Tempus AI

IThe business

Tempus does two things. It runs laboratories that test cancer patients, mostly by sequencing tumours, and it licenses the de-identified data from those tests to drug companies. It calls the first Diagnostics and the second Data and applications. Diagnostics was $854M of revenue in the first nine months of 2026 and Data and applications was $288M, or 75% and 25%.³ Total revenue was $1,271.8M in 2025, 83% more than in 2024, and $1,509M in the twelve months to September C (calculated).²³ The founder, Eric Lefkofsky, still runs it from its headquarters in Chicago, and the 10-K says it ran its first clinical test about ten years ago. It listed on Nasdaq on 14 June 2024. It runs five labs, in Chicago, Atlanta, Raleigh, Aliso Viejo and Minneapolis, and had more than 3,800 employees at the end of 2025.²¹

Diagnostics

Oncology is the original business. Tempus sequences tumour tissue (its xT test), blood (xF), RNA and the whole exome, and now tests for minimal residual disease, which is cancer left after treatment. It ran about 340,500 oncology tests in 2025, at an average revenue of $1,600 a test, up from $1,510 in 2024, mainly because Medicare paid more R (reported).² In the first nine months of 2026 it ran about 283,500, up 30% on the year before. In the third quarter oncology volume was up 32%, and about 12,000 residual-disease tests were run, up 300% on a small base.⁴³

Hereditary testing arrived with Ambry Genetics, bought on 3 February 2025 for $375.0M in cash and 4,843,136 shares valued at $61.54 each, which is about $670M C (calculated).² Ambry ran about 460,500 tests in 2025 at roughly $770 each. That is 37% of 2025 Diagnostics revenue by my arithmetic, against 57% for oncology C (calculated). Hereditary is the slower line: 143,000 tests in the third quarter against 140,000 a year before, and revenue up 3%.⁴³

Diagnostics is paid by insurance. Tempus bills Medicare, Medicaid and commercial insurers, and the claim is usually paid 30 to 120 days after the service. Because it has to estimate what it will collect, part of the revenue it books is a forecast. The 10-K says that at the end of 2025 it had been paid for about 55% of clinical oncology tests and about 50% of hereditary tests done in 2023 and 2024.²⁴ Receivables were $380.7M at 30 September, 85 days of the third quarter’s revenue, against 78 days at the end of 2025 and 71 days a year earlier C (calculated).³² That is not a warning by itself, since the mix changed when Ambry joined and I can’t tell how much of the longer wait is mix. It does mean growth in revenue arrives before growth in cash.

Data and applications

The second line has four products. Insights licenses de-identified patient records and gives customers analysis tools; it was $237.9M of the $288.2M in the first nine months, or 83%.⁴ Trials matches patients to clinical trials and runs some trials as a contract research organization, at $28.5M. Next is software that flags patients whose care has gaps, at $21.6M. The three add up to all but $0.2M of the line; Algos, the algorithm-based tests, have no revenue line of their own in the 10-Q.⁴ So when this page says data revenue, it mostly means Insights: the licence of records, and the work around them.

The large customers are named in the 10-Q. AstraZeneca has agreed to spend at least $220M on Tempus products through 2028, and can raise that to $320M. GSK has agreed to spend at least $180M through 2027, of which $70M was paid when it signed. Recursion is paying $118M over the term to November 2029, part of it in its own shares if it chooses. Bristol-Myers Squibb signed a data licence in October 2026 for $137.5M of fixed payments to the end of 2031, which it can end on twelve months’ notice and a substantial termination fee.⁴ The company says it has signed more than $550M of data licensing contracts so far in 2026 R (reported).³ I treat that as contract value. It is not revenue, and it is not cash.

Three more facts change how I read this line. Insights growth slowed through 2026: 44.1% in the first quarter, 36% in the second and 29% in the third.¹²¹³³ Related-party revenue fell at the same time, from $31.2M in the third quarter of 2025 to $22.1M. Take out Pathos AI and SB Tempus, the two related parties as the accounts define them, and the rest of Data and applications was $86.0M in the third quarter against $50.2M, up 71%, and $216.3M in nine months against $155.9M, up 39% C (calculated).³ So part of the slowdown is the two related parties shrinking, and the rest of the line is growing faster than the headline. The split is not clean: Pathos builds the AstraZeneca foundation model, and BMS uses Tempus data on Pathos’s platform.⁴ Last, the company reported Insights net revenue retention of 140% at the end of 2024 and 126% at the end of 2025, with remaining contract value of $940M and then more than $1.1B, of which about $300M is optional, and the figure assumes no one cancels.¹⁴¹⁵² Retention of 126% means the same customers spent 26% more than the year before, a smaller step up than a year earlier. I found no new figure for either in the 2026 releases.

