Tempus is still mostly a genomic testing lab, and the price only works if the data business or a reimbursement step-up turns thin cash flow into something much larger.
The thesis is free. The live trend rank on these names updates continuously in the terminal — create an account to see it.
Forward-looking commentary for a general audience, not personalised advice. Company figures are from Tempus filings and shareholder letters. Market data are StockAnalysis figures, sourced to S&P Global, as of the US close on 2 Oct 2026 unless a date is given below. Horizon: quarters to years.
Tempus still makes most of its money the old way. A doctor orders a test, a lab reads the DNA, an insurer pays. The newer story, the one in the name, is that those tests throw off a data set drug companies will pay for, and that the models trained on it make the next test more useful. The latest quarter is specific enough to show which story is in the income statement, and which one is only in the stock.
Tempus AI (Nasdaq: TEM) does two jobs that feed each other.
The first is diagnostics. Genomic sequencing, in plain words, means reading the DNA of a tumor or of a person's inherited genes so a doctor can pick a treatment or spot a risk. Inside that, oncology covers therapy selection (a broad tumor profile, often called comprehensive genomic profiling) and molecular residual disease, a blood test that looks for cancer still left after treatment. Hereditary is the inherited-risk business, mostly from the Ambry acquisition. In the second quarter of 2026, ended 30 June, diagnostics revenue was $289.3 million, up 20%, out of $382.5 million total. Oncology was $166.0 million of that, up 29%, on about 96,500 clinical tests, up 31%. Hereditary was $107.4 million, up 5%, on about 141,500 tests, up 2%. For all of 2025, diagnostics were $955.4 million of $1.272 billion, and the 10-K says oncology and hereditary tests together were 74% of revenue. The mix has not flipped.
The second job is data and applications. Tempus licenses the de-identified records from those tests, and it lets drug companies build models on the data inside Tempus's own computers rather than downloading files. The company calls the licensing and modeling piece Insights. Applications, which include care-gap software, trial matching, and algorithms that ride along with a test, are still small. Management says so. In the quarter, data and applications revenue was $93.2 million, up 28%, with Insights up 36%. That is about a quarter of the company. It was also about a quarter in 2025, when the segment did $316.4 million.
The flywheel is the reason to own both. More tests mean more records. More records make the license more valuable. A better model is supposed to make the next test more useful, which brings in more tests. In July the company said it had delivered a first oncology foundation model to AstraZeneca, trained on 2.5 million de-identified records. It also booked about $200 million of new data and applications licenses in the quarter, naming BioNTech, Daiichi Sankyo, and others alongside a list that already includes AstraZeneca, GSK, and Bristol Myers Squibb. At the end of 2025, remaining contract value was over $1.1 billion and net revenue retention on Insights was 126%. Contract value here means the signed deals, assuming options get exercised and nobody cancels early. It is a backlog of potential, not cash in the door.
Gross margin is what is left of a sales dollar after the direct cost of running the test or delivering the data. In the quarter the lab kept 62.6 cents (63.9 cents if you ignore stock pay charged to the lab). The data business kept 70.2 cents, down from 72.7 cents a year earlier. A gap of about eight points is not a software company sitting on top of a lab. And the gap narrowed for a reason management named: compute, the cost of delivering that model to AstraZeneca. The AI product was a margin drag this quarter. The CFO expects data margins back in the mid-70s for the full year. That is a forecast, not a print.
Revenue grew 22%, to $382.5 million. That is the first quarter in which Ambry is in both periods, so the growth is no longer an acquisition jumping the base. The shape inside it matters more than the headline. Oncology volume grew faster than oncology revenue, because reimbursement per test did not. Average oncology reimbursement was about $1,720, flat with the first quarter. Strip out the cheaper residual-disease tests and it rose from $1,820 to $1,850. Residual-disease volume was 9,000 tests, up from 6,500. Hereditary grew 5% because the year-ago quarter was unusually strong. Management expects that growth to settle in the mid-teens later in the year, not to reaccelerate on its own.
