The NextWave Signal
Week of 17–23 May 2026 | Issue #7
Three articles this week, three verification tests. IBM needs independent researchers to confirm its quantum chemistry result. Rentosertib needs Phase III to replicate a 12-week signal in 500+ patients. Personalized mRNA cancer vaccines need regulators and payers to accept five-year survival data as the basis for approval and coverage. The pattern: bold, credible, early-stage claims that now face the harder work of independent confirmation at scale.
This week on the NextWave Signal
IBM Is Targeting Verified Quantum Advantage by Year-End 2026. Here Is What the Claim Actually Means. — IBM’s Nighthawk processor — 120 qubits, targeting 7,500 two-qubit gate operations by year-end — is built around a claim with two formal conditions: outperform all classical methods on a commercially relevant problem, and have it confirmed by outside parties through an open Quantum Advantage Tracker run with Algorithmiq, the Flatiron Institute, and BlueQubit. The most likely first domain is quantum chemistry, where IBM-Algorithmiq have already demonstrated drug-candidate energy estimation exceeding prior methods. A verified chemistry result is not a signal to migrate enterprise workloads — it is proof that one problem class cleared one bar. The gate-count to watch is 5,000 today, 7,500 by December. Read →
The First End-to-End AI-Designed Drug Has Phase 2 Data. What Rentosertib’s Results Mean for the Pipeline. — Rentosertib (Insilico Medicine) posted a 118.7 mL FVC separation versus placebo at 12 weeks in the GENESIS-IPF Phase IIa — the first published clinical signal for a drug in which both target (TNIK) and molecule were identified by AI. The signal is credible: FVC gains tracked with profibrotic biomarker changes consistent with target engagement. The caveats are structural: 71 patients, China-only, a safety-powered trial not sized to confirm efficacy. Phase III — 500+ patients, Western populations, multi-year follow-up — is where the result holds or doesn’t. No AI-designed drug has reached FDA approval; Rentosertib is the field’s first clinical benchmark. Read →
mRNA Cancer Vaccines Have Five-Year Efficacy Data. Regulatory Submissions Are Next. — Moderna/Merck’s intismeran autogene sustained a 49% reduction in melanoma recurrence or death at five years — the same hazard ratio as year three; the survival gap has not converged. BioNTech/Genentech’s autogene cevumeran showed 87.5% of immune responders alive at six years in pancreatic cancer, a disease with 13% five-year survival. INTerpath-001, the Phase 3 pivotal trial for intismeran, has enrolled 1,089 patients; rolling BLA submission is anticipated in 2026 under Breakthrough Therapy Designation. The open questions are now operational: manufacturing cost ($100,000–$300,000 per patient), the responder-rate split (~50% in pancreatic data), and whether “vaccine” or “therapy” classification — a naming decision that determines the reimbursement tier. Read →
Signals worth watching
- TSMC’s 5.5-reticle CoWoS hits 98% yield, redefining AI packaging limits — At its May 14 Technology Symposium, TSMC announced mass production of a 5.5-reticle CoWoS packaging platform — the world’s largest chip-on-wafer-on-substrate — at yields above 98%, versus Intel’s ~90% for EMIB-T and Samsung’s mid-50% for 2nm. With 25 finalized 2nm tape-outs (70+ in development) and a 14-reticle / 20-HBM CoWoS roadmap to 2028, the AI hardware supply constraint is now publicly quantified — and concentrated. (Focus Taiwan)
- Anthropic overtakes OpenAI in US enterprise AI adoption for the first time — Ramp’s May 2026 AI Index, drawn from expense data across 50,000+ US businesses, shows Claude at 34.4% adoption vs OpenAI at 32.3% — a crossover that took Anthropic from under 8% in April 2025 to category leader in 12 months. Spending-data crossovers in the model market are rare; this one signals that CTOs are now evaluating models on code quality and API reliability ahead of brand. Lead is fragile: token economics, compute constraints, and OpenAI’s $4B DeployCo could compress the gap quickly. (Axios / Ramp AI Index)
- NIST/CAISI to pre-launch test frontier AI models from Microsoft, Google, and xAI — On May 5, the Commerce Department’s Center for AI Standards and Innovation announced it will evaluate unreleased models from Microsoft, Google, and xAI before public launch — focusing on cybersecurity, biosecurity, and chemical-weapons risk. This is the first formal US government pre-clearance framework for frontier AI, escalating beyond the voluntary post-deployment reviews CAISI has run for 40+ models. Pre-launch government testing changes release-timeline dynamics for every frontier lab. (Washington Post)
Stat of the week
118.7 mL — the forced vital capacity separation between Rentosertib 60 mg QD and placebo at 12 weeks in the GENESIS-IPF Phase IIa trial: the first published clinical efficacy signal for an end-to-end AI-designed drug. (Nature Medicine | NWI analysis)
What to watch next week
- Memorial Day weekend (Mon 25 May) — US tech industry slowdown; expect lighter news flow Mon–Tue. Plan external-signal research for Tue 26 May or Wed 27 May rather than the holiday itself.
- Computex 2026 wrap-up coverage — Nvidia, AMD, Intel, and Qualcomm Computex announcements being digested by analysts; secondary commentary typically lands the week after. Watch specifically for any Blackwell Ultra or Rubin architecture timelines that interact with the TSMC CoWoS packaging constraint covered in Issue #6.
- IBM Quantum Advantage Tracker — Algorithmiq/Flatiron Institute updates — Following the IBM Nighthawk article (WP 649), this is the primary ongoing signal to monitor. Any new community-verified circuit milestone published on the Tracker between now and end-2026 is a direct data point on whether IBM’s year-end advantage target is on track. Watch for preprints on the IBM-Algorithmiq drug-candidate energy estimation work.
— NWI Editorial