How the money works

A test costs Tempus much less than it earns, and the rest of the business costs more than the tests earn. In the nine months, Diagnostics had a gross margin of 62.1% and Data and applications 71.0% C (calculated). Research and development took 25% of revenue and selling, general and administrative expense took 59%. Stock pay across all lines was $155M, $101M of it in selling and administration.³ The first table in this section covers the tests and prices. The second shows the costs. Stock pay rose from 8.8% to 13.6% of revenue in a year, which is why a loss that looks as if it is shrinking on the company’s own adjusted measure is not shrinking much on mine.

The latest quarter

Revenue in the third quarter of 2026 was $411.5M, up 23%. Diagnostics was $303.5M, up 20%, and Data and applications $108.1M, up 33%, with Insights up 29%.³ Gross profit was $266.7M. The loss from operations was $56.9M, against $61.0M a year earlier, and the net loss was $14.9M after $42.8M of unrealized gains on shares Tempus holds in other companies and $50.8M of stock pay and payroll tax on it.³ Adjusted EBITDA was $25.6M. The company raised its revenue guide for 2026 to $1.60B to $1.605B and kept its adjusted EBITDA guide at about $65M.³ It had $854.4M of cash and securities.³

The same release says Tempus cleared antitrust review for the purchase of Personalis, a company that makes a tumour-informed test for residual disease, and that Personalis holders vote on 30 November 2026.³⁴ I leave the purchase out of the valuation and show it separately in Section VI, because it has not closed.

IIThe moat

I ask what keeps a rival from doing the same thing at the same price. For Tempus there are three answers that are real, and three that people assume and that I don’t see in the filings.

The three that are real

The data library. Since 2016 Tempus has collected about 1.4 billion documents across more than 8.6 million de-identified patient records, with imaging on more than 8 million, matched clinical and genomic records on more than 1.5 million, and full RNA profiles on more than 360,000. It stores more than 450 petabytes. The records come in through more than 700 data connections to more than 5,000 healthcare sites.² Every oncology test it runs adds a linked record, which a company that only licenses data cannot do. The 10-K says Tempus keeps the right to license the de-identified data. That is the scarce asset.

Tests that regulators and payers have approved. Its xT test has FDA premarket approval, which it got in April 2023, and Advanced Diagnostic Laboratory Test status, a Medicare pricing category.² In the second quarter of 2026 the FDA approved xT Tumor Only, which the company says will move its tissue testing to ADLT pricing.¹³ Medicare paying more is the main reason revenue per oncology test rose from $1,510 to $1,600 in 2025.² Each approval is something a rival has to earn again for its own test.

Doctors and hospitals that already send samples. More than 9,000 physicians have used Tempus, and hospital systems keep signing: Northwestern Medicine picked it in early 2026 to widen testing across its health system.¹⁵¹² Tempus sells to oncologists through its own field sales force, which a rival has to build too.²

The three that people assume

Sequencing. Reading DNA is a service many labs sell. The 10-K names Foundation Medicine, which Roche owns, Caris, Guardant, Natera and NeoGenomics as rivals in oncology, and Quest and Labcorp as possible competitors.² The company says it is one of the largest sequencers of cancer patients in the US, but nothing about sequencing is exclusive.

Exclusive data. The licences are non-exclusive. AstraZeneca, GSK, Bristol-Myers Squibb and Recursion all license from Tempus, and each can license from Flatiron, IQVIA or ConcertAI as well.⁴² A record that many drug companies can license is priced on how many questions it answers, not on who else lacks it.

Artificial intelligence as a barrier. Models improve for everyone. The AI software line, Next, was $21.6M of the $1,142M of nine-month revenue, about 2% C (calculated), and the Algos tests have no line of their own.⁴ The name of the company says AI. The revenue says lab tests and data licences.

What the numbers show

If the position were weak, price and margin would show it first. They do not. Diagnostics gross margin rose from 46.1% in 2024 to 59.6% in 2025 and 62.1% in the first nine months of 2026, helped by Ambry’s mix and higher Medicare rates C (calculated).²³ Data and applications gross margin has stayed at 71% to 72%, and Insights grew 29% in the third quarter.²³ No customer is a significant share of revenue.² These are signs of a position, but a young one. Tempus has been public for two and a half years, and I have no record of how its data business behaves when drug companies cut spending.

What matters is what drug companies will pay for the data. If drug companies pay more per record as models improve, the library is valuable. If they pay less because many sellers have similar records, the library is a cost. I can’t tell which from one customer’s contract. I come back to this in Section VII.

IIIThe competitive advantage

The question here is how Tempus wins an order today. I see three reasons in the 10-K, and I check each against what is reported.

