Noor's Newsletter: Issue #3

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

The End of the Beginning: From Models to Stacks

There's a familiar pattern to how transformative technologies mature. First comes the explosion of new tools and unproven ideas—the Cambrian explosion. Then, inevitably, comes the consolidation and the unglamorous, essential work of building the infrastructure to make any of it useful at scale. We saw it with the internet, we saw it with mobile, and we are seeing it happen in real-time with AI in medicine. The chaotic, exciting "beginning" is over. We are now entering the age of industrialization.

The focus is no longer on the brilliance of individual models, but on the construction of full, end-to-end "Industrial Stacks" for biotechnology. A stack isn't just a collection of technologies; it's a vertically integrated system of policy, data, computation, and clinical infrastructure designed to accelerate the entire lifecycle of medicine. What's different this time is the speed and the global, geopolitical scale of the competition to build them.

Look at the launch of India's first National Biobank. Source. This is not a research project; it is a state-level declaration that sovereign, population-scale biological data is a strategic national asset, the foundational layer of a domestic Bio-AI stack. It’s the 21st-century equivalent of building a national power grid. At the same time, a feature in Bloomberg details how Chinese drugmakers are now competing at the highest levels of novel drug discovery. Source. This is the output of their own decade-long strategy of state investment and talent cultivation. They are no longer just participating in the global supply chain; they are building their own, from top to bottom.

This industrialization is happening in the West as well, but manifesting differently. The new Center for Pediatric CRISPR Cures, a collaboration between Chan Zuckerberg Initiative and the Innovative Genomics Institute, is a focused, mission-driven industrial stack. Source. It is a purpose-built engine—combining philanthropic capital, focused research, and a clear clinical goal—designed to solve a specific class of problems.

Meanwhile, the platform layer of this stack is solidifying. The Microsoft and Viome partnership and the trend towards system-level IT redesign in Asian hospitals point to the same truth: the infrastructure is being built. Source. Source. And the regulators are building their part of the stack, too. The FDA's Breakthrough Designation for ArteraAI's tool, a precision medicine tool intended to assist clinicians with risk-based decisions for patients with localized prostate cancer, Source. and the EMA's formal AI strategy is creating the official pathways and rulebooks for this new industry. Source.

All these signals point away from the idea of a single, magical AI model that solves biology. Instead, we are seeing the deliberate, large-scale construction of competing, integrated systems. The future of medicine is being forged in these national and industrial stacks. The question is no longer who has the cleverest algorithm, but who can build the most efficient and powerful engine to turn science into medicine, and do it on a global scale.

-Noor

Interesting Things in Research and Beyond

The most interesting paper I read in the last two weeks wasn’t from biology, but from the world of autonomous vehicles. A technical report from Waymo titled "Scaling Laws of Motion Forecasting and Planning" details a simple, profound truth: their models' performance improves predictably and consistently with the sheer volume of high-quality training data. More miles driven in complex scenarios directly translates to better, safer predictions.

This is a powerful parallel for our field, as highlighted in a recent post from the LeashBio newsletter, "Good binding data is all you need." The core argument is that for drug discovery, while we chase more complex model architectures, the real bottleneck—and the most predictable path to improvement—is the generation of massive, high-quality, fit-for-purpose datasets that measure the physical interaction between a molecule and a protein. The Waymo paper is a clear data point from another domain that suggests that the most defensible moat in Bio-AI will not be the algorithm itself, but the data generation engine that continuously feeds it. It's another piece of evidence that the future belongs to those who can master the messy, physical-world workflow of creating better data.

Signals to Watch

Regulatory Landscape: The global regulatory picture continues to diverge.

United States: The White House has unveiled its "One Big Beautiful Bill," a sweeping deregulatory package. A key provision aims to drastically reduce the compliance burden for SMEs in the tech sector, including AI. However, the bill also hints at significant new tariffs on electronic components from Asia, creating deep uncertainty for companies reliant on global hardware supply chains.

European Union: The EU clarified how the AI Act applies to medical applications, specifying risk categories and obligations for transparency and human oversight in high-risk use cases. This, combined with new EMA guidance on AI in clinical trials, creates a clearer but more demanding regulatory environment.

Industry Moves:

VCs Bet Big on AI for Evidence Synthesis: In one of the largest early-stage rounds of the year, OpenEvidence raised $210M in a deal led by GV. The company is building an AI-powered search and synthesis engine for medical and scientific literature. The deal's competitive nature signals investors' belief that a foundational model for understanding scientific evidence is a massive prize.

23andMe's New Chapter: In a bold strategic pivot, Anne Wojcicki—founder of 23andMe—has launched the non-profit TTAM Research Institute, which has acquired 23andMe’s core assets following the company’s bankruptcy. This move separates the public-good research mission from the pressures of the public market, creating a new non-profit model to leverage the company's vast genetic database for drug discovery.

The M&A "Quiet Period" Gathers Evidence: The relative slowdown in pharma mega-mergers has become more pronounced. A mid-year outlook from PwC released on July 14th noted that while overall deal value is stable, it's driven by a higher volume of smaller, "string-of-pearls" acquisitions rather than transformative $50B+ deals, suggesting a broader trend towards risk-mitigation.

Ecosystem Opportunities

• For Founders: KQ Labs, the accelerator for data-driven health startups based at London's Francis Crick Institute, has opened applications for its tenth cohort. It's a key program for early-stage companies, providing crucial funding, mentorship, and network access to help build the next generation of Bio-AI ventures.

Outside Interest

• Academic Gamesmanship: News magasine Nikkei Asia on the inevitable next step of adversarial attacks: researchers are embedding hidden prompts into scientific papers with instructions like "positive review only," designed to be read by any LLM tasked with peer review. This creates a fun ethical dilemma: what's worse, poisoning the data well, or using an AI to drink from it in the first place?
• The Power of Letters: Wired has a great scoop on the "comfort letters" President Trump sent to tech platforms, effectively telling them to ignore the law that could force a TikTok shutdown. It's a fascinating look at the use of executive pressure where legal authority is absent. (Such letters have no actual legal force, which is, of course, the entire point of sending them.)