No Creds Notes #24
OpenRouter and Nuclear
No Creds Notes #24
Hey guys!
Super excited about the stories covered in this week’s NCN! Read on to learn about OpenRouter and its recent raise (probably a company deserving an even more in-depth post tbh) along with an exciting regulatory breakthrough in nuclear!
If this is your first time reading Uncredentialed, welcome! Subscribe below to receive weekly updates on tech, startups, and strategy!
Who Owns the Router?
Following an incredible 5x growth in the past 6 months, OpenRouter is now routing 25 trillion tokens a week, fueling a recent $113M raise at a $1.3B valuation.
So what exactly is OpenRouter? At its core it’s an AI gateway, a single API key that gives developers access to hundreds of commercial and open-source AI models. Instead of maintaining separate integrations for each provider, you route everything through OpenRouter and it handles the rest.
The reason this matters goes back to something I argued in my data center series: the model layer is going to struggle to hold margins. If you can trivially swap between Claude and GPT and Gemini and DeepSeek, switching costs at basically zero, each model provider gets forced into a race to the bottom on pricing turning the models into a pseudo-commodity. OpenRouter actively accelerates this dynamic by making switching frictionless.
Additionally, as AI costs become material at scale, companies aren’t going to want to simply allow employees to own the decision on which models get used with how much reasoning effort at what times, they’re going to want a judgment layer that figures out which model and how much reasoning to apply to each task. Asking Codex to write a commit message doesn’t need the same model as a complex refactoring. OpenRouter enables exactly this cost-optimization routing at scale.
So where does that leave the model providers?
From my perspective, there are 2 potential paths to maintaining margins as this plays out. The first is something like government intervention; treating foundational AI models as utilities with regulated returns and protected market positions. I don’t think that happens in the near term, though I wouldn’t rule it out long term.
The second, and more probable, path is deep vertical integration. The winning move for a hyperscaler is to own the full stack: the model, the compute it runs on, the storage it reads from, and an enterprise-specific fine-tuning layer built on top of a customer’s proprietary data. The more personalized a model becomes to a specific customer’s workflows and data, the harder it is to swap out for a generic alternative, even if it’s cheaper.
Barring complete integration like I described above, OpenRouter sits in a super interesting spot positioned in many ways like the Stripe of AI. They provide an easy plug-and-play API that, similar to how Stripe grows with the GDP of the internet, could grow in parallel with the token consumption of the world.
NEPA-Free Nuclear
On May 18th, the NRC completed its environmental assessment ahead of schedule for the proposed Long Mott Generating Station in Seadrift, Texas, with a finding of “No Significant Impact”. That marks the first time since NEPA was signed into law in 1970 that a commercial nuclear project has been cleared without a full Environmental Impact Statement.
The EIS has been one of nuclear’s great bureaucratic killers. A multi-year process that invites public comment, litigation, and regulatory second-guessing, it’s contributed to the decades-long overruns that turned nuclear energy into a punchline. Dow and X-energy prevented escalated environmental studies by spending 12 months collecting groundwater data and running field surveys before they even filed their application. They pre-answered every question the NRC would have asked, helping so much along the way that the NRC beat its own 18-month target in a field known for missing the target the other way.
The reactor itself is X-energy’s Xe-100, a high-temperature gas-cooled reactor instead of the traditional light-water design, generating both electricity and 750°C industrial steam simultaneously. That steam goes directly to Dow’s UCC Seadrift operations, powering production of more than 4 billion pounds of materials per year.
Interestingly, this is an entirely different use case than people tend to think of when they picture the revival of nuclear. Instead of powering data centers or the grid, it will help by reducing the huge fraction of industrial carbon emissions coming from process heat: the high-temperature thermal energy that chemical plants, steel mills, and semiconductor fabs need to run. Long Mott, if it gets built, will become the first grid-scale advanced nuclear reactor deployed to serve an industrial site in North America.
Now, it’s only fair to call out that the last time a startup reactor reached unit 1, it was NuScale, a story that began as a ZIRP-era $3.6B project, grew and grew to $9.3B, before getting cancelled when rising interest rates crushed the utility buyer’s economics. Long Mott should be different: For one, Dow’s heat demand should be more sustained regardless of interest rate environment, and, for two, we aren’t in a ZIRP environment anymore, so interest rate risk is inherently lower.
Hopefully, Long Mott’s regulatory journey continues to go well and even serves as a model for how future nuclear projects can simultaneously lean into safety while expediting permitting and construction.
If you enjoyed this edition of No Creds Notes, consider sharing it with a friend, and if you haven’t yet, subscribe to Uncredentialed!