One order, several tests. A physician can order DNA, RNA, liquid and residual-disease testing from one lab, and the report ties to the patient’s other records. The 10-K says many of its tests are connected to clinical data in a way other labs’ tests are not.² The evidence is volume. Oncology tests rose from 217,500 to 283,500 between the first nine months of 2025 and 2026, up 30%, and the growth rate rose each quarter this year: 28% in the first, 31% in the second and 32% in the third.¹²¹³³

Results that flow back. Tempus built data connections to and from doctors’ record systems, so results can return into the chart and the chart’s data comes back to Tempus. That makes ordering convenient and feeds the library. The company says it has more than 700 such connections.² I can’t see retention by physician, so I can’t say how sticky it is.

Two kinds of customer. Drug companies license data and also pay for trial matching, and Tempus can offer physicians the trials. The AstraZeneca and GSK agreements cover licensed data, sequencing and trial matching, and the 10-K says their minimums can be met with any Tempus product.⁴² A customer that spends across products is harder to lose.

Where it is not ahead

Hereditary testing is the plain example. Tempus bought a large business there and it is growing slowly: tests up 2% and revenue up 3% in the third quarter.³⁴ The same is true of trial matching, which was $28.5M in nine months, down from $28.7M.⁴ Residual-disease testing has established rivals too. The 10-K lists Natera and Guardant as competitors as Tempus expands into monitoring, and Tempus is paying $1.5B to acquire a partner there.²⁹

What I can’t judge

I can’t judge market share. The 10-K does not give it, and I did not purchase a third-party estimate. I can’t say whether the growth is taken from rivals or comes from a market that is growing for everyone. Tempus’s oncology volume is up 30%. I did not open the rivals’ filings to compare.

IVManagement

I follow the incentives. How are the people at the top paid, what does that pay them to do, how much of their own money is in, and what did they do with other people’s money?

Who runs it

Eric Lefkofsky, 56, is the founder, chief executive and chairman. Before Tempus he co-founded Groupon in 2008, was its chief executive from August 2013 to November 2015 and chaired its board from 2015 to 2020. He also co-founded the venture firm Lightbank, InnerWorkings, Mediaocean, Echo Global Logistics and Pathos AI.⁷ The chief financial officer is James Rogers, Ryan Fukushima is chief executive of the Data business, and Tom Schoenherr has run Diagnostics since February 2026.⁷ Tempus is a founder’s company, and that cuts both ways: one clear owner, and few people who can say no.

Control

Mr Lefkofsky owns all 5,043,789 shares of Class B stock, which carry 30 votes each, and 37.3M Class A shares. At 27 March 2026 that was 21.4% of the Class A shares and 57.9% of all votes.⁷ He has sold 1.85M shares since then, so his vote is now about 57% C (calculated).⁸⁷ The Class B shares turn into Class A if he stops serving as an executive or director, if he and his entities hold fewer than 10M shares, or after June 2044.⁴ In August 2025 the company moved its incorporation from Delaware to Nevada, and I draw no conclusion from that.⁴ No buyer can take Tempus without his agreement, and no vote can pass over his objection.

Who owns it

Alphabet Inc.’s Form 13F-HR for the quarter ended 30 September 2026, filed 6 November 2026, lists 1,551,102 Tempus Class A shares, valued in the filing at $127.0M and attributed to Google LLC, one of the managers it includes.¹⁷ That is 0.9% of the Class A shares C (calculated). Alphabet’s four earlier filings show the same count, so it held 1,551,102 shares at each of the last five quarter ends.¹⁸ ARK Investment Management’s filing for 30 June 2026 lists 10,022,634 shares valued at $580.6M, up from 7,175,767 at 30 September 2025; the Q3 filing was not yet on EDGAR.¹⁹²⁰ The proxy lists Baillie Gifford at 5.6% and the Vanguard Group at 5.4%.⁷ A holding is not a view, and I don’t read these as one.

How they are paid

Mr Lefkofsky’s pay for 2025 was $41.5M, of which $0.8M was salary and $40.0M was stock awards.⁷ The awards are 750,000 performance share units at target. Half depend on compound revenue growth over one, two and three years from 1 January 2025, and half on Tempus’s shareholder return against the Nasdaq Composite over the same periods.⁷⁴ His 2024 award of restricted stock units was worth $27.75M and vests in 20 quarterly installments; 525,000 units were still unvested at the end of 2025.⁷⁴ His 2025 bonus had two goals, $900M of revenue without Ambry and positive adjusted EBITDA. Revenue was $909M. Adjusted EBITDA was minus $7.4M, so that half paid nothing, and the committee still paid 90% of the $800,000 target, $720,000, in vested stock units, saying the profit goal would have been met without unplanned acquisition costs.⁷ The other four named officers received $3.6M to $10.7M each in stock awards.⁷

That pays for revenue and for a rising share price. It does not pay for cash, margin or a lower share count. A revenue measure also pays for revenue that is bought: Ambry added $363M of Diagnostics revenue in 2025 for a price of about $670M.² The first yearly test of his award shows it: the committee certified 82% revenue growth for 2025 against a bar of 30%, with Ambry in the number, and Mr Lefkofsky earned 250,000 units, the full first third of his award.⁷ I don’t say the awards are wrong. I say that they point at growth, and growth is the one thing this company has shown it can produce.

