IBM Breaks Sub-1nm Barrier; US Clears Mythos 5 for Trusted Organizations

IBM's nanostack transistors cross sub-1nm as the US restores Claude Mythos 5 access for 100+ vetted organizations.

🗄️ IBM Claims World's First Sub-1 Nanometer Chip With Nanostack Transistors

Decoded: IBM on June 25 unveiled its nanostack transistor architecture — the company's claim to the world's first sub-1 nanometer chip manufacturing process. Presented at the IEEE Symposium on VLSI Technology and Circuits in Kyoto, Japan, the technology packs 100 billion transistors onto a chip the size of a fingernail, delivering up to 50% more performance or 70% better energy efficiency compared to IBM's existing 2nm nanosheet chips. IBM researchers also demonstrated a 40% improvement in transistor scaling density over the prior generation. IBM shares rallied in premarket trading on June 25. The nanostack process represents the first demonstrated path below the 1nm node threshold, a boundary the industry has treated as the practical limit of conventional silicon scaling. (Reuters, Ars Technica, The Verge, June 25, 2026)

Why it matters: Compute density and energy efficiency are the two binding constraints on AI chip economics — and nanostack addresses both. A 70% energy efficiency improvement over 2nm silicon would directly reduce the power consumption costs that are driving AI data center electricity demand to multi-gigawatt scale and constraining hyperscaler capex plans. IBM (IBM) does not operate a commercial foundry at scale today, but the nanostack architecture uses gate-all-around transistors compatible with high-volume foundry processes, meaning TSMC, Samsung, and Intel could in principle adopt or license the design framework. For investors tracking the AI hardware roadmap, this introduces a credible successor path beyond TSMC's N2 process — which currently has no confirmed commercial follow-on node. If nanostack reaches volume production, the energy-per-inference economics of AI clusters improve materially, shifting the buildout constraint from power procurement back toward compute.

🤖 US Government Clears Claude Mythos 5 for 100+ Trusted Organizations

Decoded: The US government on June 26 allowed Anthropic to restore access to Claude Mythos 5, its most capable AI model, for more than 100 'trusted' US organizations — partially reversing a security-based suspension issued two weeks earlier over concerns the model could aid cyberattacks. Bloomberg reported that Anthropic resolved Trump administration concerns before the clearance was granted. The restoration follows the access framework applied to Fable 5 earlier in June: government-vetted organizations in defense, research, and critical infrastructure can access Mythos 5 while the restriction for foreign entities remains in place. (Bloomberg, South China Morning Post, June 26, 2026)

Why it matters: The tiered access framework the US government is applying to frontier AI models — restrict foreign access, preserve vetted domestic access — is now a confirmed policy template. For Anthropic, the Mythos 5 clearance restores commercial revenue from its highest-value enterprise segment, including government contractors and defense-adjacent firms that had suspended usage under the earlier restriction. The speed of resolution signals that Anthropic's compliance posture with the Trump administration is sufficient to maintain domestic operations even for its most sensitive capability tiers. For competitors including OpenAI, Google DeepMind, and Meta AI, the precedent is clear: any frontier model demonstrating offensive cyber potential should expect a security review before full commercial deployment, and compliance — not legislation — is the path back to market. For investors tracking Anthropic's IPO, the clearance is a positive regulatory signal that even the most restricted AI capabilities can be restored through process.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team