DeepSeek AI intensifies the AI War and disrupts the AI-Powered tech industry

Did DeepSeek really kill Nvidia's momentum, or it showed possibilities to squeeze more from the powerful Nvidia chips?

Feb 04, 2025

This article is part of my Viewpoints collection that focuses on current events in the world of technology, data and AI, a space where I can share my viewpoint.

It is interesting to see the roller-coaster of events in the AI world in January.

Nvidia provided scale but at a high cost. Is that cost still justified?

Lets find out…

Image generated by the author using Microsoft Copilot

January 06, 2025 - Consumers Electronic Show (a.k.a. CES) Day 1

Everyone was talking about the CES keynotes by Jensen Huang, the CEO and co-founder of Nvidia.

Here is a link to an article I published last month, summarizing the keynotes and key takeaways. Nvidia CES Keynotes 2025

January 20, 2025 - DeepSeek R1 launched its new open-source large-language model (LLM) in competition with ChatGPT, an AI chatbot developed by OpenAI.

DeepSeek AI, a Chinese-based AI startup, released its AI Assistant, which uses the V3 model as a chatbot app for Apple IOS and Android.

January 27, 2025 - DeepSeek sparks AI stock selloff; Nvidia posts record market-cap loss. Full story

DeepSeek R1's AI Assistant app took the entire US Technology Industry by surprise, surpassing ChatGPT to become the highest-rated free app on the iOS App Store in the United States. This unexpected success was a key factor in triggering a significant selloff in the AI stock market, and Nvidia suffered a record market loss.

So, what is so unique about DeepSeek R1?

Why did the launch of an open-source LLM by a Chinese AI company lead to a staggering $1 trillion loss in the US stock market, as per finance experts? What was the unique factor that propelled DeepSeek R1 to such heights, and why did it catch the entire industry off guard?

This includes Nvidia losing a significant chunk of its net worth (close to $589 billion with more than a 16% decline in its stock price). Such a panic sellout in a day was one of its kind.

Let’s understand this…

November 2022 - the AI war was reignited with the launch of ChatGPT by OpenAI.

A GPT-based LLM using Generative Pre-trained Transformer(GPT) 3.5 can generate content using chatbots or prompts using natural language query as input.

The new wave of Generative AI (GenAI) changed the way we create content, perform customer service, use natural language to interact with chatbots to find relevant responses, use copilot to optimize productivity and streamline workflows in our day-to-day activities.

GPT models in their pre-trained phases required costly infrastructure to train and retrain the models. Hence, semiconductor chipmakers were the driving force behind the ability to process pre-trained data at scale.

According to TechRadar, following are the popular LLMs available

GPT and GPT 4.0 (with multi-modal capabilities) from OpenAI
GitHub (GH) Copilot (best LLM for coding). Microsoft acquired GitHub, and and uses the OpenAI GPT-4 for its GitHub Copilot
Llama 3.2 from Meta. It is an open-source LLM, which performs better than GPT4.
Claude 3 from Anthropic Claude LLM is best for businesses.
Gemini 1.5 from Google is the best LLM for translations.

This is a well-known landscape of open-sourced and proprietary AI models leading the global AI wave.

Disruptive Innovation through low-cost AI alternatives is bound to happen…Why did it come as a surprise?

Chinese counterparts across many industries have taken low-cost hardware and open architecture alternatives in disrupting innovations.

DeepSeek AI is no different.

Since the beginning of November 2023, DeepSeek AI has launched a DeepSeek coder that uses an open-source architecture similar to Llama and low-cost semiconductor chips to train the foundation models. This was followed by multiple versions of foundation models with reinforced learning. The latest version is DeepSeek R1 V3.

DeepSeek R1's cost was significantly lower, ranging in $6 million, compared to similar models trained by OpenAI and Meta, which required 10x the hardware cost (in the range of $100+ million). This cost-effectiveness reassures investors about the potential return on their AI investments.

It uses less than a tenth of the hardware cost to generate similar or even better results. This is adequate to make it an overnight success.

But was it a cutting-edge architecture?

Will it change any course in the intensified AI War between open-source and proprietary AI assistants?

Are all AI investments overvalued?

Viewpoint

The real AI wave is just getting started.
- The era of pre-training LLMs with singular modality is becoming a commodity.
- The future of AI is bright, with the potential of multi-modality using vision learning, and spatial computing to understand and interact with physical surroundings. This emphasis on AI's future potential will make the customers feel optimistic and excited about the possibilities.
Going forward, it is more about AI applications and production implementations than about the LLMs and foundation models.
- Companies like Palantir Technology and its performance in the stock market certainly attest to this. The stock almost doubled in the recent earnings report in February 2025. Their core focus is implementing AI use cases in the Data and Analytics space for their customers in broader industries including US Government and National Security.
Disruption like DeepSeek is inevitable. However, this situation is bullish for AI and tech giants like Nvidia.
- Thanks to DeepSeek R1, Nvidia GPUs have 30x more potential. DeepSeek's staggering low-cost approach confirms Nvidia technology but also shows possibilities for doing more with powerful GPUs and processing powerful AI workloads.

What’s Next?

Tech giants like Nvidia, Meta, and Microsoft confirm their position and stay bullish on AI investments.
Nvidia states that DeepSeek R1 is a perfect example of using widely available models and AI advancements using test-time scaling and reasoning. They confirmed that we no longer need Inference that uses pre-trained or post-trained learned patterns and synthetic training data, which requires significant numbers of NVIDIA GPUs and high-performance networking.
However, in this turmoil of DeepSeek and how the market reacted to the sell-off,
- Microsoft defended this position by stating that energy is more critical than AI-powered chips to power their data centers and serve the demands of AI workloads.
- OpenAI is all set to launch its AI Agent for deep research. It is also making news with its plan to raise $40 billion on a market valuation of $340 billion, making it the most valuable private company.
- Google Cloud confirms the availability of Nvidia B200 (a.k.a. Nvidia Blackwell) and is the first cloud provider to offer it to its customers to support the demand for more powerful AI workloads.
- In the same week, Amazon confirmed the support for DeepSeek R1 models on its AWS Bedrock and Sagemaker platforms. Similarly, Databricks also confirmed that they support DeepSeek R1 on their AI Platform.

Key Takeaways-

It was high time for some disruption of Innovation to happen in the AI world. Disruption in terms of low-cost alternatives that provide adequate performance but are not necessarily cutting-edge.
The launch of DeepSeek R1 caused that disruption and panic of sell-off in the US stock market. However, it was a correction that was low-hanging for a long time.
DeepSeek R1 and its cost theories and performance have confirmed that widely available open-source models and Test-Time scaling with reasoning are faster ways to build AI assistants.
- Multi-modality added a new dimension to AI use cases, like processing text and images as input to generate content. Now, vision-learning and spatial computing models are reinforced in training the physical aspects, like surrounding and building relationships with the space around us.
- With test-time scaling and reasoning, AI model performance is significantly improved on complex tasks, as opposed to an inferencing mechanism that uses a huge GPU for synthetic training data and navigating pre-trained learned patterns.
Tech giants like Nvidia, Microsoft, OpenAI, Meta, and Google remain firm on their AI vision and investments.

With AI advancements, reasoning theories, and vision learning, it is time to build the future of AI with AI Agents and Physical AI, perhaps much cheaper than previously estimated.