By Ajay Raju
In late January 2025, a Chinese AI startup named DeepSeek sent shockwaves through global markets with a simple but powerful demonstration: its R1 model matched the reasoning capabilities of ChatGPT-o1 and approached the performance of OpenAI’s GPT-4o—but at a fraction of the cost. Within just seven days of launch, DeepSeek’s mobile app topped the U.S. iOS App Store’s download charts, overtaking ChatGPT for the first time since 2022.
The market reaction was immediate and historic. Nvidia shares plunged by 9%, the steepest decline since March 2020, erasing nearly $300 billion in market capitalization. It was the largest single-day loss for any U.S. company in market history. This caused a broader tech sell-off: the Nasdaq fell 3%, while the Philadelphia Semiconductor Index dropped 9%. However, sentiment quickly rebounded the following day, with Nvidia stock climbing about 8.9%, as analysts framed the plunge as an overreaction and buying opportunity. Within weeks, the panic evolved into a more complex understanding: DeepSeek’s rise, while disruptive, ultimately highlighted the structural advantages that continue to make companies like Nvidia indispensable to the global AI ecosystem.
NVIDIA’s H20 Chips
The DeepSeek crisis exposed a critical vulnerability in U.S. semiconductor strategy. The H20 chip is widely believed to have contributed to DeepSeek, a revelation that underscored how export-controlled chips were still reaching Chinese developers through complex supply chains and intermediaries.
This reality prompted dramatic policy reversals. In April 2025, the Trump administration imposed strict licensing requirements on NVIDIA’s H20 chips—export-compliant processors specifically designed for the Chinese market. NVIDIA said in May that it missed out on $2.5 billion in additional revenue that would have come from H20 sales to China during the first quarter of 2025 and expected another $8 billion revenue loss during the second quarter. The regulatory back-and-forth began in April when the Trump administration restricted H20 sales, potentially costing Nvidia $15 billion to $16 billion in revenue.
But by July 2025, geopolitical pragmatism prevailed over ideological restrictions. U.S. government officials told Nvidia they would green-light export licenses for its H20 artificial intelligence accelerator, the company confirmed in a recent blog post — a move that may add billions to Nvidia’s revenue this year. The policy reversal was reportedly tied to broader trade negotiations, with Commerce Secretary Howard Lutnick suggesting that NVIDIA’s anticipated return to selling H20 chips to China is part of bigger U.S. talks over rare earth minerals.
Again, the market responded immediately to this news. Nvidia stock climbed and rallied, leading the market, demonstrating how closely semiconductor geopolitics and market valuations have become intertwined.
Ajay Raju: Predicting future economic success of cities (June 26, 2025)
Why did DeepSeek create a market panic?
In today’s AI-driven world, semiconductors are more than technology—they’re strategic levers of national power. Nvidia’s dominance in AI accelerators represents not just commercial success, but America’s technological sovereignty. Its chips, particularly the H100, A100, and H20 models, power everything from large language model training to national defense systems.
Nvidia’s advantage isn’t just in hardware—it’s in its full-stack ecosystem. The company’s CUDA software platform, refined over nearly two decades, has become the backbone of modern AI development. It’s a closed-loop system where developers, researchers, and companies are locked into an architecture few competitors can match.
This ecosystem fuels a virtuous cycle of innovation: high-margin chips support massive R&D investment (over $7 billion annually), which attracts global talent and drives deeper software integration. Nvidia’s success creates ripple effects throughout the U.S. economy—supporting jobs, tax revenue, and cutting-edge research.
What made DeepSeek’s success so alarming was not just its technical competency, but its cost structure. The company claims it trained its R1 model for under $6 million using less compute, leveraging efficiency optimization—a dramatic contrast to the $600 million reportedly spent on GPT-4 and $1 billion+ on GPT-4o. Perhaps more striking was the speed of adoption. DeepSeek’s app soared to the top of the U.S. App Store’s free downloads, unseating ChatGPT for the first time in years—an unmistakable signal that global audiences were open to non-Western AI tools. This raised existential questions for the U.S. AI ecosystem: if Chinese firms could achieve 90% of the performance at 1% of the cost, was American dominance unraveling?
READ: Mirror movements: Democratic socialism vs. MAGA in the battle for America’s future (July 3, 2025)
Historical Context: The Semiconductor Wars
To understand the significance of the DeepSeek moment, it’s crucial to examine the broader context of U.S.-China technology competition. Since 2018, the United States has implemented increasingly aggressive export controls on advanced semiconductors, with the most significant restrictions imposed through the CHIPS Act of 2022’s domestic incentives and BIS-imposed export restrictions introduced in 2022 and updated in 2023. Meanwhile, President Trump announced a $500 billion industry-led AI infrastructure venture in late January—which dovetails with the CHIPS Act’s $52 billion aimed at boosting domestic semiconductor leadership.
These export controls specifically target China’s access to advanced AI chips and manufacturing equipment. NVIDIA’s flagship H100 and A100 chips, which dominate AI training, are restricted from export to China. The H20, designed as an export-compliant alternative, became a critical test case for whether technological containment could coexist with commercial reality. Similarly, advanced lithography equipment from ASML, the Dutch company that produces the machines needed for cutting-edge chip manufacturing, is also restricted.
The strategic logic is clear: by controlling access to advanced hardware, the U.S. aims to maintain technological superiority in AI development. China will be forced to make do with older, less capable chips, theoretically limiting their ability to compete in the AI race.
