Noor's Newsletter

This isn’t a recap of the weeks—it’s an attempt to understand the forces reshaping how we live, govern, and evolve.

Issue #1: The Sovereignty Race in Bio-AI

For years, the dominant question in Bio-AI was: Whose model is best? We obsessed over architectures, training efficiency, and the rise of foundation models. But the question has shifted. Performance still matters—but control matters more.

We’ve entered the era of Bio-AI Sovereignty.

In this new phase, value isn’t about having the best model. It’s about owning the full stack: data, compute, models—and most crucially, the closed loop between them. Better data trains better models, which create even better data. This feedback loop becomes the moat. It compounds.

We're seeing this play out not just in corporate R&D, but across government policy and research strategy. Four developments in June illustrate how deep this shift runs.

Novo Nordisk x NVIDIA: Building the In-House AI Engine

On June 11, Novo Nordisk announced a partnership with NVIDIA and the Danish Centre for AI Innovation (DCAI). Press framed it as collaboration. But look closer—it’s vertical integration.

Novo isn’t just accessing compute—they’re internalizing it. With NVIDIA’s BioNeMo and NeMo microservices, they’re creating a proprietary AI engine tailored to their discovery pipeline. Source

This is a move from API user to sovereign AI operator. They’re betting model performance is now commoditized—and that competitive advantage will come from systems trained on proprietary data, custom-built for their specific modalities.

UK’s 10-Year Health Plan: Data as National Infrastructure

The UK government’s latest NHS reforms are often described as streamlining clinical trial access. But the deeper strategy is national data control.

Via the NHS App and new contract frameworks, the UK is positioning itself to build one of the world’s most valuable public health datasets. Source

The goal isn’t just better trials. It’s to generate high-quality longitudinal data—spanning genomics, diagnostics, lifestyle, outcomes—on 67 million people. Few countries can match the UK’s mix of centralized data, universal coverage, and research infrastructure.

The message is clear: health data is a sovereign asset, like telecom or energy. And the UK wants to be more than a passive innovation market—it wants to be an AI biomanufacturing hub.

SandboxAQ’s SAIR Dataset: Raising the Floor, Forcing the Shift

SandboxAQ, the Alphabet spinout, released SAIR: over 5 million protein-ligand structures labeled with binding affinities. It’s the largest dataset of its kind—and it’s open. Source

On the surface, this is academic generosity. Strategically, it raises the baseline and forces serious players to move upstream—toward closed data, experimental validation, and custom pipelines.

Open sourcing this foundational dataset commodifies public efforts and pressures competitors to secure private data. It mirrors the OpenAI model: public pretraining, private fine-tuning.

A New Bio-AI Stack: Control as Strategy

The evolving Bio-AI landscape can be visualized as a strategic pyramid. At its base lie public models trained on open datasets—accessible and low-cost, but increasingly undifferentiated. The middle tier consists of specialized tools, proprietary fine-tuning, and selective experimental capacity. But it’s the top of the pyramid where true defensibility emerges: sovereign loops in which proprietary data continuously improves models, and those models generate new proprietary data in return. This compounding cycle of improvement not only drives performance—it creates a structural moat that is difficult to replicate. In this model, control becomes strategy, and value consolidates at the top.

Bio-AI as Statecraft

The evolving Bio-AI landscape reflects a broader shift from technical competition to structural control. At its core lies a strategic pyramid: public models and open data at the bottom—cheap but undifferentiated; specialized tools and private fine-tuning in the middle; and at the top, sovereign loops in which proprietary data continuously improves models, which in turn generate more proprietary data. This self-improving loop consolidates value and becomes nearly impossible to replicate. It’s not just corporate strategy—it’s geopolitics. China is doubling down on genomic sovereignty with enforced data localization and state-run biotech AI labs. The UK and US are aligning national strategies around AI and health data. The EU risks paralysis—ambitious in policy, slow in execution. AI in medicine has joined defense, energy, and finance as a domain of strategic autonomy.

The Hidden Layers of Sovereignty

Regulatory Arbitrage: The Quiet Race

Sovereignty is also about where you build. Singapore’s Health Sciences Authority (HSA)—recognized by the WHO as a Stringent Regulatory Authority—has developed fast-track pathways and reliance frameworks that allow for quicker approvals and regional scaling of AI-driven medical technologies. Meanwhile, Estonia’s e-Residency program streamlines EU market entry, enabling startups to establish compliant operations without a physical presence. These jurisdictions are quietly becoming magnets for Bio-AI startups that prioritize speed, flexibility, and market access.

Compliance is a hidden tax. Startups in Germany may burn 15–20% of R&D on regulatory overhead. Meanwhile, companies in lighter-touch jurisdictions reinvest into models. Time-to-approval becomes a compound advantage.