How much of their own money is in

A lot, and some of it has come out. At 27 March Mr Lefkofsky held 42.3M shares.⁷ From January 2025 to October 2026 he sold 10,715,394 shares for $660.2M, an average of $61.61, the later ones under a Rule 10b5-1 plan adopted on 4 March 2025. The 2026 sales were 2,366,013 shares for $143.3M, of which 1.85M shares came after the proxy date, which leaves about 40.4M shares, worth $3.4B at today’s price C (calculated).⁸⁷ The first three months of 2025 took 6.1M shares. Most months since then were 166,250 or 332,500 shares, and July to October 2026 brought larger lots, 250,000 to 450,000 a month. Four of the other named officers sold $20.0M in 2026. I found no open-market purchase by any officer or director in 2025 or 2026.⁸

Selling is not a verdict. Mr Lefkofsky still holds about 40M shares, which is far more than he has sold, and holding that much in one stock is a reason to diversify. I record the sales because a buyer at this price is on the other side of them.

Dealing with its own people

Pathos AI is the part to read slowly. Mr Lefkofsky co-founded it and is its executive chairman. Mr Fukushima sits on its board and was its interim chief executive from April 2023 to May 2025. Four of Tempus’s directors, Scott Gottlieb, Peter Barris, Eric Belcher and Ted Leonsis, are investors in Pathos.⁷ Tempus holds a warrant for 23,456,790 Pathos shares, about 15% of the company, at $0.0125 a share.⁴ Pathos is also a Tempus customer: it supplied $65.7M of data revenue in the first nine months of 2026, $22.1M in the third quarter alone.⁴ In April 2025 Pathos agreed to pay Tempus $200M over three years for a data licence, up to half in Pathos preferred stock, while Tempus paid Pathos $35M, $25M in July 2025 and $10M in October 2026, for Pathos’s commitments in the same arrangement. AstraZeneca paid Tempus a $35M fee in the same series of agreements.⁴ In June and August 2026 Pathos paid $17.5M of its fees in preferred stock.⁴

The company has a written policy that its audit committee reviews related-person deals, and the 10-Q discloses these ones clearly.⁷⁴ I can’t say the terms are not at arm’s length, and I have no evidence that they are not. A reader needs to know that one customer is the chairman’s other company, that part of its payment can be in its own stock, and that in October 2026 a data licence for Alzheimer’s records was signed with Pathos, with Bristol-Myers Squibb as a sublicensee.⁴ The 10-Q says that Pathos’s chief executive was also engaged by Tempus as a contractor, paid in stock units, and that the disinterested directors approved his 175,000 performance units on 22 October 2026.⁴ Mr Lefkofsky also holds 150,000 Personalis shares in his own name, which the merger prospectus reports next to Tempus’s 13.0M.²¹

What they did with the money

The cycle is short, so the record is short. Since the June 2024 listing Tempus raised $382M in the offering, $195M by selling stock into the market in 2025, $726M and $443M from two convertible note issues, and borrowed $300M on a term loan and revolver to help pay for Ambry, which it has since repaid.²⁴ It spent about $670M on Ambry, $102M on Paige, about $17M on Deep 6 and $19M on Center Health, and it agreed to pay $1.5B for Personalis.²⁴⁹ It has not run a buyback or paid a dividend on its common stock. The share count rose from 162.1M at the end of 2024 to 182.2M at 30 September, up 12.4%, and the Personalis purchase would add about 20M more C (calculated).²⁴²¹

On guidance the record is mixed. Revenue guides have been met or beaten, and the 2026 revenue guide has been raised three times in small steps. The adjusted EBITDA guide for 2025 was missed by about $12M against the first one, and the 2026 guide of $65M has not moved since February. For 2026 it needs about $34M in the fourth quarter alone, after $30.8M in nine months.^1422153

VFinancials

Ten years of history are not available. Tempus listed in June 2024, and the filings I could open give revenue from 2020, costs from 2022 and a stock-pay figure only from 2024. I show what exists and don’t pad it.

Growth, and how much of it was bought

Revenue grew from $188.0M in 2020 to $1,271.8M in 2025, and to $1,509M in the twelve months to September C (calculated).¹¹²³ The first two years had help from COVID-19 testing, which was 47.6% of 2020 revenue and 36.7% of 2021. It was 0.5% by 2023, when Tempus had stopped offering the test.¹¹ Growth in 2025 of 83% looks like a different business, but Ambry added $362.7M of the $578.4M increase. Without it, 2025 grew 31% C (calculated).² Growth in 2026 is 26% for nine months, and the guide implies 26% for the year. Part of that is still Ambry, which joined on 3 February 2025, so January 2026 is a month it did not have a year before. At the $362.7M it made in eleven months, that month is about $35M, which puts organic growth near 22% for nine months and 23% for the year E (estimate). That is the number I would build on.