Competition with China
However, what DeepSeek demonstrated is that constraints can fuel Chinese innovation. With limited access to state-of-the-art GPUs, China’s engineers focused on efficiency—producing high-performance results using leaner models, novel training techniques, and tight optimization. Their success proved that innovation doesn’t require brute computational force—it can also emerge from necessity.
This efficiency-first approach in China represents a fundamental challenge to our economic assumptions. If Chinese companies can achieve comparable AI performance at dramatically lower costs, will they undercut U.S. firms across multiple sectors? The implications extend beyond technology: cheaper AI may enable Chinese companies to dominate manufacturing automation, financial services, healthcare diagnostics, and autonomous systems—sectors where AI integration determines competitive advantage.
The H20 situation exemplifies this challenge. Even with export-controlled, deliberately degraded chips, Chinese firms demonstrated remarkable capability. This suggests that future restrictions may prove ineffective, while the revenue loss to U.S. companies may weaken America’s ability to fund continued innovation leadership.
But we may not have to push the panic button yet because there were limits. DeepSeek’s cost claims only reflected direct compute costs—not the years of R&D and infrastructure needed to achieve a stable release. Nor was its approach replicable at scale: R1 was trained on older-generation GPUs, still reliant on legacy architectures that struggle with truly cutting-edge applications.
Moreover, DeepSeek’s open-source approach—releasing model weights under an MIT license—generated massive publicity but cast doubt on its monetization strategy. Unlike OpenAI or Anthropic, DeepSeek hasn’t articulated a clear path to sustained commercial success.
Despite its headline-grabbing debut, DeepSeek’s app store dominance proved short-lived. By mid-2025, usage and engagement metrics (while not officially disclosed) showed signs of decline. This followed a familiar arc in tech: early hype without sustained ecosystem or business viability.
Critically, DeepSeek’s models face regulatory and ethical scrutiny. Studies show they embed Chinese government censorship, filtering sensitive political content. Others highlight security vulnerabilities—one paper reported a 100% success rate in jailbreaking harmful prompts. These weaknesses make global enterprise adoption problematic, particularly in sectors requiring regulatory compliance or handling sensitive data.
Most tellingly, the DeepSeek scare did not erode Nvidia’s structural advantages. CUDA’s dominance means developers are deeply embedded in Nvidia’s stack. Meanwhile, Nvidia’s chips are made by TSMC, which leads the world in 3nm and 2nm fabrication—China remains years behind in advanced manufacturing. Also, U.S. companies have brand trust, security clearance, and integration in critical infrastructure that China cannot easily replicate.
However, the H20 reversal reveals the economic unsustainability of technological isolationism. CEO Jensen Huang said China would “move on” with or without Nvidia’s chips, highlighting the risk that overly aggressive export controls could accelerate Chinese self-sufficiency rather than maintain dependence.
READ: Here’s how the $36 trillion US debt will affect your wallet! (July 21, 2025)
The New Cold War?
The broader economic threat from China’s AI advancement extends beyond individual companies or sectors. If Chinese firms can deliver AI solutions at dramatically lower costs while maintaining acceptable quality, they could capture significant market share in global technology services. This would represent a fundamental shift in economic power, similar to how Chinese manufacturing dominance emerged in the 2000s through cost advantages.
The semiconductor restrictions reveal both American strengths and vulnerabilities. While the U.S. maintains technological leadership in advanced chip design and manufacturing partnerships, the economic cost of restrictions—billions in lost revenue for companies like NVIDIA—may ultimately undermine the innovation capacity these policies aim to protect.
The H20 situation also demonstrates China’s growing sophistication in navigating international technology constraints. Rather than direct confrontation, Chinese firms are developing alternative approaches that achieve strategic objectives while technically complying with restrictions. This adaptive capability suggests that simple export controls may prove insufficient for maintaining long-term technological advantage.
Final Word
DeepSeek was not a Sputnik moment—it was a Sputnik spark. A signal flare that should remind American industry and policymakers that complacency is the enemy of resilience. The subsequent H20 policy reversal demonstrates how market forces and geopolitical strategy often conflict, forcing difficult choices between ideological purity and economic pragmatism. The fundamentals of American technological leadership remain strong, but they are not unassailable.
While Chinese firms will continue to innovate, their breakthroughs are unlikely to immediately overcome the systemic moat built by U.S. companies over decades: trusted manufacturing, mature ecosystems, policy alignment, and perpetual reinvestment.
However, the efficiency innovations demonstrated by DeepSeek, combined with China’s massive domestic market and growing technological sophistication, represent a genuine long-term challenge to American economic dominance. The H20 reversal suggests that pure technological containment may be economically unsustainable—forcing a more nuanced approach that balances security concerns with commercial reality.
In this AI age, it’s not enough to build a great model. We must scale it, secure it, monetize it, and integrate it. And that is where American dominance endures—not because we prevent innovation elsewhere, but because we’re best at turning innovation into infrastructure. The question is whether we can maintain that advantage while Chinese competitors prove increasingly capable of achieving more with less.
(Ajay Raju, a venture capitalist and lawyer, is the author of The Review, a new column that attempts to decode the patterns emerging from the unprecedented shifts reshaping our world. In a world where adaptation is survival, The Review offers a compass for the journey ahead).