Talent Hoarding: The Bottleneck

All the data and compute in the world can’t help if you don’t have people who understand both biology and AI. This rare talent—true bilinguals who can navigate both molecular biology and machine learning—is in short supply, and increasingly, it’s the real bottleneck to progress. We’re seeing new models emerge: Fractional Chief AI Officers—senior scientists who split their time across multiple companies—are becoming a popular workaround for firms that need strategic AI leadership but can’t afford or attract top-tier talent full-time. Universities are beginning to respond with cross-disciplinary programs, but most PhDs and MDs are still trained in silos. The result: talent becomes the new terrain for competition—hoarded, cross-leased, and fiercely protected. And unlike software engineering or content generation, this kind of systems-level, cross-domain expertise is not easily replaced by agentic AI tools. These roles require judgment, institutional context, and biological nuance that even the most advanced models struggle to replicate.

Unit Economics: Still Theoretical

The sovereignty model remains capital-intensive.

Novo’s AI stack is a long-term bet. Clinical ROI is uncertain—drug development still sees sub-15% success rates. Recursion Pharmaceuticals reported a net loss of $203 million in Q1 2025, with an operating cash burn of $118 million, and ended the quarter with $509 million in cash reserves (Source). The company also laid off 20% of its workforce to extend runway. Relay Therapeutics, meanwhile, raised $520 million to date, but trades at a ~$544 million market cap as of June 2025, with Q1 revenues of just $7.7 million (Source). These cases underscore the financial strain of building sovereign Bio-AI stacks and highlight the risk. Relay Therapeutics raised $520M and now trades below that.

Even Caris Life Sciences—this month’s rare Bio-AI IPO success—took years of real-world data monetization to get there.

Who Needs to Care?

• Startups: If you don’t own your data loop, you’re just an API user.
• Pharma: Your AI engine is core infra. Don’t outsource it blindly.
• Governments: Treat health data like sovereign infrastructure.
• Academia: Public datasets matter—but expect the frontier to disappear behind closed doors.

We’ve seen this before in cloud. First came shared tools. Then private clouds. Then custom chips. Bio-AI is following the same curve.

Building a self-improving biological AI stack is brutally hard. But once it works, it’s hard to copy—and nearly impossible to catch.

— Noor

Signals to Watch

🧠 Novo at London Biotech Show: Novo Nordisk demonstrated AI/ML-powered drug discovery workflows, integrating proprietary biological datasets (London Biotechnology Show).

📉 E&Y Biotech Report: EY’s Biotech Beyond Borders 2025 shows 39% of biotechs risk running out of cash within a year—highlighting growing investor focus on platform sustainability and sovereign data strategies (EY Global Report).

🧬 ASCO 2025: AI-enhanced diagnostics for HER2-low breast cancer broadened treatment eligibility. Precision improvements reported in session abstracts; real clinical impact still under follow-up (ASCO Meeting Abstracts).

🇬🇧 UK Moonshots: The UK government pledged £82.6M to fund new AI health hubs like PharosAI and Bind Research, part of a broader life sciences industrial strategy (UK Government Press Release).

🚪 IPO Window Still Shut: As reported by BioPharma Dive (June 20), biotech IPOs remain limited. Sovereign positioning is increasingly a survival strategy (BioPharma Dive).

New Fronts in the Race

1. Viome: Consumer-Generated Sovereignty

Viome’s 500,000+ at-home testing kits generate proprietary longitudinal health data—from the consumer up (Viome Company Overview).

But consumer sovereignty is fragile. 23andMe’s bankruptcy exposed how easily user data can become a tradeable asset. Legal control trumps marketing promises.

Strategic Lens: Consumer-scale datasets could rival national ones—if platforms can maintain user trust and withstand regulatory pressure.

2. Caris Life Sciences: The Sovereign Dataset IPO

Caris raised $494M in a strong June IPO on the strength of its real-world clinicogenomic data—built via revenue-generating services, not grants (Reuters | FierceBiotech).

Their dataset—50+ petabytes strong—powers both diagnostics and model training. This is sovereignty in practice: clinical utility funds the dataset, which reinforces clinical value.

Strategic Lens: The market rewards defensible datasets, not speculative AI.

3. AstraZeneca + CSPC: Hybrid Sovereignty

AZ’s $5.2B partnership with China’s CSPC fuses local data/AI capability with global pharma infrastructure. It’s a hedge—regulatory risk is real, and AZ is navigating it by co-owning sovereignty (Reuters).

4. US–China Friction: The Sovereignty Opportunity

As Washington tightens access to Chinese CROs/CDMOs like WuXi AppTec, US firms are reshoring testing and doubling down on early-stage licensing. $18B in Chinese-origin deals signed in 2025—9x growth (FiercePharma).

Strategic Lens: Friction is a forcing function. The US is using it to repatriate control—and attract global partners seeking stability.