The loss is shrinking slowly, and by less than the adjusted figure suggests. Operating margin after stock pay and before acquisition amortization was minus 22.5% in the first quarter of 2025 and minus 9.8% in the third quarter of 2026. Those two quarters are not a straight line. The second quarter of 2026 was worse than the second quarter of 2025, and the fourth quarter of 2025 was minus 12.1% against minus 9.8% now.³¹³¹⁵ Over nine months the margin was minus 14.6% against minus 16.2%, a gain of 1.6 points C (calculated). The adjusted EBITDA margin gained 4.9 points in the same time.

What adjusted EBITDA leaves out

Adjusted EBITDA adds back interest, taxes, depreciation, amortization, stock pay, payroll tax on stock pay, acquisition costs, gains on investments and some other items.³ In the nine months of 2026 it was plus $30.8M, against minus $20.3M a year earlier, a gain of $51.1M. Stock pay over the same period rose from $79.4M to $154.9M, a rise of $75.5M C (calculated).³ So every dollar of improvement in the adjusted figure was matched by $1.48 more of stock pay that the figure ignores. I put stock pay back in, because it is a real cost: the company pays its people in shares, and I get a smaller share of the company for it.

The bridge below starts from the company’s figure and ends at what an owner could take out of the business: operating cash flow, less everything spent on equipment and software, less stock pay. It is minus $351M for the twelve months to September, against plus $44M of adjusted EBITDA, a gap of $395M C (calculated).

Operating cash flow itself has improved. It was minus $181.3M in the first nine months of 2025 and minus $71.6M in the same months of 2026, and the second and third quarters of 2026 together were slightly positive, plus $1.7M C (calculated).³¹³ Some of that came from timing. Accrued expenses added $58.5M to cash in 2026 against a $5.2M use a year earlier, and receivables took $71.8M.³ I treat the improvement as real and partly borrowed from suppliers.

One item deserves its own paragraph. Tempus pays for part of its Google Cloud use by reducing what it owes Google under the convertible promissory note. The 10-Q says the reductions are booked against cloud and compute spending in selling, general and administrative expense. They were $32.7M in the first nine months of 2026, $22.7M a year earlier and $32.0M in all of 2025.⁴² The credit is not cash. It lowers reported cost by the same amount and lowers what Tempus owes Google, and nine-month adjusted EBITDA of $30.8M includes it. It is worth about 2.9 points of margin over the nine months, and it ends when the note is gone. The cash flow statement has no separate line for it, so I take operating cash flow as printed. I count the note as debt in Section VI and take reported margins as given, so I count it once.

How it has been paid for

A business that uses cash needs someone to supply it. The record since the listing is below. About $1.17B net was raised from the two convertible issues, $382M from the offering and $195M from stock sales. There are no buybacks.²⁴ The bank loans with Ares Capital were repaid in July 2025 and May 2026.⁴

Both convertible notes are close to the money. The price of $84.44 is above the $69.26 conversion price of the 2032 notes and just above $84.19 for the 2030 notes.⁴ If holders convert, 15.55M shares are issued, 8.5% of the shares outstanding, less what the capped calls offset up to $98.94 and $111.20.⁴ In my valuation I treat both notes as debt up to those caps and count extra shares only above them. That matters only in the bull case. The Google note comes due on 31 December 2030 and Tempus can pay up to half the balance in stock.⁴

The company reports $854.4M of cash and securities, which includes $4.8M of restricted cash. Without that it is $849.6M, and $590.5M of it is cash. At a loss of $350M a year, that is 2.4 years of money if the securities are sold C (calculated). The company says it has enough for at least twelve months and may raise more.⁴ It has an unused $300M stock program. If the Personalis purchase uses cash for half the price, about $0.8B would go, nearly all the cash and securities on the balance sheet.⁴²¹ I count that as the main capital need after the loss itself.

Shares

Tempus has 182.2M shares outstanding, 177.2M Class A and 5.0M Class B, up from 178.3M at the end of 2025 and 162.1M at the end of 2024, a rise of 12.4% in 21 months C (calculated).⁴² Another 7.9M restricted and performance units are unvested, and each January the 2024 plan tops its share pool up to 5% of the shares outstanding, subject to board limits.⁴ The stock pay is $200M a year. At today’s price that is about 2.4M shares a year C (calculated). The Personalis purchase would add about 19.8M at today’s price, 10.9%, and as many as 34.5M if the price falls to $48.42 or less C (calculated). The prospectus states a maximum of about 38.9M, which appears to count all 108.4M Personalis shares, including the 13.0M that Tempus owns and the merger cancels; I leave those out.⁹²¹

VIValuation

First, how this goes badly wrong

Before the numbers I asked what would most likely hurt an owner here, and worked back. Three things stand out.

