SambaNova has pulled in another billion dollars, and the detail that matters sits underneath the headline number. On 8 July the AI chip company closed the first tranche of a Series F round, $1 billion at an $11 billion post-money valuation, led by General Atlantic with T. Rowe Price, Capital Group, BlackRock, Intel Capital and Qatar's sovereign fund all taking part. Chief executive Rodrigo Liang said a second close should follow in the coming weeks.

The raise lands only five months after SambaNova's last mega-round, a pace that says as much about investor appetite as about the company. What sets this one apart is the customer it named alongside the cash. JPMorgan Chase has picked SambaNova as an inference-infrastructure partner, putting the company's SN40L and SN50 systems to work running AI models on the bank's own premises rather than in a public cloud.

The fight is moving from training to inference

Most of the money and attention in AI hardware has gone to training, the enormous one-time job of building a model. Inference is the other half, the cost of actually running that model every time someone asks it a question. As companies move from experiments to daily use, inference is where the bills pile up, and it is the market SambaNova has aimed at. Its pitch to a bank like JPMorgan is speed and control, with sensitive data staying inside the building.

That puts SambaNova in a crowded and expensive lane. Nvidia still dominates, though it has troubles of its own after slipping its next rack system to 2028, while memory suppliers like SK Hynix raise record sums to keep pace with demand. The through-line across all of it is the same. The capital is chasing whoever can run these models cheaply and reliably, and right now investors are willing to write very large cheques to find out who that will be.

Sources

  1. i. techcrunch.com
  2. ii. www.generalatlantic.com
  3. iii. www.hpcwire.com

Commentarii · 0

Add · a · Comment