Bottom line:
$TEM is the rare AI name where the revenue is real, the data is proprietary, and the margins are expanding. If MRD works, this isn't a diagnostics company. It's the operating system for cancer care.
The bear case, because you should know it:
• Personalis is a 100% stock deal, so dilution is coming
• $460M in converts on the balance sheet
• Profits are thin. One bad quarter and the "profitable AI" story cracks
• MRD is crowded. Natera and Guardant aren't going anywhere
What most people miss:
Every AI company is fighting over the same public data. Tempus owns the data everyone else wants to license. That's why pharma pays them, not the other way around.
5/ September alone
Sept 21: Recursion expanded its data license to include its RNA foundation model.
Sept 11: launched a 100,000 whole-genome dataset, target 1 million.
Sept 9: $9.5M from ARPA-H to deploy the first autonomous AI agent in cardiology.
Aug 24: another FDA clearance (ECG-PH).
Aug 4: PRISM2 foundation model published in Nature Medicine.
3/ It turned profitable
Net income $5.6M vs a $42.8M loss a year ago. Adjusted EBITDA positive. Gross margin 73.5%. $820M in cash. Full-year guide raised to ~$1.6B revenue and ~$65M EBITDA.
4/ The $20B move
Tempus is buying Personalis for ~$1.5B. Their NeXT Personal test detects cancer recurrence from a blood draw before a scan can see it. That's MRD, a $20B market with Medicare coverage already in place.
Tempus ran 9,000 MRD tests in Q2, up from 6,500. Personalis ran 10,384, up 33% QoQ. Put them together and you have the fastest-growing MRD franchise in the space.
2/ The flywheel is spinning
Q2 2026: revenue $382.5M, up 22%.
Oncology test volume up 31%.
Data & Applications up 28%. Insights up 36%.
~$200M in new data licenses signed in a single quarter.
Everyone's arguing about which AI stock is overvalued.
Meanwhile $TEM quietly became one of the only AI healthcare companies that actually makes money. Almost nobody noticed.
The story in 60 seconds:
1/ The moat
Tempus spent 10 years sequencing cancer patients and pairing their DNA with real clinical records. The result is one of the largest multimodal oncology datasets on earth. You can't scrape it. You can't buy it. You build it patient by patient, and they already did.
The last time the Fed hiked, banks got paid to sit still.
This time they got left with a bill.
XLF hit 58.60 on September 3. Tuesday it closed $54.78 on 70 million shares.
Look at the two charts:
March 16, 2022 — first hike from zero. XLF bounced that day. Then it sold for six months. June was already down double digits from the hike print. The year finished −11%. Checking paid nothing… Deposit beta lagged a year. Industry NIM jumped the most since 2002. That money showed up in 2023 and 2024. That is when XLF actually ripped.
September 16, 2026 — first hike from 3.50%. No bounce. Hike day $55.93. Tuesday $54.78. Three weeks from the high: −6.5%.
That book is gone. Time deposits 17% → 30%. 87% of CDs roll inside a year. JPM NIM already −10 bps in Q2. Huntington cut NII the same afternoon as the hike.
2022: stock sold first, NIM paid later. 2026: NIM already spent, stock is selling off now. Very different situation.
Use this with it:
Thursday they buried crypto at $76,000. Tonight it’s $86,000
First print above $85k since January. ETH off the $2,400 dump to ~$2,750. Six hundred million in shorts paid the bill in a day. Market cap back near $3T.
The Clarity Act still failed. Nobody brought it back. Atkins and Selig just kept talking. Oil came off. The 10-year slipped under 5%. Risk came back
Scoreboard since Thursday’s post:
$COIN $174 → $201 (+16%). Armstrong said agencies would move. They did. CFTC registration is still the on-ramp.
$CRCL $85 → $94. The bill that wasn’t even about stablecoins still gave you the discount. GENIUS already passed.
$HOOD $110 → $123. Chain + prediction markets + whatever leveraged-retail rule Selig actually ships.
$PURR $12.83 → $13.55 (tagged $14.76 today). Only U.S. ticker on the thing they keep saying they want onshore. HYPE token ~$93.
ethereum:native — the one that already had its classification in March and still got sold like it was on the ballot.
Nothing in the original call required a 60-vote cloture. It required the agencies to stop hiding behind Congress. That’s what printed.
The trade is not “buy crypto because the Senate blinked.” It’s the same five names, now with the de-rate partially closed.
Original thread if you’re just walking in:
$COIN — Armstrong said before the vote that failure would be fine because the agencies would move. He was right within 24 hours. CFTC exchange registration is a direct on-ramp for the biggest U.S. exchange. -8% on the vote was a gift.