Mickey McManus

shredding pixels, shifting shapes

In 2026, recruiters will be hiring AI Agents for companies, not human employees. Companies will opt to pay $10,000+ per month for agents that work 24/7 instead of employees that work 40 hours per week.

Apparently the market is not too happy to see US AI supremacy in question by DeepSeek's R1 reasoning LLM. NASDAQ futures are down 1.77% and S&P 500 futures are off by 0.84% as I write this and news of DeepSeek continues to circulate. My thoughts on the situation: 1) We don't know how many actual GPUs were used by DeepSeek due to export restrictions. China could have massively understated them to create this exact impact: having people question US AI supremacy. 2) If we are to assume the $500M figure is accurate, that would indicate that DeepSeek has found a significantly more efficient path for training, fine tuning and inference 3) Such a situation would call into question CapEx on AI by mega caps and other companies, as results have been much less impressive (again, assuming DeepSeek actually has a much smaller budget, less GPUs in actuality vs stated amounts) If indeed DeepSeek has pulled off what seemed to be impossible with a much smaller amount of computing power, and specificially Nvidia GPUs, that would certainly have some major impacts. On the one hand, it would be excellent news for AI as a whole, as it would potentially usher in a much more sustainable era of resource utilization. It would also significantly lower the barrier to entry for smaller research groups to build their own models. On the other hand, it would be terrible news for the current monoliths like $NVDA, $META, $MSFT, $GOOGL and OpenAI as it would illustrate that enormous spend on GPUs and R&D has been grotesquely inefficient. It would also pressure data center CapEx and other planned spend as a lot of reconsideration would be needed. A lot of this is still very hypothetical, however, as we don't have any way to know what resources DeepSeek actually had when building R1. We can only take the information received and either parse it at face value or decide to add some healthy skepticism. After all, it is China. Export restrictions on technology have also been unevenly effective. Data center-grade GPUs would likely be quite easy to smuggle out in large quantity with rather small amounts of effort by would be smugglers to disguise them as lower value products, like generic gaming video cards or otherwise. We know the export restrictions placed on China have been defied repeatedly, and in large scale, to feed China's growing AI ambitions. There exists a rather large cache of Nvidia H100 GPUs in the country. Easily tens of thousands if not more. So there is reason to believe that China has much more than what they are 'supposed' to have based on these restrictions. As always I hope my thoughts are helpful. As the situation develops I'll add additional analysis. Until then, be skeptical. We don't know enough to make an informed judgment with credible information about what is actually going on.

I am no longer injecting my son's blood. I've upgraded to something else: total plasma exchange. Steps: 1. Take out all blood from body 2. Separate plasma from blood 3. Replace plasma with 5% albumin & IVIG Here's my bag of plasma. Who wants it? 🧵