer a good value for money for light deployments.

✅ Conclusion

By announcing $150 billion a year to Taiwan, Jensen Huang made official what everyone knew but didn't dare say: the AI revolution is Taiwanese or it isn't. America designs the chips, but it's a 36,000 km² island that manufactures them. This geopolitical paradox is the blind spot of the entire Western tech strategy, and NVIDIA just invested the equivalent of a small country's GDP into it. Per year.