$1 Trillion Wiped Out from US Stock Market – Nvidia’s $500 Billion Loss
Yesterday, on 27th Jan, $1 trillion were wiped out from the US stock market. Nvidia alone lost more than $500 billion, and the reason was a Chinese AI model called Deep Seek R1. In this video, I’ll explain what exactly Deep Seek is, why it’s such a big deal, and what it means for the future of AI, as well as your future, in case you’re planning a career in AI.
What Do the Big Players Think About Deep Seek?
Let’s first check what these big players have to say about Deep Seek:
- Nvidia, whose stock was tanking yesterday, calls this an excellent AI advancement.
- Microsoft’s CEO calls it a big win for tech.
Why Is Deep Seek Such a Big Deal?
The number one reason is that it offers low-cost training and inference. It took 2,000 of Nvidia’s H00 GPUs to train this model in 2 months at a cost of $5.5 million. This is much lower than the cost OpenAI spends in training their GPT models. If you compare the cost that US companies are spending on training their LLMs, this cost is peanuts.
Not only that, it’s faster and more accurate. If you look at the benchmark performance, it is performing almost at the same level as GPT, offering very fast inference (the time it takes to generate an answer) and accuracy at par with OpenAI’s flagship models.
Efficiency and Cost Advantage of Deep Seek
Jared Fredman, a partner at Y Combinator, says that Deep Seek is 45x more efficient. The reason for this efficiency is that it uses 8-bit instead of 32-bit floating-point numbers and employs some amazing compression techniques. It can also do multi-token prediction instead of single-token prediction, along with distillation and a mixture of expert models, which decompose a large model into smaller models.
Breaking the Myth of Compute Supremacy
Deep Seek achieved algorithmic efficiency, breaking the belief that compute supremacy is necessary to build the best AI model. Deep Seek and its creators came up with a very fast, optimized way of training so that you don’t need so many GPUs. This is the reason Nvidia’s stock was crashing.
Former Intel CEO on Deep Seek
The former Intel CEO stated that he has already started using Deep Seek instead of GPT. He mentioned that engineering is all about constraints, and Chinese engineers, with limited resources, had to find creative solutions. Due to trade restrictions and bans on high-end GPUs by the US, Chinese engineers had to innovate and prove that the US’s belief in GPU supremacy was wrong.
The Open-Source Advantage of Deep Seek
Deep Seek is an open-source model, which means you can see their approach, research papers, PDFs, and more. If you’re concerned about data safety, you can download Deep Seek locally and run it without sending any data to China.
Performance Comparison
Deep Seek’s performance benchmarks (blue dotted line) compare favorably to OpenAI’s models (gray line) across various tests, such as math, coding, and more.
How to Safeguard Your Data When Using Deep Seek
You can download Deep Seek’s model, run inference locally, and turn off the internet, ensuring that your data won’t go anywhere. It’s also available on platforms like AMA and Gro Cloud, which host the model while ensuring that data doesn’t go to China.
What Does This Mean for the Future of AI?
- Democratizing AI: Deep Seek is democratizing AI, making it more accessible to small companies that can’t afford OpenAI’s API.
- Faster Adoption: Smaller companies in tight-budget regions, like Africa, can now adopt AI faster, which will lead to faster global adoption.
- Geopolitical Impact: China is gaining momentum in the AI race against the US, as they’ve found intelligent workarounds to the GPU export restrictions.
Environmental Impact of Deep Seek
The optimization in the software layer has resulted in a significant reduction in the power needed for training these models. As a result, CO2 emissions are expected to go down, leading to a positive environmental impact.
What Does This Mean for Your Career?
As AI adoption accelerates, companies will need more developers and AI engineers to build AI solutions faster. If you are an aspiring engineer or data scientist, this is a great opportunity to join the growing field of A