What The Banking Desks Are Watching

The bank desks are circling the same three pressure points: oil, AI credit, and the dollar.

Across today's banking desk notes, the cleanest macro read is that oil and Hormuz are back in the rates conversation. JPMorgan's rates desk noted that renewed Middle East stress pushed Brent from roughly the low $70s to above $78 and warned that oil may be re-emerging as a principal driver of yields, swap spreads, curves, and volatility. MUFG, Mizuho, ANZ, and Rabobank were all working around the same problem: energy shocks do not stay in energy when central banks are already jumpy about inflation.

The second theme is AI capex turning into a credit story. TS Lombard argued that new capex, information processing equipment, software, and R&D, has not behaved like old capex in relation to the yield curve, but the equity market is starting to question whether that exception survives flatter curves and tighter policy. Separately, market notes around Amazon's latest bond financing and hyperscaler spending point to the same thing we have been talking about: AI infrastructure is becoming project finance. GPUs and data centers are getting levered up.

The third theme is the dollar. Deutsche Bank argues that foreign official appetite for US debt is weakening while foreign private appetite for US equities, especially the AI trade, is rising. That means the dollar may become more tied to US equity exceptionalism and the AI cycle than old-school Treasury recycling. That is a subtle but important shift. The reserve currency is still dominant, but the funding mix is changing.

Prime Intellect Raises $130M To Build The Open Superintelligence Stack

Why it matters: The AI sovereignty story is moving from slogans to funded infrastructure.

Prime Intellect announced a $130 million Series A to build what it calls the Open Superintelligence Stack, led by Radical Ventures with NVIDIA, Intel Capital, Dell Capital, and existing investors. The company's pitch is straightforward: train, deploy, and continuously improve your own models on an integrated compute, training, inference, and sandbox stack. Their homepage repeats the line that matters most: Own Your Intelligence.

This is the right direction. The world does not need every company, developer, and operator renting intelligence from a handful of closed frontier labs forever. We need people training workflow-specific models, running evals, owning inference, and improving agents around the actual work they do. That does not make the capital intensity of AI disappear. It does point to the more durable side of the trade: operators using AI to build leverage they control.

Tap-To-Pay Ecash Shows What Private Digital Payments Can Feel Like

Why it matters: The best answer to surveillance finance is payments that are fast, local, and cash-like.

Calle showed off tap-to-pay bitcoin ecash using Cashu, describing it as free and open source, faster than credit cards, cash-like in its privacy properties, and able to work even when the sender does not have internet. His explanation is simple: ecash can be stored on your phone like physical cash sits in your wallet, then moved from one device to another over NFC.

This is the constructive side of today's lead. Governments are building reporting machines. Builders are making private payment tools feel normal. Cashu is not a replacement for self-custody cold storage. It is a privacy-preserving spending tool that can make small payments feel like cash again. That is exactly the sort of thing we need more of.

Spencer Pratt Says The Fraud Is Upstream Of The Count

Why it matters: If the public does not trust the ballot pipeline before votes are counted, recounting anonymous ballots after the fact will not restore confidence.

Spencer Pratt posted a long video after his California race arguing that the real issue is not a sloppy count at the ballot processing center, but everything that happens before ballots get there. His core claim is that late-arriving mail ballots erased a roughly 40,000-vote election-night lead, that ballot harvesting around vulnerable populations created openings for abuse, and that California has removed too many mechanisms for detecting fraud before ballots become anonymous.

The important distinction in the video is evidence versus proof. Pratt says an anomalous late vote swing, loose voter rolls, permissive signature rules, ballot harvesting, and reports of people being paid for ballots are evidence that should trigger serious investigation. He also says a recount is the wrong remedy if the alleged fraud occurs before ballots are separated from envelopes. You can recount anonymous ballots forever and still miss the upstream problem.

I am not going to adjudicate every claim in a newsletter paragraph. Watch the video. The broader point is obvious: elections require more than accurate counting machines. They require a chain of custody that normal people can understand and trust. If officials design a system where fraud is hard to detect, then scream that no one has proved fraud, they should not be surprised when citizens stop believing them.

Token Prices Are Falling While AI Infrastructure Gets Levered Up

Why it matters: The AI story is splitting between cheaper model usage and more expensive infrastructure financing.

One of the more interesting market notes floating around today was the token-unmaxxing argument: AI model companies are discounting aggressively, startups are getting large credits, and cheaper open or Chinese models are putting pressure on frontier pricing. That is happening at the same time hyperscalers and neoclouds are financing massive data center and GPU builds.

Both things can be true. The workflow layer can get cheaper and more useful while the infrastructure layer gets more levered and more fragile. That is why I keep coming back to the credit structure. AI is real. The productivity gains are real. The question is whether the debt stack being built underneath it is as durable as the demos.