Reimbursement is simply what Medicare and private insurers pay per test. The lever Tempus is counting on has a clumsy name. An advanced diagnostic laboratory test, or ADLT, is a Medicare category for a newer lab test. It can start at a company-proposed rate and later reset to the middle of what private insurers actually paid. CMS granted that status to Tempus's tumor-normal xT test in July 2024, at an introductory $4,500 from 1 July 2024 through 31 March 2025. That introductory rate has already expired. In July 2026 the FDA approved a tumor-only version, so the test can run when a matching normal sample is not available. Tempus says it is the first lab with companion-diagnostic approval for both versions, and that the approval lets the rest of the tissue portfolio move onto ADLT pricing. Management's own bridge, which is an estimate, is about $200 more per test and about $85 million a year starting in 2027. If the liquid biopsy, xF, is approved and on the market in the back half of 2027, they add another $550 and talk about oncology reimbursement, excluding residual disease, moving from about $1,850 toward about $2,600, and something like $400 million of extra revenue in 2028. None of that is in the $1,720 yet.
The profit line needs the same care. GAAP net income was $5.6 million. Inside it sit $55.6 million of stock pay and related payroll taxes, and a $98.5 million unrealized gain on marketable securities, which the CFO ties to the Personalis stake. Adjusted EBITDA, profit before interest, taxes, and depreciation, with stock pay, deal costs, and those mark-to-market swings taken out, was $8.0 million, against a $5.6 million loss a year earlier. The company's own non-GAAP net loss was $7.7 million. The quarter did not become a profit in the way a screen of "net income turned positive" suggests. It became a smaller operating loss, plus a gain on a stock they have agreed to buy.
Guidance is $1.595 to $1.605 billion of 2026 revenue, about 25% above 2025's $1.272 billion, and about $65 million of adjusted EBITDA. The guide excludes Personalis. The first half produced $730.6 million of revenue and $5.2 million of adjusted EBITDA. On the midpoint, the second half is about $869 million of revenue, on our arithmetic, and something like $60 million of the year's adjusted EBITDA. The sales step is ordinary. The profit step is not. They need a margin jump, not just more tests.
Cash is better than it was, and less simple than the headline. Operating cash use fell to $7.5 million in the quarter from $73.3 million in the first quarter. The first half still used $80.8 million of operating cash, plus $14.3 million of equipment and $6.8 million of capitalized software. The CFO expects the company to be cash flow positive in the fourth quarter, helped by retiring a loan from Ares and by data deals that now pay quarterly instead of as large prepayments that then burn off. They also say the new convertible saves more than $30 million a year of interest. Trailing free cash flow on StockAnalysis is still negative $263 million, which includes the weaker half before this one.
The balance sheet can fund a wait. Cash, equivalents, and marketable securities were $820.7 million on 30 June. About $216 million of that is equity securities, not cash in a bank. Spendable cash and equivalents were $600 million, plus $4.7 million restricted. Against that sit $750 million of 0.75% convertible notes due 2030, $460 million of 0% convertible notes due 2032, and a Google cloud note carried at $188 million that shrinks as Tempus uses Google's cloud. StockAnalysis puts total debt near $1.44 billion and enterprise value at $14.45 billion. Shares outstanding are 180.4 million, up about 5% in a year. That is the dilution already happening, mostly stock pay.
The converts are a claim on later years, not an automatic share issue. The 2032 notes convert at about $69.26, and the company bought a hedge that offsets dilution up to $98.94, covering about 6.6 million shares. At the 2 October close of $76.63 the shares sit above that conversion price and below the cap. Conversion ahead of 2032 is not automatic. It generally requires conditions, including the stock at or above 130% of the conversion price, about $90 on our arithmetic, for 20 trading days in a 30-day stretch. The 2030 notes convert at about $84.19, with a hedge up to about $111, so they are still out of the money. The nearer dilution question is how Personalis gets paid.
