Export controls, leaks redraw AI power map

The US export controls on Anthropic’s Claude Mythos AI models have marked a pivotal inflection point in the global AI arms race, sharply intensifying techno-nationalist sentiments and geopolitical tensions worldwide. By restricting foreign access to leading AI technologies, these bans have disrupted international collaboration and ignited fierce competition among nations striving to assert sovereignty over AI development, with Europe notably accelerating efforts to build homegrown AI capabilities to avoid dependency on US-controlled tech. This shift underscores a growing fracture in global AI governance as regions grapple with maintaining autonomy amid mounting export restrictions.

Anthropic’s Claude Mythos leak has compounded the challenges posed by US export bans by exposing critical cybersecurity vulnerabilities that exacerbate governance crises and complicate sovereign AI strategy formulation. As the US clamps down on Anthropic’s technology for foreign entities, the resulting governance chaos and market jitters have intensified, revealing deep fractures in international regulatory frameworks and fueling a volatile AI arms race. These developments highlight the intertwined risks of technological insecurity and geopolitical techno-nationalism, as nations confront not only competition but also infrastructure and energy constraints amid escalating AI tensions.

The AI sector's explosive growth is driving a massive expansion in compute and data center infrastructure across the U.S., with hyperscale projects underway from Texas to the Mountain West, signaling that the AI infrastructure buildout is still in its infancy rather than approaching saturation. This surge is underpinned by unprecedented capital commitments, such as Google's $200 billion compute deal with Anthropic and SpaceX's aggressive $60 billion acquisition of AI coding startup Cursor, reflecting a strategic realignment of investment flows toward AI-driven technologies amid a volatile macroeconomic backdrop marked by Federal Reserve rate uncertainties and geopolitical tensions.

Investor enthusiasm for AI has sparked an IPO frenzy and soaring valuations, exemplified by Getty Images' 200% stock surge following its OpenAI partnership and SpaceX's planned $2 trillion IPO, which together have injected both excitement and turbulence into the market. Yet, this fervor carries significant risks reminiscent of historical tech booms, as many AI startups compete in a crowded field with only a handful expected to succeed, making the landscape highly speculative and vulnerable to sharp corrections once commoditization pressures and market concentration intensify.

The AI revolution is reshaping value creation by shifting it downward into capital-intensive base layers such as semiconductors, advanced packaging, and data infrastructure, compressing margins at the application level and prompting a capital rotation back into physical assets after decades favoring asset-light models. Hyperscalers are now spending over $500 billion annually on capex, with companies like Micron at the epicenter due to their critical role in supplying scarce high-bandwidth memory essential for AI workloads, making upcoming earnings reports a bellwether for the sustainability of AI infrastructure investments despite the sector's recent 300% stock surge and elevated valuations.

Venture capitalists are navigating the AI boom with a mix of cautious selectivity and recognition of the sector's unpredictability, focusing on fewer, high-quality deals in burgeoning areas like generative and creative AI while grappling with an overwhelming volume of similar startups rapidly generating revenue. As Chamath Palihapitiya observes, competition is inevitable without monopolies, and while leaders like OpenAI and Anthropic will hold significant market share, the landscape remains fragmented and volatile, underscored by spiraling operational costs such as token usage that have forced companies like Uber and Microsoft to rethink their AI spending strategies.

The AI arms race is increasingly defined by a fierce competition for compute resources and infrastructure, exemplified by the U.S. Department of Defense’s rapid AI adoption reaching 1.5 million users and a $46 billion budget boost aimed at expanding compute capacity. Yet, despite this massive investment and vendor clearances including Google and Microsoft, persistent governance and talent shortages underscore the complex challenges of scaling AI infrastructure securely and effectively within government and industry alike.

Control over foundational AI infrastructure—model weights, compute power, and distribution networks—is concentrated among a handful of dominant players like NVIDIA, OpenAI, and Anthropic, who act as the durable 'railroads' of the AI ecosystem rather than the fleeting 'fashion houses' of applications. This capital-intensive, patient infrastructure layer contrasts sharply with the fast-moving but fragile application layer, making ownership of compute assets critical for startups seeking long-term competitive advantage rather than merely renting resources and exhausting their creative energies.

The next AI boom is unfolding in the physical economy, with Goldman Sachs projecting a staggering $7.6 trillion global investment in AI infrastructure—including data centers, compute, and power—between 2026 and 2031. This expansion into industrial sectors such as factories, mines, and utilities is driving a surge in demand for specialized hardware and compute resources, while fostering unprecedented collaboration between tech and traditional industries. However, capital allocation remains a critical bottleneck, as the pace of AI advancement outstrips available investment, determining which sectors will thrive or stall in this infrastructure race.

Talent scarcity poses a profound challenge amid the AI arms race, as leading companies like OpenAI, Anthropic, and Google DeepMind attract top researchers and ethicists away from academia and public service by offering salaries nearly triple those of university professors. This brain drain risks undermining independent oversight and ethical governance, creating conflicts of interest reminiscent of the regulatory 'revolving door.' To mitigate these shortages, firms are investing heavily in upskilling and leveraging AI augmentation tools—some paying as much as $20,000 per month—to boost engineering productivity, while emphasizing foundational well-being practices like sleep and nutrition to sustain high-performing teams.

Anthropic’s Claude Mythos leak has starkly illuminated profound cybersecurity vulnerabilities and governance failures within the AI ecosystem, triggering a global crisis that exposes the fragility of current protective frameworks. This breach not only revealed deep flaws in safeguarding sensitive AI technologies but also catalyzed an open-source arms race, intensifying competition and risk across borders. As the industry grapples with soaring AI token costs and strategic upheavals, these events underscore the urgent need to fortify AI governance to prevent further destabilization.

The entanglement of U.S. government equity stakes in leading AI companies has further complicated governance, eroding trust within the industry and spotlighting regulatory fragmentation. This intertwining of public and private interests disrupts transparent oversight, fueling skepticism about impartial governance and amplifying calls for clearer, more accountable frameworks. The resulting governance fractures have been exacerbated by export restrictions on Anthropic’s technology, which, while intended to mitigate risk, have instead intensified geopolitical techno-nationalism and fragmented regulatory responses.

Beyond governance and trust issues, the fallout from Anthropic’s AI shutdown and U.S. export bans has spotlighted critical infrastructure and energy challenges that underpin AI’s operational stability and cybersecurity resilience. The surge in energy demands and infrastructure battles, driven by the escalating AI arms race, reveal systemic vulnerabilities that current governance failures have yet to address. This confluence of technological, geopolitical, and infrastructural pressures calls for sovereign AI strategies and enhanced oversight mechanisms to safeguard both national security and the integrity of global AI ecosystems.