The company does not reach real profit. It loses $351M a year after stock pay and capital spending, and its adjusted measure hides that. If operating margin after stock pay stalls near zero and the company must keep paying people in stock, an owner is diluted every year and the $1,210M of notes in 2030 and 2032 must be refinanced. Each point of 2036 margin is worth about $1.80 a share in my base E (estimate). That is the main reason for the $3 bear value.

The data revenue does not repeat. A quarter of Data and applications revenue is the chairman’s other company and a joint venture. The 10-K says customers can end contracts for convenience, and only $514.0M sits in multi-year contracts with no early-termination clause, about 44% of it due to be recognized in the next year.²⁴ Data is 25% of revenue and 28% of gross profit.³ If the largest buyers pause, the loss lands on the highest-margin line. I can’t put one number on it, because the company reports one segment and gives no profit by line.

Medicare or the competition lowers the price of a test. Oncology revenue per test rose $90 in 2025, mostly from Medicare rates.² A 10% cut in that revenue is about $55M a year at 2025 volumes, and nearly all of it would fall to profit C (calculated). That is large against a business whose operating loss after stock pay is about $212M.

None of these needs anything unusual. They set my bear case, and they are why I start from the cash that reaches an owner and not from the adjusted figure.

What an owner could really take out

I value cash after stock pay and after the capital needed to stand still. Operating income after stock pay is the starting point, because stock pay is in it. I take out the amortization of acquired intangibles, since it is not cash, and I count the purchase prices instead as capital spent in the past. From it I take tax, add back depreciation, and subtract capital spending, software and the cash tied up as receivables grow. Standing still costs little here: about 3% of revenue goes to equipment and software, a little more than depreciation. The cost is in the receivables and in stock pay.

The twelve months to September lost $351M of owner cash, against a net loss of $189M and plus $44M of adjusted EBITDA C (calculated). In my base, owner cash is positive in 2029, $180M in 2030, and $553M in 2036.

What the price implies

Put the other way, what does $84.44 need? At a 10% discount rate and 3% growth after 2036, and holding my base margins, it needs $19.8B of revenue in 2036, which is 12 times the 2026 guide and 28.6% growth a year for ten years. At my bull margin of 28% it needs $9.5B, 19.5% a year. On my base revenue it would need an operating margin of 51%, which is more than three times my base and above the 43% operating margin Palantir reports.²³

What I can’t judge

• How much data revenue repeats. The company says it signed more than $550M of data licences in 2026 and the 10-Q shows $514.0M that cannot be cancelled. The 10-K says customers can end contracts for convenience. The company gave Insights net revenue retention of 140% at the end of 2024 and 126% at the end of 2025, and I found no later figure; it compares spending by the same customers, and it is not a count of renewals. A quarter of data revenue is related-party.³⁴¹⁴¹⁵
• Profit by line. Tempus reports one segment. I can see gross profit by line but not what each line costs to sell, develop or support.⁴
• Future Medicare prices. The ADLT rate for the tests moving into that category in 2026 is not in the filings I read.¹³
• What AI does to the price of data. More buyers with models that need data can raise the price per record. More sellers can lower it.
• The Personalis business, the Pathos warrant and the SoftBank joint venture. I count the joint venture at its book value, $77.6M, and Pathos preferred stock at cost. I put no value on the warrant for 23.5M Pathos shares because I can’t price a private company.⁴

Because I can’t know these, I show a range and where the price sits in it. A fair reading of my work is that the base is somewhere from about $15 to $35, not a point.

The discount rate

I ask 10.0%, the rate I use for Palantir and IDEXX.²³²⁴ A case could be made for more, because Tempus has never made money and has $1.2B of notes to repay or refinance. I don’t add a premium, because the bear case already carries those risks and a premium would make the base look more exact than it is. At 9% my base is $28 and at 11% it is $18. At 9% my bull is $86, which only just reaches the price.

Three cases

Bear

Oncology volume growth falls to the mid-teens by 2028 as rivals match the tests. Medicare lowers what it pays per test. Data customers use their right to end contracts early, and the related-party revenue stops. Revenue grows 16% in 2027, then fades to 3%, to $3.3B in 2036. Operating margin after stock pay reaches 7% in 2033 and stays. Value $3.

Base

Revenue grows 20% in 2027 and fades to 6% by 2036, to $4.9B, a little more than IDEXX’s revenue in 2025.²⁴ Operating margin after stock pay rises from minus 13% to plus 16% in 2035. Owner cash turns positive in 2029. The step from about minus 13% in 2026 to minus 4% in 2027 is nearly nine points in one year, against a gain of 2.3 points from 2025 to 2026, so the near term in my base is not harsh E (estimate). The gap to the price is in the long run. That needs gross margin of about 66%, research at 13% of revenue instead of 21%, selling and administration at 30% instead of 45% (both before stock pay), and stock pay at 7% instead of 14%. No acquisition beyond Personalis, which I show apart. Value $22.

