SM Research Lab

What if?

$IAM picked up a lot of new eyes yesterday with four banks starting coverage. Here’s what you get for ~$665M, with ~$410M of that in cash. The AlphaFold comparison. NeuralPLexer was built with NVIDIA researchers at Caltech. NVIDIA co-owns a patent family and put $25M into the company ~15x faster than AlphaFold 3 by Iambic’s numbers. An independent benchmark clocked the first version at ~100x on the same chip Isomorphic, Alphabet’s AlphaFold drug company, is reportedly raising at ~$50B (Bloomberg) with no drug in humans Iambic, with a drug in 122 patients, is ~$665M. That’s about 1/75th of Isomorphic’s price, and the business beyond cash (~$255M) is under 1/190th, for a model that’s arguably on par on accuracy and ~15x faster by Iambic’s numbers. The drug IAM1363, a HER2 pill, 122 patients dosed. Iambic says it’s the only one known to bind the inactive form of HER2 It concentrates in tumours: 20-30x more drug in the tumour than in the blood, in biopsies from two patients 3 confirmed brain responses in breast cancer patients who’d already failed Enhertu and Tukysa. Nothing is approved for them Registrational trial as early as 2027, and two more drugs entering the clinic The flywheel Molecules designed Monday, made Tuesday, tested by Friday. A traditional pharma design cycle takes weeks to months. Iambic puts ~1,000 molecules through one every 7-10 days, and the CEO puts its pace at ~50x a typical pharma programme Every result goes straight back into the models, so each week’s designs start smarter than the last. More data makes better models, which make better molecules, which make more data That’s how IAM1363 went from programme start to the clinic in under 2 years, with the drug candidate found by month 7. The industry norm is ~6 years Enchant v3: 41B parameters, 4.5T tokens of training data, 6,000+ properties predicted, ~12M predictions a month. Google’s TxGemma tops out at 27B Against TxGemma, by Iambic’s numbers: kinase activity 0.69 vs 0.55, rat clearance 0.70 vs 0.35 With ~5% of the human training data, Enchant matched the best published human PK models trained on 100% The coverage All four of the banks that took Iambic public started coverage on Monday, all at Buy or Overweight Jefferies $35, BofA $30, JPMorgan $22, Citi $22. The average is ~$27, about 95% above $14 BofA’s case: Iambic uses AI to design new drugs, not just to find targets, and it trades below its listed AI peers The stock closed up 31.6% on the day Jefferies has the CEO on its AI drug discovery panel on 18 Nov Commercial validation AbbVie signed up the day Enchant v3 launched: $40M upfront, up to $670M in milestones Bayer’s deal names Enchant and NeuralPLexer. RevMed paid ~$25M for NeuralPLexer alone. Takeda signed a separate technology agreement on top of its drug deal Upfronts keep climbing: $10M (Lundbeck) → $27M (Takeda) → $40M (AbbVie). ~$2.4B in potential milestones across the book Isomorphic’s Lilly and Novartis deals brought $82.5M upfront. Iambic’s bring ~$77M. A similar deal book, at roughly 1/75th of the valuation Last December the CEO told investors more deals were coming. Three signed within nine months. The balance sheet ~$410M pro forma cash, plus AbbVie’s $40M upfront still to come. No debt Funded into late 2028: enough to start the registrational programme for IAM1363 and take two more drugs into the clinic The company is locked up too. No new shares before April without its banks’ sign-off Partner money stretches it further. The CEO says every new deal adds quarters of runway Therefore minimal dillution risk. Who’s buying IPO 5x oversubscribed (Bloomberg) Cathie Wood’s ARK has bought on 8 of 15 days since the IPO, ~2.4M shares across its funds (~5%) Druckenmiller’s Duquesne indicated interest in the IPO A director’s fund bought $11.5M in the IPO. No insider has sold The flow 80% of shares locked until April. The 10 biggest IPO buyers took 80%+ of the rest. Possible Russell add: preliminary list Friday, index buying 11 Dec CEO on Jefferies’ AI drug discovery panel, 18 Nov Of course NFA, Dyor

Iambic took IAM1363 from programme start to human trials in under 2 years. The usual timeline, by the CEO’s numbers, is about 6 years. $IAM The lead series came within a few months and the clinical molecule by month 7. Tom Miller, on Catalio Conversations, puts their pace of AI-designed experiments at roughly 50x a typical pharma programme. A drug’s patent clock starts early in discovery, so every year saved before the clinic is an extra year of protected sales at the end, when sales are at their peak. His answer on why big pharma can’t just do the same: train a model on every dataset that exists and you hit a wall, because the data runs out. Iambic generates new data for each programme on demand through its weekly lab loop. Anyone can buy compute. You can’t download data that doesn’t exist yet. On NeuralPLexer vs AlphaFold: AlphaFold gives you a protein’s shape. NeuralPLexer predicts where a drug binds, what it does to the protein and how strongly. He called it step one of the 40-50 properties a drug has to get right, and Enchant covers the rest.

$IAM AI hype? Look at the drug it made. Most HER2 breast cancer pills also hit a near-identical protein in skin and gut. That causes the side effects that force dose cuts. Iambic's AI found a pocket on the cancer protein that its twin lacks. IAM1363 is the only known drug of its kind that binds there. It hits HER2 2,800x harder than the twin in lab tests. Closest rival: ~1,200x. It also reaches the brain, where up to half of these patients see their cancer spread, and beat the two leading drugs there in animal studies. It's in 122 patients now. For scale: Seagen paid $614M in 2018 for tucatinib, a brain-penetrant HER2 pill in a pivotal trial. IAM1363 is earlier but more selective. It's one of three $IAM drugs, before you count the NVIDIA-backed AI or the Enchant model AbbVie and Bayer pay to use. At ~$10: ~$480M market cap, ~$449M cash ($40M from AbbVie). Everything else is priced at ~$71M.

This is big for the $IAM thesis. Isomorphic, a competitor Iambic names in its own prospectus, is reportedly raising at $50B. No drug in human trials yet. Iambic's NeuralPLexer is ~15x faster than AlphaFold 3 with similar accuracy, has a drug in 122 patients, and trades at ~$640M. 1/78th of the price.

$IAM bull case: • Its AI matches Google’s AlphaFold 3 on accuracy and runs ~12-15x faster (Iambic’s numbers) 
• Alphabet’s Isomorphic: reportedly raising at $40B +, no drug in trials. IAM: ~$600M (1/67th of the price), ~$190M after cash 
• Enchant v3: a 41B-parameter model trained on 4.5T tokens, fed by data from Iambic’s own automated lab. Frontier-scale for drug discovery 
• Lead cancer drug in 122 patients, tumours shrinking in Phase 1. Trial expanding to ~380 patients across the US, Europe and Asia 
• $NVDA -backed: investor, and co-wrote the first version of the model
• Cathie Wood’s ARK was in before the IPO and bought more at it. Druckenmiller indicated interest too. 
• Backers include Sequoia, Coatue and Qatar’s sovereign fund, some of the same investors behind Anthropic 
• Insiders bought substantial amounts at the $16 IPO price, above where it trades today 
• Cash runway into late 2028 (company guidance) 
• Pharma deals with AbbVie, Takeda, Lundbeck and Bayer Ai / Bio. NFA, DYOR

Started @SmResearchLab This account will stay for the shitposts + ideas, while the new account will be purely for investment theses and longer-form research. I want it to serve as a portfolio of my work and a personal scorecard, somewhere I can learn as I go, continuously improve, track my thinking over time, and look back years from now to see how I’ve developed as an investor. Appreciate everyone who follows along. First deep dive coming soon. $GAP