Frontier AI model compute costs remain very high, limiting profitability for AI companies despite high GPU vendor margins.

Elon Musk's proposed space-based data centers would require approximately $100 per B200 GPU-hour to make financial sense, compared with $2-3 spot pricing on Earth.

“Elon is focused on moving the chips to moving the data centers to space and I think on the numbers that he has I think they are looking at like a $100 an hour per GPU hour for a B200 for it to make sense.”

The open-source model GLM2 is 90% cheaper than Claude Opus.

“It is 90% cheaper to use these products.”

Frontier labs spend 10 to 20 times more on compute than on data.

“My sense is that the split is something like 20 to 1 or 10 to 1.”

Decagon runs 90% of its workflow on open-source models, not frontier models.

“Today 90% of our workflow is on open source.”

Startups are migrating off frontier AI APIs to locally-hosted open-source models, using last-generation models at significantly cheaper rates because most use cases don't require the latest capabilities.

“I work with startups. They are all moving off of these and they're using open source and they're using it at much cheaper rates because a lot of the jobs don't need the latest models. They can use these the last generations models and people are moving them local. They're hosting them themselves.”

AI model performance advantages evaporate within weeks, as open and closed competitors replicate or exceed frontier capabilities almost immediately after benchmark publication.

“These models are getting commoditized much faster than anybody thought. And how do we know this? Because there is no meaningful sustained advantage once a model publishes their performance criteria. What you see is literally within weeks other models some open some closed some open weight who are able to match and in some cases exceed the performance.”

Within the next year, teams will be able to dial in compute per task, cutting token costs by over 90%.

“I think you'll get to a place probably over the next year where you can like really dial in hey how much compute do I want to spend on this because I have certain like cost considerations and certain latency considerations and get to like the exact optimal amount of cost. and so if you do that like your token costs go down 90% plus.”

Grok 4.6 already surpasses Fable 5 in quality at a lower price, breaking the perceived OpenAI-Anthropic frontier duopoly.

“You can see like even on this, you're well ahead of Fable 5, which you know, I think most people would kind of agree is the you know, the gold standard. Like this is you're you're you're higher quality at a slightly lower price.”

Compute can be converted into proprietary training data, meaning a company without partner data is not necessarily blocked from building specialized models.

“There are ways to turn compute into data and get more and we're we're doing those, right?”

Grok 4.5 uses just one-third the tokens of GPT-5.5 or Fable while achieving a similar intelligence score.

“Grok 4.5 uses just one-third the amount of tokens as GPT-5.5 or Fable while achieving a similar score.”

Training a model close to frontier capability is not technically hard today, and many actors are achieving this not just via distillation.

“It's not hard to train a model that is close to frontier capability.”

Replicating Nvidia's Neatron Nano pre-training run costs approximately $100,000, making it affordable for nonprofits to experiment with pre-training interventions.

“We're looking at scaling up pre-training filtering and going to be doing not full but pretty close to replic full replicas of something like Nvidia's Neatron Nano and it only costs maybe like $100,000 per run.”