The NextWave Signal
Week of 12–18 April 2026 | Issue #2
This week on Next Waves Insight
Anthropic’s Mythos Model Finds Thousands of Zero-Days — Then Gets Locked Behind a 40-Org Consortium — A model scoring 83.1% on cybersecurity benchmarks found a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg bug that fuzz testing missed. Anthropic chose not to release it publicly. Access goes through Project Glasswing — a named-membership consortium of 40+ organisations including Apple, Google, Microsoft, and CrowdStrike. The question: when Mythos finds zero-days in non-member vendors’ products, who discloses, and when? Read →
Google Gemma 4’s Apache 2.0 Licence Removes a Key Enterprise Deployment Blocker — Apache 2.0 means teams can self-host, fine-tune, and embed in commercial products without royalty exposure. That changes the self-hosted vs. hosted calculus for every enterprise still evaluating whether to run inference in-house. Read →
The Packaging Bottleneck: Why TSMC’s CoWoS Lines — Not Its Fabs — Now Control AI Chip Supply — NVIDIA has booked TSMC’s CoWoS advanced packaging capacity through 2027. Every chip fabricated in Arizona still ships to Taiwan for packaging — no US-based CoWoS facility exists. Earliest credible date for US packaging capacity: 2028. Infrastructure teams ordering AI accelerators today are dealing with a lead-time problem, not a price problem. Read →
When OpenAI Goes Public, the API Relationship Changes — and Not in Enterprise Buyers’ Favour — At $852B valuation and $25B annualised revenue, the implied multiple is ~34x — roughly 5x where growth-stage enterprise software stabilises post-IPO. Public company status means every pricing decision becomes a disclosure event. Teams building on OpenAI’s API should model what pricing looks like when OpenAI starts filing 10-Ks. Read →
Spider Silk Hits Industrial Volume: Kraig Biocraft’s 1.3-Ton Month — Recombinant spider silk has crossed 1.3 metric tons per month of production — the first time the material has been at supply chain scale. Near-term buyers are in defence and industrial composites. Read →
Signals worth watching
- NIST post-quantum standards now enforceable — FIPS 203, 204, and 205 were finalised in August 2024. NSA guidance targets RSA phaseout by 2030 for national security systems. The 2026–2027 window is when cryptographic inventory work needs to be complete — not started. Organisations still running RSA-2048 in critical infrastructure are inside the risk window.
- Section 232 semiconductor tariff probe moves toward recommendations — The Commerce Department investigation has a July 2026 review. A reclassification of advanced packaging services as taxable imports would raise the landed cost of US-delivered AI accelerators materially. Infrastructure procurement teams should model a 15–25% tariff scenario now. (Our coverage)
- Frontier Model Forum activates coordinated security intelligence sharing — OpenAI, Google DeepMind, Anthropic, and Microsoft are now sharing threat intelligence on model extraction attacks across lab boundaries. If your AI security posture assumes each lab operates in isolation, it needs updating. (Our coverage)
Stat of the week
$852 billion — OpenAI’s post-money valuation ahead of its Q4 2026 IPO target, on $25B annualised revenue. The implied ~34x price-to-revenue multiple is roughly 5x where comparable enterprise software trades post-listing. Every API contract signed today is priced against a company still burning capital; post-IPO pricing answers to public market investors. (CNBC)
What to watch next week
- Q1 2026 VC concentration data (publishing Mon 21 Apr): Four companies took 65% of all global venture capital in Q1. The question is not the record total — it is what that concentration means for the rest of the startup ecosystem.
- EU High-Risk AI Rules: August 2026 compliance deadline (publishing Tue 22 Apr): What financial services and healthcare teams need to have done — and what is already late.
- Why cheap inference has not made enterprise AI profitable (publishing Wed 23 Apr): Token prices have fallen. Enterprise AI margins have not followed. The gap between inference cost curves and enterprise ROI is the most important unresolved question in AI product economics.
— The Next Waves Insight editorial team
Forwarded this? Subscribe at nextwavesinsight.com.
