A day after AMD's record data-center quarter, its 10-Q put the first real price on the Taalas acquisition it announced in August: about…
A day after AMD's record data-center quarter, its 10-Q put the first real price on the Taalas acquisition it announced in August: about $2.9 billion, with up to roughly $1.4 billion of that as contingent consideration. The deal still needs regulatory approval and other conditions, and AMD expects it to close by year-end.
Taalas is a three-year-old Toronto startup with an unusual approach. Instead of general-purpose AI chips, it designs silicon built for one specific model, hardwiring the model into the chip to make inference faster and more energy efficient. Its first chip was built for Meta's Llama family. AMD says it will fold the technology into future inference chips.
The price is what caught my eye. Taalas's Series B in November 2025 implied a valuation above $750 million, so this is a steep markup in about a year. It's also bigger than Nasdaq's $2.75 billion Verafin deal in 2020, and the largest known Canadian tech deal since Ecolab bought CoolIT for roughly $4.75 billion in March.
My read: AMD is paying up for the idea that inference eventually fragments into per-model silicon, and the large contingent piece shows it wants proof before paying in full. That seems like a sensible way to place the bet. If hardwired chips really do win on cost per token, I'd expect other chipmakers to start looking for their own version of Taalas.
$AMD
Sources:
The Globe and Mail: https://www.theglobeandmail.com/business/article-amd-to-pay-29-billion-for-ai-chip-firm-taalas/
Benzinga: https://www.benzinga.com/news/26/11/62856168/amd-paid-2-9b-acquire-specialized-inference-silicon-technology-provider-taalas-including-1-4b-contin
Not financial advice. For informational and educational purposes only.
Snap Q3: drops daily users as its headline metric even though DAU/MAU held at ~50.7%; North America lost 6M daily users since mid-2025
92 million. That's Snap's North American daily users, down from 98 million a year ago, and it's the number $SNAP will stop reporting next…
29% vs 14%. That gap between Duolingo's daily user growth and its paid subscriber growth is the number I keep coming back to after the…
29% vs 14%. That gap between Duolingo's daily user growth and its paid subscriber growth is the number I keep coming back to after the $DUOL rally. Users are piling in faster than they're paying, and the next few quarters decide whether that's a funnel or a…
The FIS reset is priced in fast.
The FIS reset is priced in fast. What I'm watching now is who becomes permanent CFO, and whether the bank-side headwinds behind this cut stay contained to 2026. On November 4, $FIS cut FY2026 adjusted EPS guidance to $5.82–$5.89 from $6.15–$6.24, and revenue guidance to $13.445–$13.505 billion from…
Volcker Shock & the 1980s Bull — 20% Rates to 3.75%-4.00%
In June 1981, the Federal Reserve pushed the Fed Funds rate to 20% to break inflation running at 14.8%, the most aggressive tightening cycle in Fed…
In June 1981, the Federal Reserve pushed the Fed Funds rate to 20% to break inflation running at 14.8%, the most aggressive tightening cycle in Fed history. The market absorbed the full cost of it. The S&P 500 bottomed at 102.42 on August…