Kelsus pivoting because of AI
Frontier model lock-in puts enterprises at risk and creates an opportunity for builders
Enterprises have spent 2026 test driving AI
Within weeks of my first use of ChatGPT, I predicted that software engineering would become cheaper, speed up, and my company Kelsus would end up needing more clients to keep the same number of developers busy. This year, when this actually started happening, even though I knew it was coming, it was a shock. But at the same time, there was another movement brewing, and I was paying attention.
Just in 2026, enterprise companies fully started to embrace Anthropic’s Claude. The proof is in the numbers: Anthropic’s revenue jumped from $9BN at the end of 2025 to $47BN at the end of May. Those dollars came from enterprises newly enthusiastic about Claude’s capabilities.
These enterprises are still new to the AI world, and while they might be scrambling today because of token costs, and the comparative lack of new value those tokens have created, they haven’t seen the next part of the lock-in rachet the the frontier model providers are building.
Enterprises won’t outsource their entire intelligence
Simply put, if an enterprise uses Claude to achieve something, and it works well, then a little bit of that enterprise’s intelligence has moved from the brains of their employees to the servers operated by Anthropic. And it works well, so they add another piece. Maybe they start with customer service, then they do invoice ingestion, then something marketing related, and finally they’re even driving mid level management decisions with Claude. Before they know it, they look around and a significant part of their business operations and decision making are happening at another company.
Even if Anthropic hadn’t recently done sudden price changes, secretly rerouted requests from one model to another, and rug-pulled a whole model class, enterprise CTOs and CISOs would still frown at the idea of outsourcing running the brain of their enterprise to another company they can’t control or audit.
So, to avoid outsourcing every agentic action to companies whose business models favor token-hungry workloads, the enterprise world will be making the switch to self-hosted, sovereign AI.
This is where Kelsus comes in. Working with self-hosted AI is still something that requires systems engineering experience. We have a whole team of really intelligent systems engineers that are thrilled by the idea of sharing hard-won knowledge with new clients and helping them own their own AI and send their data only to servers they control.
We’ve updated our website at https://kelsus.com with our new productized service offerings, an open-source kubernetes-orchestrated reference architecture, and benchmark report of typical financial services workloads across a broad range of open weight models. My favorite finding of the benchmarks is that the models perform the same regardless of the substrate they run in. What I mean by this is that regardless of whether a model like Kimi K2 is run via AWS Bedrock or self-hosted, it scores the same given the same workloads. In other words, AWS doesn’t have some secret configuration they use to juice their models in Bedrock.
Lastly, we’re focusing our pivot on financial services companies. In the past few years we’ve built an entire accounting package, a cash flow prediction system, a personal injury financing system, and an investment banking data room and task management system. We’re well positioned in the financial services sector to deploy AI to greatest effect.
My ask of you
Do you work at a heavily regulated financial services company? Are you considering switching to sovereign AI? Please get in touch. If not, please introduce me to someone who does.
The next few years are going to have an incredible amount of activity in the sovereign AI building space, and Kelsus has a head start.
Thanks for reading!
—Jon Christensen