The Silicon Shield, “GPU Refineries,” and the New Superpowers — The Second Dawn of Corporate Geopolitics
In recent years, Nvidia has become the symbol of a tectonic shift in the global economy — from a designer of graphics cards for gamers to the central pillar of artificial-intelligence computing. The company’s rise in value has been dizzying: from the first chipmaker to break the one-trillion-dollar mark to a reported five-trillion valuation by the fall of 2025. This surge is not merely investor euphoria; it is an indicator of power moving. Where states once dominated, corporations whose products shape both the economy and foreign policy are increasingly taking the stage. Nvidia has in the process become both an instrument and an independent actor in geopolitics.
A brief history explains today’s dominance. Founded in 1993, the company conquered the PC-gaming world with its series of GeForce graphics cards, but the decisive turning point came in 2006 with the launch of CUDA (Compute Unified Device Architecture — a software platform that allows graphics processors, originally intended for image and video processing, to be used for complex scientific and industrial computations). Nvidia thus transformed the GPU from a gaming accessory into a universal tool for parallel computing, the key technology for developing artificial intelligence. Over time, an entire ecosystem of auxiliary libraries emerged, such as cuDNN (for deep learning) and cuBLAS (for advanced linear algebra), creating strong network effects — the more researchers use Nvidia’s tools, the more useful they become and the harder they are to replace. This phenomenon, known in the industry as the “CUDA moat,” arises from a unique fusion of silicon and software and explains why the company built a long-lasting technological advantage.
The hardware pace was driven by dramatic performance leaps: from the Tesla/V100 accelerator to the A100 (Ampere architecture, 2020) and the H100 (Hopper architecture, 2022), chips with tens of billions of transistors and dedicated tensor cores (specialized units for the mathematical operations needed to train neural networks). In parallel, Nvidia began offering complete systems under the DGX label — turnkey AI servers that connect multiple GPUs into a unified whole. The acquisition of Israeli company Mellanox in 2019 gave it access to InfiniBand — an ultra-fast networking technology that enables thousands of GPUs to be linked into single AI supercomputers (via NVLink and InfiniBand). Although its attempt to buy the British company ARM (owner of the processor architecture powering most mobile devices) failed due to regulatory and geopolitical pressure, Nvidia responded by developing its own Grace processor for data centers, itself based on ARM technology. In short, Nvidia grew from a “chipmaker” into a “full-stack” player — a company that controls the entire layer from silicon to software and delivers complete AI supercomputing clusters ready to use.
Why is its value rising so sharply now? Generative AI exploded in 2022–2023 and triggered unprecedented demand for accelerators. Data-center revenue overtook gaming: in 2023 revenue jumped to around $60.9 billion (more than doubled), and quarterly records fell one after another. Estimates that about 80% of GPUs used to train most modern models come from Nvidia made the company a de facto standard. At a time when every industry — from biotechnology and energy to finance and transportation — is becoming an “AI industry,” Nvidia is the supplier of the essential fuel. The market has rewarded not only current results but also a future in which no competitive economy can emerge without these chips.
But this centrality has catapulted Nvidia from economics into geopolitics. U.S. export controls from 2022 and 2023 formalized chips as strategic goods: top-tier accelerators (A100/H100 and equivalents) are defined by technical thresholds and cannot be exported to China without special permits. The company tried to adapt its products (A800/H800, later H20) with intentionally reduced performance for the Chinese market, but the rules were subsequently tightened. In 2025 the issue reached the highest levels of bilateral diplomacy, and in September the Chinese regulator formally banned the purchase of the H20, effectively breaking the “compromise.” Chips have thus become a tectonic fault line in U.S.–China relations.
China’s response did not stop at statements. Domestic GPUs are being rapidly developed (Huawei, Biren, and major ecosystems within tech giants), and authorities earlier introduced export controls on critical minerals such as gallium and germanium — dry levers over global value chains. In turn, the U.S. and its partners are locking down access to advanced lithography and tools and encouraging reshoring through the CHIPS Act. This mutual pressure opens space for the “middle world”: Global South countries want to avoid choosing between blocs but know that without advanced accelerators they cannot build their own AI ecosystems. Nvidia is thus both an instrument of U.S. policy and a company trying to retain access to the largest growing market in the world.
All this brings us back to the Achilles’ heel — manufacturing. Nvidia is fabless: it designs in the U.S. but relies on TSMC in Taiwan for leading-edge nodes. This means that the fate of key AI chips is tied to the most sensitive point in East Asian geopolitics. Taiwan’s “silicon shield” holds as long as all sides avoid escalation that would wreck global supply chains. Yet even without war, advanced packaging and testing have become bottlenecks, and any epidemic, earthquake, or political shock sends waves through the entire economy. Diversification is underway (TSMC fabs in Arizona and Japan, occasional production at Samsung, potential services from Intel), but critical products are still made where geopolitical risks are highest.
(The “fabless” model refers to companies that design and sell hardware devices and semiconductor chips while outsourcing their physical fabrication to specialized semiconductor foundries.)
The security dimension sharpens the picture further. Top-tier accelerators are dual-use: they power chatbots but also surveillance systems, cryptanalysis, autonomous platforms, and simulations of nuclear and hypersonic systems. States are therefore buying and building AI supercomputers with tens of thousands of GPUs while simultaneously trying to prevent the same hardware from reaching competitors. It is no surprise that black markets and “grey” routes are emerging. When chips become strategic fuel, logistics take on the characteristics of the energy sector: licensing, monitoring, and political conditionality shift from exception to norm.
The economic consequences are ambivalent. On one hand, concentration of capacity in a few companies creates extremes of valuation and dependence — entire industries (healthcare, science, automotive) schedule their plans around GPU deliveries. On the other hand, productivity and innovation genuinely accelerate: cheaper training and inference lower the barrier to entry for startups and the public sector. But the cost of the “AI economy” is material: megawatt-hours for data centers, optical networks, cold-chain logistics for silicon. If the 20th century was ruled by refineries and tankers, the 21st century is developing its own “GPU refineries” — dense computing campuses around which local politics, tax incentives, and diplomacy intertwine.
The geopolitical emphasis remains the same: whoever controls the design and distribution of advanced chips shapes the order of the “Near Future.” The U.S. strategy relies on technological choke points (design and tooling), China responds by accelerating self-sufficiency and mobilizing domestic capital, and Europe seeks autonomy between the two gravitational poles — with its own strengths in tooling and optics. For the rest of the world, timing is crucial: how many years will it take for serious alternatives to appear to the ecosystem Nvidia has built over a decade? While the cost of leaving CUDA is high, political pressures and industrial consortia are already working to reduce it.
In the end, the Nvidia story confirms that the new “super-powers” are not necessarily states. Companies that control key technologies are once again entering the arena of geopolitical competition alongside traditional actors. This echoes the era when corporations were integral to imperialism — now reappearing in a new form, in a new dawn of corporate geopolitics. Profit remains their primary mission, but at a certain scale elements of ideology, industrial sovereignty, and foreign-policy gameplay emerge: standard-setting, high-level negotiations, and control of market access. The battle for chips and AI — the battle for the “Near Future” — is fought simultaneously in cabinets and laboratories, in regulations and in microcode. Nvidia, whether it wants to be or not, participates in both worlds. And as long as computing power is treated as a strategic resource, companies of its kind will not be merely market barometers but instruments — and even creators — of the new geopolitics.
About the Author