On 20 July Tempus agreed to buy Personalis, whose residual-disease test it already distributes, at $16.25 a share, about a $1.5 billion enterprise value. The deal is stock, with an option to pay up to half in cash, a maximum exchange ratio of 0.3356 Tempus shares, and a collar that is not set until just before closing. Management has said they would rather borrow than issue the whole thing, and that the close is late 2026 or early 2027. The 2026 guide assumes it does not happen this year. Only about 10% of the sales force is selling the residual-disease test today.
The close on 2 October was $76.63. Market value was $13.83 billion. Enterprise value was $14.45 billion, about 10 times trailing sales and, on our arithmetic, about 9 times the midpoint of this year's revenue guide. Set that enterprise value against about $65 million of guided adjusted EBITDA and you are at roughly 220 times that profit. Trailing free cash flow is negative.
A reverse DCF starts from the price and works backward to the growth and profit that price needs, instead of guessing a future and hoping the market agrees. Ours is simple on purpose. A 9% discount rate, growth that fades to 3% after year ten, and a constant share of revenue that becomes free cash flow, compounded off the $1.60 billion midpoint of this year's guide. On that arithmetic the enterprise value needs about 25% annual growth for a decade if only 10 cents of each sales dollar becomes free cash, which would be a strong lab. It needs about 20% growth at a 15 cent cash margin. It needs about 13% if a quarter of each dollar becomes free cash, which is closer to a data business than to a lab. The outputs are approximate. They are a sanity check, not a model, and they are ours, not the company's.
Read against the guide, the price is a fork, not a verdict. Today's plan is about 25% growth and an adjusted EBITDA margin near 4% of sales, with free cash flow still negative. The price can be earned by a lab that keeps something like today's growth for ten years and learns to keep a dime. It can be earned by a data company that grows more slowly and keeps a quarter. It cannot be earned by the profit Tempus has actually put in the guide.
Natera, the large residual-disease testing company, is the useful comparison, and it cuts against the idea that Tempus is already priced as a pure AI platform. Natera closed the same day with a $59 billion market value on $2.7 billion of trailing sales, about 22 times sales, a 65% gross margin, and positive free cash flow. Tempus is the cheaper sales multiple, and the one that still burns cash. The market is not paying a premium to the best lab. It is paying a discount that still assumes cash Tempus does not yet produce.
A screen that sorts on a first GAAP profit, a 64% gross margin, and 22% growth will file the quarter under "healthcare AI inflection." The same documents argue with that.
The $5.6 million of net income includes a $98.5 million gain on Personalis shares, the company Tempus had agreed to buy ten days earlier. The company's non-GAAP net loss, which removes that gain and the stock pay among other items, was $7.7 million. Of the $821 million headline for cash and securities, $216 million is that kind of equity, not spendable cash. And in June, Pathos, a related party, settled $8.8 million of data-license fees in its own preferred stock. A slice of the data business is collected in paper.
The margin tell is the other skip. Data gross margin fell because the foundation model cost money to deliver. Volume in oncology ran ahead of price. Growth is still "do more tests," not "get paid more per test." The reimbursement bridge is a letter, not an average selling price.
Customer concentration is the one cartoon that does not fit. The 10-K says no customer was a significant portion of 2025 revenue. That is the right answer to "one pharma client is the whole company." The finer point is inside the data segment. Related-party data revenue was $22.0 million in the quarter, about a quarter of the $93.2 million. For 2025, related-party revenue was $65.3 million in total. The SoftBank joint-venture data license ended this quarter, and Insights still grew 36%. That is evidence the book is broader than the one contract a short seller fixated on. It is not evidence that related parties have become a rounding error.
The strongest case against the stock does not need a plot. It needs the mix.
Three-quarters of revenue is testing. Hereditary just grew 5%, and the comp it is lapping was the hot one. Oncology is adding tests faster than dollars, and the Medicare category people point to is a mechanism that resets to private-insurer payments, not a permanent $4,500 price. The $4,500 figure was an introductory rate that ended in March 2025. Data margins are lower than a year ago, and the applications that actually look like software are still, in management's words, relatively small. A business can be a good lab and a promising data set and still be a lab.
