AMD Goes 2nm, Japan Gets Mythos, and the White House Blinks on AI

Week of May 25, 2026

Three stories this week illustrate the current trajectory of AI—and none of them concern a new chatbot. AMD challenged NVIDIA's chip dominance by putting the first 2nm server processor into production. Japan's government secured access to Anthropic's most powerful model, Claude Mythos, while simultaneously establishing a 36-entity cybersecurity task force to manage its implementation. Additionally, President Trump cancelled the signing of a new AI executive order, citing concerns that the language could provide China with a competitive advantage. The AI race is no longer solely about building the most intelligent model; it is about controlling hardware, managing access to powerful tools, and establishing the regulatory framework.

AMD's 2nm Gambit: The First Real Challenge to NVIDIA's Crown

On May 22, AMD announced the start of production for its 6th Generation EPYC processors, codenamed "Venice," built on TSMC's 2nm process technology. This is the first high-performance computing product to enter production at this node—a significant manufacturing milestone that places AMD in direct competition with NVIDIA's dominance in AI compute.

The technical specifications are critical. Transitioning to 2nm allows for more transistors per square millimeter, which translates to faster AI training and inference with lower power consumption. AMD is already planning a follow-on "Verano" processor to extend its roadmap. For enterprises running AI workloads, this represents the first credible alternative to NVIDIA's hold on accelerated computing since the AI boom began.

The timing is strategic. Demand for AI compute has exceeded the most optimistic forecasts, and NVIDIA's supply constraints have become a bottleneck for hyperscalers and startups alike. While AMD's Venice chips will not ship in volume until later this year, the signal is clear: the era of single-vendor AI hardware is ending.

Japan's Mythos Deal: Access Meets Anxiety

Japan's Finance Minister Satsuki Katayama announced this week that the Japanese government and the country's three megabanks—MUFG, SMBC, and Mizuho—will gain access to Anthropic's Claude Mythos model within two weeks. The agreement followed a meeting with US Treasury Secretary Scott Bessent, signaling high-level coordination between the two governments.

Of particular note is that Katayama simultaneously announced a 36-entity public-private working group specifically to address the cybersecurity risks Mythos poses to Japan's financial system. Japan is not treating this as a routine software deployment; rather, they are approaching Claude Mythos with a balance of enthusiasm for its capabilities and caution regarding its risks.

This model has caused concern in Washington and Wall Street. Anthropic has been selective regarding access, as Mythos represents a leap in autonomous reasoning and tool use that has cybersecurity experts concerned about the potential impact of a compromised instance within a financial network. Japan's approach—granting bank access while building a dedicated defense task force—is likely to become the template for how sovereign governments handle frontier AI.

The Executive Order That Wasn't

President Trump cancelled a scheduled Oval Office signing ceremony for a new AI executive order this week, stating that he was concerned the text could undermine the United States' competitive edge over China. The proposed order had emerged from growing pressure within the banking and financial sectors regarding AI cybersecurity risks—specifically those posed by models like Claude Mythos.

The cancellation suggests that the White House is balancing two powerful constituencies: financial institutions demanding stronger AI security guardrails, and national security advisors who fear that regulation will slow US AI development relative to China. This has resulted in a period of hesitation at a time when other governments—including Japan, the EU, and China—are moving decisively.

This is a significant development, as executive orders shape federal procurement, research funding, and export controls. Without one, the US government lacks a coordinated AI strategy while allies sign bilateral access deals and adversaries restrict foreign chips from state data centers. This vacuum is unlikely to persist, but the delay allows other nations to help set the global rules.

The Bottom Line

Three themes defined this week: hardware competition is accelerating, frontier AI access is becoming a matter of national security, and the US government is struggling to keep pace with both. Expect more countries to follow Japan's model of granting access to powerful models while building parallel defense structures. Furthermore, AMD's 2nm announcement is expected to trigger increased investment in alternative AI chip architectures. The next six months will determine whether AI hardware remains a monopoly or evolves into a competitive market.