lnofeisone

You are right. I had to refresh my memory. LMT merged its consulting with Leidos about 10 years ago and absorbed whatever was left. LMT is a product company that provides services around products, so it is the complete opposite of BAH. CACI is somewhere in the middle.

BAH is a pure play gov't consulting (though they have been trying to build their asset plays). LMT's IT competes with BAH. The rest of LMT doesn't. In many ways, CACI and LMT are similar. Both sell military hardware and offer consulting that focuses on armed forces IT. CACI's consulting is about 50%, and LMT's is about 25%.

BAH has something like 35% of their portfolio in civilian, and a lot of them are getting hit very hard (cfpb is $0 now). CACI is a bit more insulated because of its technology play but some of their service contracts are looking weak (e.g., part of USCIS that handles inbound immigration, etc.).

What's your thesis? BAH is a total mess internally. They are losing contracts and people are jumping ship. BAH is also pure gov play and don't have commercial contracts to offset (unlike ACN which lost $1.5B or so of contracts already).

Same. I'm holding steady but have dry powder to pick up extra shares if needed.

CPNG is a big position for me, and I'm aware of its silly trade-down on a good earnings pattern, but I am hoping this time is different. Ha!

Naked but with tight stops.

Shorting UNG calls.

Overall, considering they had nothing last year and now they are at the front of the pack is impressive. It's substantially better than Deepseek (33%) and a little bit ahead of the other models. However, that's all in the vacuum and theoretical. It still remains to be seen how it will perform in the wild.

CNC

Perplexity doesn't have a model. You are likely benchmarking it against something under the hood of Perplexity and you can adjust to go with GPT-4o, Claude, etc. Mistral, by design, uses small models and focuses on text. The downside here is that it doesn't keep up with GPT on complex tasks, and it's not good for coding, math, or reasoning. They are coming out with Pixtral but you can already see costs starting to go up with that. Realistically, the clients I have, always get jazzed about Mistral but when you put things in production and benchmark, Mistral tends to underperform. I do agree with you that US start ups are spending like drunken sailors. Revenue is coming in but the constant need to update and retrain really eats away any revenue that comes in.

I was part of the fund that invested in the latest (F) round.

Full disclosure - I have Andrul shares. Anduril is a great product company in the mold of SpaceX and they are great at marketing. While they purport themselves as product builders trying to build ahead of what the military wants, it's primarily exaggerations. They shape and hustle with the generals/admirals like the rest of the military complex industry. They solicit pilots to be funded by the Gov't. They differ because they try to drive to fixed-priced contracts instead of the huge time and materials that the likes of LMT like.

Sold my NEP puts and buying NEP long.

Inference is where the rubber meets the road but it's a volume game. I wouldn't be surprised if inference is going to be the new browser for big tech to bring people into their ecosystem.