Bull

Data licensing becomes a large, high-margin line and oncology keeps growing in the high teens. Revenue grows 26% in 2027 and 9% in 2036, to $7.7B. Margin reaches 28% by 2036: gross margin near 72%, research 11%, selling and administration 26%, stock pay 7%. Owner cash is $1.5B in 2036. Value $69.

I checked the cases against what the company is doing. Base growth of 20% in 2027 is five points under the 25% the guide implies for the fourth quarter of 2026 C (calculated). Bull growth of 26% repeats this year’s reported pace, above the organic 23%, and then loses about two points a year, and its 28% margin is under the 32% operating margin on my IDEXX page.²⁴ Bear growth of 16% is below every quarter of 2026. What I can’t check is the long run: nothing in the filings says whether a lab with a data business settles at 7%, 16% or 28%.

The Personalis purchase, shown apart

Personalis has not closed, so I leave it out of the values above. Its holders vote on 30 November 2026, and the 10-Q expects it to close in the fourth quarter of 2026.⁴ The price is $16.25 a share in Tempus stock, an enterprise value of $1.5B, for a business that Tempus’s own projection has losing $112M of cash this year. It would make Tempus the owner of a residual-disease test it already sells under an agreement with Personalis.⁴²¹ Tempus already owns 13.0M Personalis shares, which are on the balance sheet at the deal price and which the merger cancels.

Paying in stock matters. The share count the deal issues depends on Tempus’s own price: about 19.8M shares at today’s close. By my numbers those shares are priced at 3.8 times my base value, so the deal costs existing holders little in my terms. Personalis needs to be worth about $0.65B to me for my base value per share to stay at $22 E (estimate). If instead the market price is right, Tempus is paying $1.67B of value, the shares it issues at today’s close, for a business that burns cash for years. The same arithmetic runs the other way if the price falls: at $48.42 or lower Tempus issues 34.5M shares. I can’t value Personalis. Tempus’s projection for it, which has revenue rising 52 times by 2040, is worth $4,269M at 10%, and the burn alone is worth −$447M. Personalis had $211M of net cash at 30 June 2026, which none of these figures count, about $1 a Tempus share.²¹

Other uses of the same dollar

The closest comparable on this site for how the price is set is Palantir: a data and software company priced on growth, with heavy stock pay.²³ Palantir’s price is 122% above my base and returns 6.6% a year on my base cash flows. Tempus’s is 284% above and returns 5.4%. For what it sells, the nearest is IDEXX, a diagnostics company whose revenue repeats and whose owners take out about $1.1B a year; its price is 128% above my base and returns 6.2%.²⁴ Intuitive Surgical, at 96% above, returns 7.0%.²⁵ By my numbers none of the four clears the 10% I ask. Tempus gives the least per year and has the widest gap between price and base.

Tempus sells at 10.2 times sales. IDEXX is at 12.8 on 2025 revenue and Intuitive Surgical at 13.0. That makes it look cheaper than them, until the next row. IDEXX and Intuitive generate cash for their owners. Tempus uses it up. A lower multiple of revenue is not a better price when the revenue has not yet paid for itself.

The margin of safety

I set it from the bear case: I want a price at which my $3 bear value is at least three quarters of what I pay, which is $4. The price is 21 times that. The stock has never closed below $24.96, on 25 June 2024, so no past price comes near it.¹ I am not saying the price will fall to $4. I am saying a buyer there would pay under one times sales for a lab growing at 26%, and get the data library for nothing.

Two different questions

The first is how good the business is. The testing business is real and growing: 30% in oncology volume, with a gross margin of 62%.³⁴ The data business has a scarce asset and customers that include most of the largest drug companies. The financial record is a loss that has not yet shrunk by much after stock pay, a chief executive who has sold $660M of stock and is executive chairman of the customer behind 23% of data revenue, and a habit of paying for growth with new shares and notes. I would say the business is real and the evidence for its profits is young.

The second is how good the price is. At $84.44 the price is 284% above my base value and 22% above my bull, which already assumes the data business becomes a very large, high-margin line. A good business can be a poor purchase at the wrong price. This looks like a business I can’t yet judge at a price that assumes it is excellent. The market may be seeing something I am not, and the strongest form of that is in Section VII.

