Two Playbooks: Go Broad or Go Niche — Who Wins? | Tempus AI vs. Valar Labs

Two companies chasing the same mission — getting the right cancer treatment to the right patient — with two completely opposite playbooks. One raised $1.3 billion to build an empire. The other ships FDA-recognized diagnostics with a seven-person engineering team. So which way should a startup go: broad or niche? And who actually wins?

📧 Connect with Host 

Host Deepti Kalghatgi : https://www.linkedin.com/in/deepti-kalghatgi/

🌐 Visit: https://idaahub.com

In this episode of the Innovation & Startup Series, we unpack Tempus AI and Valar Labs — how each was built, how each grew, and three lessons any founder can borrow.

Tempus, founded by Groupon co-founder Eric Lefkofsky after his wife's cancer diagnosis, went all-in on owning the whole stack — genomics, clinical data, labs, and AI — and now does over $1.27B in annual revenue. Valar Labs went the opposite way: one sharply focused question — will this treatment work for this patient? — answered by running AI over routine pathology slides that already exist, built by seven engineers who optimized everything for one thing: speed of iteration.

We dig into why capital intensity drives the broad-vs-niche decision, why data is the real moat (and the two ways to win it), and why fast iteration toward validated evidence beats a perfect initial design.

What you'll learn:

• How Tempus and Valar chose broad vs. niche — and the role capital played
• Why proprietary data is the moat, whether you generate it or unlock it
• Why speed of iteration is the quiet engine behind both companies
• How a seven-person team out-ships rivals many times its size

Sources & further reading:

• Valar Labs — "5 Years, 7 Engineers, No Cloud": https://www.valarlabs.com/engineering/7-engineers
• Valar Labs — $22M Series A: https://www.valarlabs.com/news/series-a
• Tempus AI: https://www.tempus.com

If you enjoyed this, follow the show and share it with someone building in healthcare or AI.

📧 Connect with IDAAHub 

Follow us on:

LinkedIn: https://www.linkedin.com/company/idaahub/

https://www.youtube.com/@IDAAHUB

https://open.spotify.com/show/3V8Vuhqkibej5fUwtwkUMx

https://podcasts.apple.com/us/podcast/the-idaa-hub-podcast-ai-in-finance-healthcare/id1848710327

