evening wrap

market vibes

January 27 …

“I always wanted to be somebody, but now I realize I should have been more specific.” Lily Tomlin

What just happened!?

In a nutshell, Liang Wenfeng, residing in Hangzhou, Zhejiang, China, and his hedge fund “High-Flyer,” built an open-source AI product that operates on par with existing technologies but is significantly cheaper to train and deploy. Just like that, the global AI arms race is a victim of its own success. NVDA, and by extension the entire U.S. tech sector, is facing an unsettling question: what happens when AI, the most sought-after asset of the decade, is suddenly (free!?) democratized by China?

The implications are immense, according to X posts and bloggers. Not just for stock markets but for the broader narratives of AI dominance…for example, Trump’s $500 billion AI project with Ellison and SoftBank. Marc Andreessen called it a "Sputnik moment," meaning China has won the AI race, or at least this leg of it. However, there’s ambiguity in how "free" it actually is. Licensing conditions and implicit costs (like data-sharing requirements) are unclear. However, if the narratives are correct, DeepSeek is a huge uptick.

According to posts on X, DeepSeek's latest model, R1, reportedly costs only about $5.6 million to train, which is a small fraction of the costs associated with U.S. models. For instance, OpenAI's costs for training ChatGPT are estimated to be in the range of $100 million to $1 billion for a model like GPT-4. This makes DeepSeek's training costs approximately 18 to 178 times cheaper than those for ChatGPT.

Operational Costs:

Per Token DeepSeek-R1 is about 27.4 times cheaper per token compared to OpenAI's O1 model for operation.

API and Monthly Subscriptions:

DeepSeek offers its services at a significantly lower price point. For instance, DeepSeek starts at just $0.50 per month for its subscription, while ChatGPT's premium models begin at $20 per month. This makes DeepSeek's operational cost for users approximately 40 times cheaper in terms of monthly subscription fees.

It is ironic how quickly climate change faded into silence as Nvidia shot higher last year. AI’s appetite for power was simply ignored. Now what? Another iterative phase of Moore’s Law, with many more to come. (Sources BBG, Google AI, x posts and Chat GPT)

As for trading today:

Most markets were lower. Gold fell $50, silver fell a dollar, crude fell $2.50, NDX and S&Ps were very weak early in the session. However the Dow was strong full stop, finishing up >300 points. Bonds and Yen rallied.

The obvious thing about trading today is that every market did what it wanted to do basis Friday’s close anyway, either up or down, albeit for the same reason. Let’s take a look at the damage and closes and see if anything jumps out for a trade: