The Merck and Moderna drug that reached the news as a possible skin-cancer vaccine is manufactured after the individual patient is sequenced, from that person's own molecular profile.
The headlines treated it as a vaccine, which is the part that travels. Eric Lefkofsky's point is that the word hides the change: this is not one drug given to everyone, it is a drug built against one person's tumor, and it is the first real example of the thing AI in oncology was supposed to deliver.
"Yeah. What's happening is you're sequencing a patient and you're actually building the drug based upon that person's unique profile."
Lefkofsky runs Tempus AI, which he says is one of the largest users of frontier models anywhere in health care, and he has been on the investing side of the same question at Lightbank since before the models existed.
The full segment is covered here so you can skip it.
Here are the 5 insights that matter.
👤 Guest: Eric Lefkofsky, Managing Partner at VC firm Lightbank Capital, co-founder and Chairman of Groupon, and Founder and CEO of Tempus AI
🎙️ Host: Kelly Evans, who anchors The Exchange on CNBC
📰 Published: 14 September 2026 on CNBC's The Exchange
🟣 Apple Podcasts | 🔗 Episode page | ⏱️ 4 min
Key Takeaways
He expects health care to be the biggest eventual beneficiary of AI, because the data is exactly what the models are good at
Unstructured, disparate and messy is the problem description, and it is also the model's best use case
The personalized cancer vaccine is the concrete version of that claim
The patient is sequenced and the drug is then built from that person's unique profile
Immunotherapies already work extraordinarily well in some patients and not at all in others, and this is an attempt to fix the second group
Tempus has been making frontier models work at scale for three or four years across diagnostics and data
The genome was sequenced 22 or 23 years ago and did not unlock cancer the way people expected
His framing is that health care is slow and drugs are slower, not that the premise was wrong
A slowdown argument runs into the fact that the models of six months ago were already extraordinarily powerful
1. Health Care Benefits Most
Kelly Evans opened on Anthropic's expected listing rather than on health care, reading out that "CNBC has confirmed they will list on the Nasdaq. Reuters reports they seek to raise up to $100 billion at a $2 trillion valuation, with Nvidia considering investing up to $10 billion in the IPO." She introduced Lefkofsky as someone who knows the IPO game as both investor and founder, then asked whether health tech is still early days or whether real things are happening.
His answer put health care at the top of the list rather than on it: "I mean. I think it's the area that most likely will eventually be the biggest beneficiary."
The reason he gives is a fit between the problem and the tool: "Because if you think about it, all these amazing models and their capability are just perfect for healthcare where you have, you know, unstructured, disparate, messy data. And these models are really good at making sense of all that data."
He did not pretend the appeal is purely technical: "And the application is just like, what's, what's better than keeping people living longer and healthier lives?"
2. Diagnostics and Data
Evans asked what he had been excited about investing in around the time AI arrived, and about his involvement with Moderna.
Lefkofsky described a company that was already pointed this way: "Yeah, well, we've been focused on AI for a while. I mean, Tempus has two main businesses, diagnostics and data."
The business model in one line: "And basically we're in the business of using, you know, big data and AI to essentially make diagnostics smarter and to help drug companies make better drugs."
What changed was the arrival of general-purpose models rather than a change of direction: "When these big frontier models showed up, they just were like, we just were fortuitous that they were perfectly designed to help us in our two main businesses."
The work since then has a timeline: "And so we've been pretty active for the past 3 or 4 years at making these models work at scale."
The destination he names is a change in what a drug is: "The whole dream of AI and healthcare, at least as it relates to our world, which is cancer, is how do you go from targeted therapies to like personalized therapies?"
3. Personalized Cancer Vaccines
His example is the Merck and Moderna drug, which he introduced as "probably no better example" of that shift — a drug where "you're essentially building personalized vaccines."
Evans said she had seen the headlines about a possible vaccine against skin cancer but had not registered the personalization in them, and asked whether it is based on the patient's molecular profile
It is, and the sequencing comes first: "Yeah. What's happening is you're sequencing a patient and you're actually building the drug based upon that person's unique profile."
The problem it is aimed at is the unevenness of existing immunotherapy: "These immunotherapies are amazing, but they don't work in all patients where they work well, they work incredibly well where they don't work well, they don't work well."
The mechanism he describes is about making the tumor visible to the drug: "And so this is a vehicle to basically say, based on this person's own unique molecular characteristics and immune profile, how do we find some way to kind of light the tumor up so that these drugs know, okay, this is what I'm going to target."
4. The Genome, 23 Years On
Evans asked how many months or years away more announcements like this one are. Lefkofsky deferred on the specifics and then widened the frame considerably.
He was careful about what he does not know: "You know, I think obviously, I mean, Merck and Moderna know more about their portfolio than, than, than others and when, when these trials are likely to read out."
On this trial he was not hedged: "But I think the success of this particular trial was pretty dramatic." He added that the companies have intimated more are coming and suspect more will come
The comparison he reaches for is the one that has disappointed everybody: "We sequenced the human genome, like, I don't know, 22, 23 years ago. And we had all thought back then, okay, this is going to unlock the mysteries of cancer."
His verdict on that promise is honest and unresolved: "And it really hasn't in the way people thought, but it just takes a long time. Health care, slow drugs are slow."
5. What a Slowdown Costs
Evans turned the conversation to the weekend's slowdown headlines, and did it by way of his own history — she noted he started the company because of personal experience with his wife, and that for people in critical moments, delay is not abstract.
Lefkofsky's only joke in the segment was about her transition: "Good segue, by the way. Do you like that?"
Her question was genuine rather than rhetorical: she said she listens to the cancer-vaccine story and thinks it could help people dealing with this firsthand right now, and wanted to know what impact slowing down, whatever that means, would have on those efforts
His answer starts from scale of use: "Yeah. I mean, we're one of the largest users of these models in, you know, that's out there certainly in healthcare."
And the frame he sets up for answering is the capability that already exists rather than the capability that is coming: "And so I think for us, you know, if you look at the models we were using, let's say six months or a year ago, they were extraordinarily powerful."
Bonus Insights
The segment was booked off an IPO story and spent almost none of its time on IPOs. Evans introduced Lefkofsky against Anthropic's Nasdaq listing, a reported raise of up to $100 billion at a $2 trillion valuation and Nvidia weighing up to $10 billion, and then asked him about health technology instead.
The personal detail is the host's, not the guest's. Evans raised his wife's illness as the reason the company exists; he did not bring it up and did not expand on it.
His enthusiasm was consistently about the category rather than his own company. Asked what is happening in health tech, he answered about frontier models and messy data before mentioning Tempus at all.
Lefkofsky's bottom line is that the payoff from AI in medicine is already visible in one place — a cancer drug built from an individual patient's sequence rather than prescribed off a shelf — and that the models capable of that work were extraordinarily powerful six months ago, before any of the current argument about pace began.
Products, Companies & Tools Mentioned
Tempus AI (His company, built on two businesses, diagnostics and data, and among the largest users of frontier models in health care)
Merck and Moderna (The partners behind the personalized cancer vaccine he calls the best available example of going from targeted to personalized therapy)
Anthropic (The subject of the host's introduction: a confirmed Nasdaq listing, a reported raise of up to $100 billion at a $2 trillion valuation, and Dario Amodei's slowdown call)
Nvidia (Reported by Reuters, per the host, to be considering an investment of up to $10 billion in that IPO)
Lightbank (The venture firm where he is managing partner, and the investing side of the same question)
Groupon (The company he co-founded and chairs, cited as part of why he knows the IPO process from both sides)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: