Interest, GLP-1s, AI

Interest Expense

I’m not sure this story is going to get the coverage it deserves, especially as the election news begins to dominate news cycles over the next 18+ months. The US government’s interest expense has reached a $1 trillion annual run rate. This is the interest paid on Treasury securities. You can keep track of the reported figure here. For 2023, the Congressional Budget Office (CBO) forecasts that US government revenue will be $4.8 trillion vs. outlays of $6.9 trillion. Interest expense represents 20% of revenues and is on track to be the largest single line item, along with a sizable deficit. At $2 trillion, interest expense would represent 40% of tax receipts, funded by taxpayer dollars and not exactly something that produces a return on investment.

In 2021, the CBO forecast interest expense to reach $900 billion by 2031. The continuous justification for growing US debt from $5 trillion to $25 trillion over the past 25 years was that the interest expense would be reasonable with low rates. We’re way ahead of schedule on all accounts, and the math today simply does not work. This is something to watch, especially as the new cycle focuses on other stories.

GLP-1s / Ozempic

NYTimes reporter Gina Kolata had a detailed report on Ozempic last week with the opening line, “Every so often a drug comes along that has the potential to change the world.” I recommend reading her report for the details. According to conversations that I’ve had recently with reputable and knowledgeable parties, GLP-1s appear to be on par with other major medical milestones, like the discovery of penicillin, for example.

Obesity is linked to a litany of health issues. Covid deaths are higher for obese people, as are heart attacks, liver disease, cancer and depression. Genetic lottery clearly plays a role, as many people are more prone to obesity regardless of intent, diet, exercise, discipline or willpower. GLP-1s appear to be the first real effective treatment for obesity. Treatment for uncontrollable desire and the role of the dopamine cycle is also on the table. It’s early days, but it’s safe to say that given the coverage, comments and people involved, this qualifies as a true leap forward.

AI Bites

• NVDA crushed earnings again with its AI data center service bringing in an additional $6B of new revenue this quarter that is almost entirely profit. Next year’s guidance includes an estimated 100% growth of top-line revenue. The AI data center service is the singular growth area for obvious reasons.
• There’s a new AI model trained to recognize keystrokes via audio recording. It was specifically trained on MacBook keyboards, which are ubiquitous. If I were Apple, I would immediately start developing a silent keyboard. We also have to accelerate the rollout of passwordless and biometric-only security systems. 
• McKinsey trained an AI model on their internal data to share insights internally without needing to talk directly with any particular subject matter expert. 
• A few of the largest AI companies now spend $1 billion each to train large language models.