The S&P 500 is currently in its 11th-worst uninterrupted decline since 1940. Over the past four trading sessions, the index has dropped 12.1%. Here are the other instances of large, continuous drawdowns and what happened next

The conspiracy theory that Trump is crashing the economy to bring down interest rates to refinance the national debt is complete nonsense! The U.S. is currently paying an average interest rate of about 3.3% on its $36 trillion national debt. That’s already lower than where market rates are today (the 10-year Treasury is yielding around 4% and the 30-year is closer to 4.4%). So for this plan to even begin to make sense, interest rates would need to plunge well below current levels, down to 2% or 3%. And even if that happened, it wouldn’t be some game-changing win. Yes, over time, some of the debt matures and gets rolled over. Refinancing that portion at lower rates (3% instead of 5%) could help around the edges. But you can’t just call up America’s creditors and refinance the entire $36 trillion like it’s a mortgage. That’s not how any of this works. Most of the national debt is in short-term Treasury bills and notes that are constantly rolling over. Even if rates fall, you’d still be refinancing gradually, over time, as debt matures, not in one big bang. And you definitely can’t shift all that short-term debt into 30-year bonds. The demand just isn’t there. Long-term Treasuries (those maturing in 20 to 30 years) make up only about 17% of the total market. Try cramming trillions into that, and you’d blow up the bond market. Right now, Treasury bills (which mature in less than a year) make up about 22% of U.S. debt. Treasury notes (2–10 year maturities) make up around 51%. Each part of the bond market has its own supply and demand dynamics. Bills are basically cash-like instruments; they don’t have interest rate risk and their prices barely move. That’s a big reason there’s always strong demand for them. You can’t simply shift a huge chunk of short-term debt into long-term bonds. Those carry a lot of interest rate risk, and the pool of buyers is different (mostly pension funds and insurance companies). If the Treasury tried to flood the market with long-term bonds, demand might not keep up, which would actually push long-term rates even higher. We’re already seeing some of that. Despite rising recession fears, 30-year bond yields haven’t dropped much, suggesting limited demand at those durations. Now, at the margins, the Treasury can and does tweak the debt mix…maybe shifting a little from bills to notes, or notes to bonds. But it’s a balancing act, not something you can radically change overnight.