Spruce Point Capital published a strong-sell report on 28 May 2025. They were short. They argued the AI story was thin, that a SoftBank joint venture looked like capital recycled into revenue, that AI applications were about 2% of revenue, and that the shares could fall 50 to 60 percent, to a range they put at $26.35 to $32.95. The shares closed at $76.63 on 2 October 2026, well above that range. The specific license they flagged ended this quarter, and Insights still grew, which undercuts the round-trip as the whole story. It does not undercut the valuation. Short interest was still 28.1 million shares at the 15 September 2026 settlement, about 26% of the float on ChartExchange's FINRA-based figure. A lot of investors are still paying to be on that side.
The rest of the bear is the capital structure. Convertible principal is $1.21 billion before the Google cloud note. Personalis can be paid in shares, and management's preference for debt is a plan, not a closed financing. The lab relies on Illumina as the sole supplier of its sequencers, about 33% of vendor payments in 2025. The founder controlled about 58% of the vote at the end of 2025 through high-vote shares. None of that stops the tests from being ordered. It means the equity is a long-duration claim on reimbursement, on data contracts that have to stay broad, and on a deal that can issue stock, with a controlling shareholder on the other side of the table.
We would take the platform story seriously if three things showed up together. Data becomes a larger share of revenue and its margin expands, while related-party fees and fees paid in stock shrink. Oncology reimbursement per test moves toward the bridge in the shareholder letter, not just the volume. And the second-half profit, plus fourth-quarter cash flow, arrives without a fresh equity raise.
We would drop even the "interesting" label if adjusted EBITDA stays near the first-half run rate, if Insights growth fades now that the SoftBank license is gone, or if Personalis is paid for mostly in shares and residual-disease volumes do not follow the wider sales force. A soft quarter in hereditary, or one more quarter of compute cost in the data margin, changes the timing. It does not, by itself, decide which company this is.
StockAnalysis estimates the next report for 3 November 2026, after the close. The company has not posted that date in the materials we used, so treat it as a calendar, not a confirmation. That print is the one that has to start closing the gap between $5 million of first-half adjusted EBITDA and a $65 million year. We will read oncology reimbursement per test, the data gross margin, Insights growth without the SoftBank license, and whether operating cash is actually heading toward the fourth-quarter surplus the CFO described.
After that, the tumor-only approval has to show up in the price per test. xF is in front of the FDA, with management hoping for a launch in the back half of 2027, and a PMA submission for the RNA assay is in progress. Management also says CMS has proposed a "Software as a Medical Service" category that would matter for the algorithm products, if it becomes a real payment path. Personalis is supposed to close in late 2026 or early 2027, and the mix of cash, debt, and stock matters more than the $16.25 headline. Bookings were about $200 million this quarter. The tell is whether that pace holds once the contracts have to become revenue.
Tempus is interesting, and it is not a core holding. The flywheel is visible enough to respect: oncology volumes are accelerating, the data business grew after a large license rolled off, a first model is in a drug company's hands, and management has sized a reimbursement step-up in dollars. The income statement has not caught up. The equity is still a genomic lab, with about a quarter of sales in data, a quarterly profit that was a mark on the company they are buying, and a price that only works if one of those two businesses becomes much more profitable than it is. Our terminal reads the name the same way: tradable rather than core, scored as very expensive, and extended, in a healthcare sector that is weakening even while the stock has been stronger than its group. It moves up the list if the fundamentals improve, meaning the cash and the price per test, or if it starts to lead that group for a reason that shows up in the numbers. The risk in the price is that the data company arrives late, or not at all, while the lab keeps doing the earning.
Every name we write about is scored continuously in the terminal — market regime, sector rotation, fundamentals and technicals resolved into one Conviction × Timing verdict.
Open the terminal