Tempus AI's First Adjusted Profit Runs on an $88.9 Million Data License — While Quest Diagnostics's Revenue Per Test Fell 4.2%

The licensing engine inside that number has a name: Insights, Tempus's data licensing and modeling line, and it did $88.9 million of work at a 72% gross margin. That is 21.6% of company revenue, and it grew 29% against the 20% growth of the diagnostics base it sits on top of — $303.5 million of oncology and hereditary testing that still supplies three-quarters of revenue and all of the reimbursement risk.

Here is the arithmetic behind the profit. Data and applications was 24.3% of revenue a year ago and 26.3% in the quarter, so a business grossing 72% instead of 62% became two points heavier in the mix. Decomposed, the licensing segment's own 300 basis point margin gain contributed about 73 basis points of company gross margin, diagnostics contributed about 76, and the shift in mix added roughly 19.

Now the part that decides whether this is a business or a rounding error. Tempus AI's GAAP net loss narrowed $65.1 million, to $14.9 million from $80.0 million. The operating loss narrowed just $4.1 million, to $56.9 million from $61.0 million, and the gap is a non-cash mark: $42.8 million of unrealized gains, mostly on the stake in Personalis that the company agreed to buy for $16.25 a share in July.

This is where the print stops being a Tempus AI story. Quest Diagnostics grew third-quarter revenue 10.0% to $3.10 billion on 14.5% more requisitions while revenue per requisition fell 4.2%. Labcorp grew 6.9% to $3.81 billion, with diagnostics laboratories up just 4.4% organically and biopharma laboratory services up 9.6%. Both sell the test. Neither sells the file.

Washington sets the next price. Medicare clinical lab fees took no cut in 2026, and from Jan. 1, 2027 through 2029 payment may be reduced by as much as 15% a year against the prior year's rate. The reprieve expires inside the first quarter of 2027, when both labs must reprice contracts that already yield less per requisition — and neither has a revenue line that a pharma licensing budget can substitute for.

The supply chain explains why the margin gap exists. Upstream, Tempus AI buys sequencing, computing and minimal residual disease technology, then sells the output as licensed insight. Downstream, drugmakers pay for the right to train models on linked clinical, molecular and imaging records. The toll collector gets volume; the data owner gets margin.

That stack is the real discovery in this print. A drugmaker with roughly $50 billion of trailing revenue is paying about $36 million a year, near 0.07% of sales, for the right to train on linked clinical-genomics data. The budget is small enough to be repeatable and large enough to compound, and the holder of the data sets the price. At the Nov. 2 close of $78.48 and 182.2 million shares outstanding, Tempus AI's market value was about $14.3 billion, roughly 9.5 times its $1.51 billion of trailing revenue.

The same business model is being attempted in Asia on a different balance sheet. Lunit, the Seoul-listed AI pathology company, reported first-quarter 2026 revenue of about 24.0 billion won, up 25% year over year, with 97% of sales from outside Korea, an operating loss narrowed 35% and an EBITDA loss narrowed 54%; first-half revenue reached 45.8 billion won, up 23%. It targets EBITDA break-even within the year and FDA clearance for its INSIGHT Risk mammography model in 2026.

The next two quarters are about plumbing, not narrative. The raised guide requires $458 million to $463 million of fourth-quarter revenue, about 25% growth, to land at $1.60 billion to $1.605 billion for the year with about $65 million of adjusted EBITDA — and that guidance assumes no contribution from Personalis.

So what is proven? Not that healthcare AI is profitable: Tempus AI still lost $14.9 million on a GAAP basis and its operating loss is $56.9 million. What is proven is narrower and more useful — a licensing line worth 21.6% of revenue can grow 29% at a 72% gross margin, and that mix alone can carry a company from an $80 million quarterly net loss to positive adjusted earnings. The labs that spent two decades selling the test now face a competitor selling the file.