VIIThe case for the price

What most people miss

Two facts cut in Tempus’s favour and get little attention. First, outside the two related parties as the accounts define them, Data and applications grew 39% in nine months and 71% in the third quarter, while related-party revenue fell from $31.2M to $22.1M C (calculated).³ The split is not clean, because Pathos builds the AstraZeneca foundation model and BMS uses Tempus data on Pathos’s platform.⁴ Second, cash from operations has nearly stopped draining: minus $73.3M in the first quarter, then plus $1.7M for the second and third together C (calculated), though part of that is timing.³¹³¹²

The bull case

I found no published model or price target on Tempus from any holder or analyst that I could open and check, so I give the case as the company makes it. In its third-quarter release of 2 November 2026 Mr Lefkofsky says the results show a durable model in which the diagnostic platform strengthens the data and AI business, citing 32% oncology volume growth and 33% growth in Data and applications.³

In plain words, Tempus owns a scarce input for medical AI: a large, linked and refreshed record of real cancer patients, and it earns the money to refresh it by running tests that doctors order. If better models make that data the bottleneck, the price per record can rise. Data revenue grows 33% at a 71% gross margin, and oncology volume grows 30% with a price step still to come. If that is right, my bull case is a fair description of the result, and the price asks for more than that.

In numbers, the market needs about $9.5B of revenue and a 28% margin after stock pay by 2036. One way to build that is Data and applications compounding at 25% a year, from $0.39B to $3.6B, and Diagnostics at 18% a year, from $1.12B to $5.9B E (estimate). Both rates are below this year’s reported pace, but both would have to hold for ten years, and with hereditary tests growing 3% the rest of Diagnostics would need about 22% a year.

Against reality

If Tempus grows revenue about 19.5% a year for ten years, to $9.5B, and reaches a 28% operating margin after stock pay, the price is possible. Growth has a record: about 31% organic in 2025 and about 23% this year, 26% as reported E (estimate). The margin has none: its best quarter, minus 9.8%, is 38 points short, and the company loses $351M of owner cash a year and issues about 2.4M shares a year to pay its people C (calculated). A 28% margin is a little below the 32% on my IDEXX page, on more than twice IDEXX’s 2025 revenue.²⁴ Multiplying it out: revenue near $9.5B, about 30%; a 28% margin after stock pay, about 30%; payers holding the price of a test, 85%; data prices holding as sellers multiply, 70%. That is 5%, or one in 19. These are my guesses, they are not independent, and I do not use them in the values E (estimate).

What a 10x would take. Ten times $84.44 is $844, or $154B of market value on today’s shares. At 30 times earnings that needs about $5.1B of net income. At a 28% operating margin taxed at 24%, net margin is 21%, so revenue would be about $24B, 15 times the 2026 guide and 31% growth a year for ten years. That is more than twice the $11.6B of revenue at Intuitive Surgical, and $154B is about what the market pays for Intuitive today, $150.9B.²⁵ IDEXX’s revenue was $4.3B in 2025.²⁴ Tempus grew revenue 47% a year from 2020 to 2025, with Ambry and a COVID-era start in the mix, and the ten years ahead would need 31% a year from a base eight times bigger. The shares have already shown how far a price can swing: $103.25 on 8 October 2025, $41.55 on 29 July 2026, $84.44 now.¹ I would not call a tenfold plausible, and I would say the same of any lab or data company this size.

Where I come down

Four things moved me: oncology growth that speeds up each quarter, data revenue outside the two related parties growing 39%, operating cash flow that has stopped getting worse, and an approved path to higher pricing that is not in revenue per test yet. They did not move my base, because stock pay doubled in nine months, Insights growth slowed each quarter, and the data line leans on a customer the chairman founded and chairs. A year of margin above minus 5% after stock pay, with stock pay under 10% of revenue and a growing non-cancellable backlog from unrelated customers, would raise my values. Until then the price sits above my bull case, and the market is right only if the data library becomes worth far more than any contract I can read today.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• Rules on lab tests. The 10-K says a future administration may try again to put FDA oversight on laboratory-developed tests, and that the FDA could treat parts of the software that reports results as a regulated device.²
• Data and privacy. The data business depends on de-identifying patient records under HIPAA and on state laws that are changing, including laws on artificial intelligence. A breach or a ruling that changed what can be licensed would hit the line that earns the highest margin.²
• Refinancing. The 2030 notes, the 2032 notes and the Google note total about $1.4B and come due between July 2030 and May 2032. Owner cash is negative today. The conversion prices help only if the share price stays up.⁴
• Cloud costs. Selling, general and administrative expense included $22.6M more of cloud storage cost in nine months than a year earlier, and Tempus owes $142.6M under non-cancellable data and cloud agreements, $74.6M of it in 2027.⁴ It took on a further $39.5M of Azure commitments with Paige.²
• Control. One holder has 57% of the votes. Outside holders can exit but cannot outvote him.⁷
• Related parties. A quarter of data revenue comes from Pathos, which the chairman co-founded, and the SoftBank joint venture, which Tempus co-owns, and Pathos can pay up to half its fees in its own stock.⁴
• Insiders. Mr Lefkofsky and other officers and directors sold $167M of stock in 2026. No officer or director has bought in the market since the start of 2025.⁸