Transcript

SPEAKER_00 0:02

Hello everyone. Welcome back to the Innovation and Startup Series. Today is our second show where we actually pull apart and look at how real companies are actually getting built and what we can learn from their playbook. Picture an oncologist sitting across from a patient who has just been diagnosed with cancer. There's a treatment decision to make, and that's the decision that matters the most to the patient. For a long time, this choice has been filled with a lot of uncertainty. So, will this therapy work for this patient? Should we go with chemo? There are so many different types of therapy for cancer patients now. So most of the time it's actually, let's try and see. Today we're going to talk about two companies and two completely opposite ideas. The first begins in 2015. And he's stunned by how little data is shaping her treatment. So he sends out to build an empire, a company that will own all of it, the genetics, the records, the labs, the AI. And that company is Tempus AI. The second, which we were talking in the beginning about the uncertainty of treatment, is all just very young, just three years old. Three young Stanford engineers look at the very same problem and make a different kind of solution. They did not go ahead to build an empire. They just built one sharply focused tool that answers a single question. Will this treatment work for this patient? And the answer is using the tissue slide, the lab already has sitting on a shelf. That company is Valerie Labs and it's just barely five-year-old. So one went ahead in a very broad way to build almost an empire that tries to do everything, but the other is master, it's very specific and it's a specialized one. And the fascinating part is both are actually winning. So today there are two playbooks and three lessons you can learn from them. So let's hold on to that contrast because it's it's part of the whole episode. Tempest tries to own everything, like I said, right? Weller is one is one thing and does it beautifully. Same mission, the right treatment for the same patient, for the right patient. Two radically different bets and how to get there. So let's meet them properly, starting with Tempest AI. So let's go deeper on Lewkovsky. This is he's a serial entrepreneur. He had co-founded Groupon and already taken multiple companies publicly. He did not need another startup. But watching Liz's treatment, one thing made it clear. He was perplexed at how little data had permeated her care. The information doctors needed existed. It was just locked away, scattered and impossible to use in the moment. That frustration became the mission. And Temple set out to fix it at scale. Here's what it does: it uses AI to read a patient's genetic code, their molecular data, and then compares it against millions of others. So an oncologist can pick the treatment most likely to work for that specific person. But the real genius isn't the AI model, it's the data underneath it. Temple spent years doing the unglamorous work of taking all that messy, what we call unstructured clinical information, which is being buried in notes, electronic records, and structuring it, combining that with genomics, clinical data, and imaging into something an algorithm could learn from. And it went all in on owning the whole stack, what's called a closed loop model, controlling both the data generation and analysis. That's enormous. And actually, that requires enormous capital. Tempus raised roughly 1.3 billion privately from backers, including Google, before going public in June 2024 at a valuation at 6 billion. The growth has been remarkable. Revenue climbed from 693 billion in 2024 to roughly 1.27 billion in 2025. With a net revenue reduction of 126%, meaning existing customers keep spending more. So in 2025, they actually acquired Page to feed its data engine. The one asterisk, Tempest is not completely profitable, it's still an empire and it's still being built. Now let's look at the specialist here. Now, those three Stanford engineers. And it is a good thing. These are the three of them. Valor CEO CEO and CTO. Who took the company out of stock 2023. It was exactly the same problem. Left cost keys, but opposite instinct, but how they're gonna approach it. Where Tempest tries to do everything, Valor does exactly one thing. It predicts whether a specific cancer treatment will actually work for a specific patient. Here's a good part. It doesn't predict new molecular data. It reads the green the routine pathology slide, the standard strain tissue sample or lab already that's been sitting there and runs AI over it. They call this computational histology AI. It's a first AI-based test to predict treatment response in bladder cancer, specifically to a therapy called BCG, which matters even more right now because there's a nationwide BCG shortage. If you can predict who it will help, you don't waste a scarce drug. And WLR is doing that genuinely really well. Not just funded, it also is validated. In 2024, it raised 22 million Series A, co-led by top-tier firms DCVC and Anderson Horwitz, Bio Plus Health Fund. The founders framed the mission plainly: reduce the uncertainty in cancer treatment decisions. And they place themselves at a real inflection point in the field. Precision oncology moved from sequencing tumors to liquid biopsies, and now they argue into AI and pathology era. A brand new category of diagnostics. The evidence backs the claim. FDA granted Westar breakthrough device designation, and Weller has published in journals, including Journal of Clinical Oncology. It's been validated across a cohort of thousands of patients on four continents. Then investors put the promise in one line, matching the right treatment to the right patient. And here's the detail that makes Wailer so striking. The entire engineering team is just seven people. Over five years, those seven build almost the whole stack pencils. Their own slide, their own lab, their own operating system, their own clinical data tool data system. And they run there, and they're on their AI, AI hardware in their own office. Not the cloud. Building a couple that can take three to five to five. Three to five feet of training runs. So they opt in plusly for one thing. Iterations, iterations, seven inch. So sit with the contrast. Using the already exists. So why why why do both work? Because what makes them so so instructive? These are just it's we're not talking about a winner and lose. Two different two different right answers here. Tempest is betting there, betting there. Whoever owns the deep integrated data set. Data set wins. So it's worth a fortune to build it. Well R is betting the opposite. In the age of powerful AI, you don't need to fall the ocean. You can actually extract enormous value from what is already there, from cheap routine data that exists. So if you're laser focused on one right question, that's what WLR is. I notice Tempus itself validates WLR's thesis. When Tempus bought Dage, it was buying a specialist here, a company that did one thing, pathology AI, extremely well. The Empire grew by absorbing specialists, which tells you a specialist model isn't just a price here. It's often the most exit. So Tempest is actually doing really well. They went really well public in 2023. What can we learn from these companies? Even though they're targeting a very similar problem here. So what are the three big lessons we learn from them? Lesson one lesson one is how do we how we choose broad or need or need. I would say lesson one. Lesson one broad how much capital the broad path demands. The biggest fork these two faced is the same one. Are any other founder's gonna face? Go broad and own the whole stack, or go very specialized and nail one thing. Tempest went broad, genomics data, clinical data, labs and AI all in-house. Valar went very niche, one prediction pulled from the slide that already exists. And the factor that really drove each choice was capital. Going broad is enormously capital intensive. Tempest had to raise 1.3 billion to build what it built. Going niche, let Valor reach a real validated FDA-based six in 26 million. Both can absolutely win and are doing great, right? So which one, which path you choose, or which is better, it all depends on how much capital you can realistically. So let that so let that let that be some let that be something about think about to answer a question. Or I want to be niche. Second, second, second lesson. Second lesson. Data is definitely for both of them. For both of them. But no which companies the real asset is data. Building labs and costs, huge costs. Unlocks and squeezing in squeezing insight already exists. Neither is wrong. Neither is the answer. The less is to be delivered. Are you building a generation? Or are you just mining the data? Mining the data is already out there. Both can be a both can be a mode. The third big lesson is also speed of iteration beats just perfection of disguise. This is quite engineer under everything under well art. Especially when it comes to iteration. In their own words, in their own words, isn't the quality isn't the quality of speed of iteration. Speed of iteration. So they build, so build, learn, learn what's broken, learn what's rebuilt. Owning the hardware, owning the hardware, every layer, every layer, an idea, an idea, and the chip that compresses the loop. They compress the loop. Just at a just much broader scale. Broad scale. Constantly launching. Constantly launching absorbing absorbing complex. But here's the whole but here's the clearance. Just iterate towards validated evidence. Evidence. FDA clearance. FDA clearance or studies studies or um these are the these are both compliance. We're reaching out. Reaching out. In medicine, fast truth. Fast iteration. Fast iteration. And hard truth. And hard truth. We learn earn the evidence. And the capsule under the thing. Underneath all three, whichever path you choose, choose it on a purpose. The failures in this space are usually the companies that choose neither. So this is what we have learned between two companies today, which were tackling. Which were dealing with the same problem. The uncertain uncertainty in a cancer treatment decision. Cancer treatment. One is winning by owning everything, the other by owning one thing completely. And it looks like the market has room for both. But it all depends on personally what you can, what you can, what path is right. What path is right for you. Whether you're building an empire or mastering. Match your ambition, match your capital. We deliberate about iterate faster. Because that's what matters the most matters the most. That's not it's not just healthcare AI strategy. This is something which probably is needed for everyone. So thanks for listening. Thanks for listening. If you like what I'm saying, if you like what I'm saying, please follow Ialthcare and Healthcare Podcast, I'm host deep deep. Thank you. Thank you.