DeepSeek made some clever, maybe even groundbreaking, improvements to large language model architecture. The cost of LLMs will likely fall because of that. But that was already the trend: model costs have fallen significantly; this just accelerates that. And if that’s the case, then it’s actually a positive for AI adoption. Why then is there so much panic in the air? One reason is the geopolitical angle—the “China has caught up to the U.S. in AI despite restrictions on U.S. chip exports” narrative. This remains to be seen. Some might recall that there was a similar level of hand-wringing when SMIC (China’s biggest chip fab) manufactured 7 nm chips for Huawei a couple of years ago. “How did they do that despite U.S. export controls? China has caught up!” a lot of people thought. But upon further inspection, it turns out that the process behind those chips was far less advanced and far more expensive than initially thought. China did not, in fact, catch up to TSMC. Now, obviously, chip manufacturing and LLMs are different. It’s harder to replicate the entire chip manufacturing supply chain than to develop AI models (or so I’m told). But it’s still true that people often panic about things before the full picture has emerged. The second angle to all this is the impact on the market for AI chips. “If DeepSeek can create a model on par with OpenAI’s o1 at one-tenth the cost, why bother buying Nvidia’s Blackwell chips?” Hence, the 16% slide in shares of $NVDA. There are a few ways to look at this. First, there are reports that DeepSeek used a lot more resources than the $5.5 million training number suggests. AI researcher Nathan Lambert says that “tracking the compute used for a project just off the final pretraining run is a very unhelpful [and deceptive] way to estimate actual cost,” while adding that “the CapEx on [DeepSeek’s] GPUs themselves, at least for H100s, is probably over $1B (based on a market price of $30K for a single H100).” Lambert estimates that the cost of a year of operations for DeepSeek AI is probably “closer to $500M (or even $1B+) [rather than] the $5.5M numbers tossed around.” Still, he acknowledges that that’s a success considering they are “relevant in an industry where American tech giants are spending what is approaching or surpassing $10B per year on AI models.” OK, so the $5.5 million number is misleading, but DeepSeek is still pretty dang efficient. That point, I think, you have to concede. But why would DeepSeek efficiently using the hardware available to them be a negative for the chip industry (let alone the Mag-7 outside of Nvidia)? “You can now use fewer Nvidia GPUs to accomplish the same thing.” You could have said the same thing about any generation of Nvidia chips. Hopper was more efficient than Ampere. You could accomplish the same thing with fewer Hopper GPUs than Ampere GPUs. But that didn’t stop companies from buying many more H100 chips than A100 chips, leading to much more powerful AI models. The same will likely be true of the Hopper to Blackwell evolution. That’s why I don’t think the DeepSeek developments are negative for Nvidia. I think they could certainly dent sentiment on the stock and feed the narrative that was already festering even before DeepSeek arrived on the scene— that AI capex was out of control, hyperscalers will cut back, etc. (still no evidence of that but the worries aren’t going away). For Nvidia, I think the more concerning narrative isn’t DeepSeek making AI more affordable. It’s the slowing of traditional scaling laws (more data and more powerful computers equals better models). This could be a data issue (we’re running out of data to feed the models) that will be solved with synthetic data and other techniques. But while improvements from pre-scaling slow down, other dimensions of scaling seem to be just getting started, like inference scaling (time test compute). That could be an issue for Nvidia to the extent their dominance isn’t as firm in inference as it is in training. This is the whole “Nvidia’s customers are building custom chips with Broadcom/Marvell and that will hurt Nvidia narrative.” I think that is a legitimate concern, especially with Nvidia sitting on sky-high profit margins. So, that’s the bigger issue for me with the stock, not DeepSeek. If anything, DeepSeek is good for the AI chip industry overall. Demand for AI will explode, leading to more demand for chips—whether to be GPUs or custom AI chips. And for the rest of the Mag-7? Great news. Cheaper AI means they can deliver better products without killing their margins.

The panic surrounding DeepSeek kind of reminds me of what happened a couple years ago when China’s largest chip maker, the Semiconductor Manufacturing International Corporation (SMIC), was found to have manufactured 7 nm chips for Huawei. That discovery shocked the world and triggered a wave of concern about China’s technological capabilities. Is it a similar story in this case? Are people overreacting about DeepSeek the same way they did with SMIC before the full picture emerged?

Even if you take Deepseek's accomplishments at face value, lower costs for running AI models are bullish for most of the Mag-7. It might be negative for $NVDA, but it's positive for $META and $MSFT. They might be overspending now, but they can stop at anytime and their margins will go up

Crypto is a backend technology. Tokenizing existing stocks and bonds won’t visibly change the user experience for those assets if the apps we use to interact with them don’t change. But they will change. Even in the traditional financial system, innovation is happening in the app layer. The Robinhood experience of today is significantly different than the Schwab experience of five or ten years ago. Fractional share trading and 24-hour trading aren’t things that were available several years ago, but are readily available on Robinhood and other apps today. In the future, if stocks and bonds are tokenized, maybe you’ll be able to do all sorts of cool things with them by interacting with DeFI protocols (whether you should is a different question). With tokenization, the assets will exist in an open, modern system where much more innovation is possible compared the traditional financial system, which is relatively inflexible (for better or worse). And of course, tokenization let’s you bring all sorts of other assets into your portfolio that either didn’t exist before or weren’t accessible. Music royalty rights, deeds to physical property, items in video games, and yes, meme coins/gambling tokens like $TRUMP and $FART -- the list is endless. Sure, none of this might be appealing for those who simply like to buy and hold index funds (a tried and true strategy), but it holds appeal for many others