Trump's AI Race Vs. China's Leaks: Why Containment Could Backfire

Forbes

3 days ago

WASHINGTON, DC - JULY 23: Michael Kratsios, U.S. President Donald Trump and David O. Sacks speak ... More onstage at the All-In and Hill & Valley Forum "Winning The AI Race" at Andrew W. Mellon Auditorium on July 23, 2025 in Washington, DC. (Photo byfor Hill & Valley Forum)
"AI is the new geopolitical lever"—this was the framing used in recent Washington briefings, where insiders described America's AI strategy as doing "two things fast, one thing slow—deliberately." The fast elements include granting FAST-41 fast-track permitting status to every data center above 100 MW, slashing timelines from years to months, and imposing stepped-up disclosure rules on shipments of advanced chips like the B200, H100, and H200 to list end-users in detail. The slow part is a deliberate study on "compute choke-points," due in December 2025, to determine whether to cap Chinese access to American cloud capacity. As these insiders noted, "We have the best technology, and we want to share it"—but only under an America First approach of techno-nationalism that ensures global infrastructure depends on U.S. foundations while restricting adversaries.
On January 20, 2025—while inaugural crowds were still finding their seats—DeepSeek released R1. No Silicon Valley fanfare, just a GitHub drop and iOS icon that became America's most-downloaded free app within 48 hours. Five months later, as the White House unveiled its AI Action Plan on July 23, Alibaba-backed Moonshot dropped Kimi K2 days before, claiming 100x cost advantages over Western models.
Coincidence? Hardly. It's Beijing projecting parity, if not superiority, precisely when Washington asserts dominance. The message: We can build frontier AI with whatever silicon you fail to contain.
At its heart, the U.S. strategy stands on three legs: Promote American technology globally, Protect it from adversaries, and pursue Techno-Nationalism. The revelation of over $1 billion in Nvidia B200/H100 chips reaching China through Singapore-Ho Chi Minh City shell networks exposes this as wishful thinking.
Here's the paradox: Failed containment doesn't just weaken "Protect"—it actively accelerates Chinese capabilities. Export controls create what analysts call "innovation incentives in captivity"—spurring Beijing to develop cheaper, more energy-efficient alternatives. DeepSeek's claim of training R1 for $6 million versus GPT-4's $100 million isn't just cost efficiency—it's containment blowback.
U.S. insiders emphasize that what has historically slowed China wasn't lack of technology but the Chinese Communist Party's ideological conformity, requiring models to pass rigorous alignment with official doctrine. Yet even with these self-imposed limits, Beijing's efficiency gains show how U.S. restrictions can backfire, pushing adversaries to innovate around ideological and technical barriers.
AI policy is energy policy. While Western democracies debate environmental reviews and grid reliability, authoritarian systems can rapidly direct energy flows to computing infrastructure. Wood Mackenzie's projection of 15-20% annual U.S. electricity demand growth from AI through 2030 isn't just a technical constraint—it's a strategic vulnerability.
The U.S. response—expedited FAST-41 permitting for >100 MW data centers—assumes regulatory streamlining can match centralized resource allocation. It can't. Regulation isn't necessarily the enemy here; it's the necessary guardrail against social, political and industrial infrastructure that could collapse under its own power demands.
Recent Washington briefings highlight growing concerns about Gulf-based corporate partnerships with U.S. AI projects serving as potential infiltration vectors. While specifics remain under wraps, the pattern involves sophisticated state actors using third-party relationships to access American technological developments—extending beyond chip smuggling to strategic information gathering.
The Biden-era export controls and AI safeguards, such as Executive Order 14110, faltered due to structural bottlenecks that delayed infrastructure buildouts and allowed adversaries like China to gain ground through persistent evasion. This erosion is accelerating under Trump's deregulatory push, which has revoked those "burdensome government requirements" to prioritize speed. The result creates a feedback loop: domestic acceleration exposes enforcement gaps, which in turn necessitate more aggressive deregulation to maintain competitive advantages.
Trump's approach represents the essential path forward to contain China, even if imperfect, as it addresses these foundational weaknesses head-on.
We're witnessing the emergence of parallel AI ecosystems: U.S. resource-intensive models versus Chinese efficiency-driven alternatives. Rather than a single dominant technological stack, we may see competing approaches based on cost, performance, and regulatory philosophy.
The question facing American strategists transcends tactical adjustments: Should U.S. strategy be defined by speed or security? Every month shaved off U.S. permitting may be offset by continued Chinese access to cutting-edge silicon through gray markets. Every export control may be accelerating the very multipolarity Washington seeks to prevent.
In the end, the AI race isn't static—the course shifts quicker than anyone can sprint.