Geoffrey Gewurz

Not reimbursed yet, small revenue in 2026. The product is actually very much needed and is a true novel approach. With reimbursement 100-200 million category.

Another important post by Alex Dickinson

Goldman initiates Caris at a buy 27 target.

Natera is doing something real here. Phased variants are a clean solution to the hardest MRD problem: distinguishing true signal from noise at extremely low levels. If multiple mutations are found on the same DNA fragment, the probability of random error drops dramatically. That directly improves confidence at ultra-low allele frequencies. This is not marketing. It’s a meaningful technical advantage in MRD today. The bigger question isn’t whether this works. It does. The question is how this approach scales across broader use cases beyond MRD, and how it competes with multi-omic, tumor-naive architectures over time. But on the core MRD signal extraction problem, this is a strong move by Natera.

Something big is converging. Most people haven’t connected the dots yet. Last night, after market close, Caris Life Sciences finalized ACHIEVE 1 — their multi-cancer early detection study. Blinded validation held up. The numbers are real. I covered that this morning. But what’s converging in the background is something else entirely. At GTC this week, NVIDIA made clear that biology is the next great compute problem. Jensen Huang didn’t just talk about chips. He talked about accelerating the entire stack — variant calling, multimodal AI, the fusion of genomic sequence data with imaging. NVIDIA’s Parabricks platform runs genomic pipelines 50x faster than traditional compute. What took hours takes minutes. What was prohibitively expensive at scale becomes routine. Then there’s Roche. Their new Axelios sequencer — built on a fundamentally new chemistry called Sequencing by Expansion — just launched at $150 per genome list price. The trajectory points toward $30 or less. Faster too: sample to variant call in under five hours. And it natively unlocks methylation mapping, RNA sequencing, and spatial multiomics in ways previous platforms couldn’t reach affordably. Ultima Genomics is pushing hard in the same direction. The competitive pressure on sequencing cost is structural and permanent. Now think about what Caris actually is. Caris has spent 18 years building something that cannot be purchased or replicated quickly — over one million cancer cases, 50 billion molecular markers, longitudinal outcomes data tied to real treatment decisions. Their core business — Molecular Profiling — is already generating nearly $1 billion in annual revenue. And their pipeline products — Detect, MI Clarity, ChromoSeq, MRD — are all being built on the same infrastructure already deployed for that core business. Here is the insight most people are missing: cheaper sequencing and faster compute don’t require Caris to rebuild anything. COGS compress. Throughput scales. AI models get richer input data, faster and cheaper to train. Every dollar NVIDIA and Roche invest in the infrastructure layer flows directly toward Caris’s economics — because Caris sits on top of that infrastructure and already owns what sits above it. A new entrant can buy a cheap sequencer tomorrow. They cannot buy 10 years of annotated, outcomes-linked, longitudinally followed patient records. That dataset is the moat. And the infrastructure tailwinds don’t erode it — they deepen it. The convergence of these three things — validated clinical data, collapsing sequencing costs, and GPU-accelerated interpretation — creates conditions for a very different kind of company than the market is currently pricing.

@carisls Congrats to management and all the employees of Caris. A real milestone and much to be proud of.

Caris Life Sciences announced the final results of the Achieve 1 Study. These results represent a major milestone in Caris’ goal to detect cancer earlier, through the future launch of Caris Detect™, its multi-cancer early detection (MCED) test. Learn more: ow.ly/4bec50YBwVY

The ACHIEVE-1 final numbers from Caris Life Sciences came out last night. A few thoughts, just how I see it. First, the early-stage detection. ~60% in Stage I–II, with high specificity. Given how little signal exists at that stage, that’s a real result. Not theoretical, not directional. Real. Second, and just as important, the specificity. It’s high, and it held. That matters more than people think. Without it, none of this works. You lose trust, you create noise, and the whole thing breaks in practice. This didn’t. Third, the control cohort. This is simply the group without cancer. The test stayed quiet there. That’s the whole game in screening. Not just finding cancer, but not finding it where it doesn’t exist. Fourth, the line that stood out to me the most: “1 of 9 pillars.” That’s the story. This is not a finished test. It’s the first version of a system that’s meant to improve. Add layers. Get better over time. That framing is very different from how most diagnostics are built. And finally, execution. There was some noise about timing drifting into Q2. They delivered March 31. In this space, that’s not small. It tells you how the company is operating internally. — What actually changed for me after this isn’t just the data. It’s confidence. Confidence in David Halbert and David Spetzler. Because from here, it’s not about whether this works in principle. It’s about whether they can keep improving it, move it into real use, and execute step by step. That’s the part that matters now. And I feel better about that today than I did yesterday.

Very important read for all investors in diagnostics systems

On #NationalDoctorsDay, we recognize the impact of all doctors. Learn how Dr. Leo Shunyakov of Central Care Cancer Center uses Caris testing to help extend lives and improve cancer care: ow.ly/SMTg50Yt7EI

NEW STUDY These Diets May Lower Your Risk of Several Cancers "A Study of Nearly 2 Million People Found That This Diet May Lower Cancer Risk Here’s what the research means for your plate" Joy Manning March 25, 2026 @DrRishabhOnco @GG57TLV eatingwell.com/vegetarian-die… #OrsiniUniversity

I wrote this piece to answer a narrower question than the usual MCED debate: not whether multi-cancer detection belongs in broad screening, but where the first real reimbursement wedge is most likely to emerge. I think the survivor population deserves much more attention. open.substack.com/pub/gg86/p/the…

Weekend read: what if the core assumption in diagnostics is wrong? open.substack.com/pub/gg86/p/liv…

Thanks again to Dr. Jain. This matters because it compresses the value of standalone biomarkers. If PD-L1 reading becomes automated and reproducible, the moat is no longer in scoring one marker. The moat moves to who can integrate pathology, genomics, and downstream clinical decision-making.

Again, another great one from Dr. Jain. If this holds, it suggests detection benefit isn’t confined to tightly defined high-risk cohorts. That has implications beyond CT — it challenges how we think about population-level screening more broadly.

This has been going on for days now. Very high cadence. Not immediate revenue but platform strength.