Palo Alto Networks will acquire AI startup Applied Compute for $4.5bn in stock, the security company said on 18 November. Applied Compute, which says it helps businesses post-train and run open-weight AI models on their own data, is being bought at about 38% above the valuation from its funding round earlier this month.
Key facts
- Palo Alto Networks is buying Applied Compute for $4.5bn in stock, announced on 18 November 2026, with closing planned for its fiscal third quarter of 2027 (February to April 2027).
- Applied Compute post-trains and runs open-weight AI models for companies and reports more than $100m in annualised recurring revenue.
- On 3 November Applied Compute was valued at $3.25bn, according to SiliconANGLE; the purchase price is about 38% higher.
Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →
The deal is all-stock and subject to adjustments, according to the press release, which Palo Alto Networks also filed with the US Securities and Exchange Commission. The company expects to close in its fiscal third quarter of 2027, which runs from February to April 2027, pending regulatory clearance. Applied Compute chief executive Yash Patil is set to become co-chief AI officer at Palo Alto Networks, and the research team will join too, the release said.
What Applied Compute does
Applied Compute says it runs a platform for post-training and inference of open-weight models, meaning models whose weights are freely available; Llama, Qwen and Kimi belong to this class. Companies can use it to train, serve and keep improving their own models, in public clouds or sovereign environments. According to SiliconANGLE, the product includes an autoscaling inference engine and a model router called Shift-1 that sends each request to the cheapest suitable model.
The company was founded in 2025 by former OpenAI researchers Yash Patil, Rhythm Garg and Linden Li, SiliconANGLE reported. On 3 November it raised $350m at a $3.25bn valuation in a round led by Elad Gil, with Nvidia, Shopify Ventures and HubSpot Ventures taking part, the outlet said. Applied Compute puts its annualised recurring revenue (ARR) at more than $100m, and SiliconANGLE lists Microsoft and DoorDash as customers. That puts the price at no more than 45 times ARR.
Palo Alto Networks bets on mixed models
“Every enterprise is going to run a mix of open and closed models,” chief executive Nikesh Arora said in the release, adding that every one of those models needs to be secured. The company argues that no large provider currently delivers open-weight models securely. Palo Alto Networks announced the deal alongside quarterly results: revenue rose 39% to $3.44bn, SiliconANGLE reported, and the shares gained almost 10% in regular trading.
Open-weight models were already in the news in the days before: Hugging Face handed routing for open AI models to OpenRouter, and Reuters reported that Washington is weighing a disclosure rule for Chinese AI models such as Qwen and Kimi.
Analysis
Open-weight models have been seen as a way out of dependence on OpenAI, Google or Anthropic: download the weights, adapt them with company data, run them on company infrastructure. If a firewall vendor sells post-training, hosting and routing as one package, a new dependency emerges elsewhere. Anyone running open models through such a platform hands control over data, training runs and model choice to the platform operator. Companies in Germany, Austria and Switzerland should therefore read those contract terms as closely as those of a large model provider.
A post-trained model is also only as good as the data it learns from. Marketing and sales teams that want to adapt a model with brand language, product knowledge or CRM data need clean data, clear sign-offs and a named owner first; such projects tend to stall there rather than on the platform. The guide AI agents for marketing sets out where a first project can sensibly start. A price of up to 45 times revenue also indicates how much value the market now places on the layer between model and application.
Got AI news? Submit news →
FAQ
What does Applied Compute do?
Applied Compute says it runs a platform that lets companies post-train, serve and improve open-weight AI models on their own data, in public clouds or sovereign environments. According to SiliconANGLE it includes the Shift-1 model router.
How much is Palo Alto Networks paying for Applied Compute?
$4.5bn, entirely in stock and subject to adjustments. Closing is expected in Palo Alto Networks' fiscal third quarter of 2027, pending regulatory clearance.
Who founded Applied Compute?
Former OpenAI researchers Yash Patil, Rhythm Garg and Linden Li founded the company in 2025. Patil is set to become co-chief AI officer at Palo Alto Networks after closing.
Source
More sources
- SiliconANGLE – Gründer, Finanzierung, Quartalszahlen, Kursreaktion
- SiliconANGLE – Finanzierungsrunde 3.11.2026: 350 Mio. $ bei 3,25 Mrd. $, Produkt, Kunden




