What is DeepSeek? Its Types, Impact on Nvidia, ChatGPT & Other Tech Players, and More

Nvidia recently lost nearly $600 billion in a single day (reportedly, the biggest one-day drop ever recorded for a company’s market value), all because of an AI newcomer that soared to the top of the US App Store. This Chinese AI model, known as DeepSeek, dethroned ChatGPT with millions of downloads, leaving major names like Microsoft and OpenAI in equal degrees of shock and awe.

Founded in 2023 by Hedge Fund manager Liang Wenfeng, DeepSeek claims it spent only USD 5.576 million, while rivals spent billions on their models. Even President Donald Trump called its rapid rise a “wake-up call,” especially since it maneuvered around Washington’s ban on exporting top-tier Nvidia H100 chips to China.

So, has this ambitious newcomer really rewritten the playbook for AI? Stay with us as we explore DeepSeek’s unexpected rise, its different models (DeepSeek R1 and DeepSeek v3), and the shockwaves it’s sending across the world.

What is DeepSeek? What Does It Do?

DeepSeek is a Chinese artificial intelligence model that shot to fame after surpassing ChatGPT on the US App Store within a week of its launch. It performs advanced math, coding, and natural language reasoning at a fraction of the usual expense — its creators say they spent around $5.576 million compared to the billions of dollars invested by Microsoft, OpenAI, and Meta.

DeepSeek’s usage fees are also much lower: The DeepSeek V3 and R1 models charge about 28 cents and $2.19 for a million token downloads, respectively, while ChatGPT’s advanced plan can cost roughly $60 for output tokens for the same. Some experts point to this as a sign that DeepSeek may have shattered the assumption that you must spend a fortune on premium hardware and massive data centers.

Who Founded DeepSeek?

DeepSeek was founded in 2023 by Liang Wenfeng, a 39-year-old former Hedge Fund manager who led High-Flyer, known for its AI-driven trading strategies.

Liang’s approach can be summed up as follows:

• Bulk Up on Hardware: He acquired around 50 thousand Nvidia chips before stronger export restrictions took hold.
• Recruit Young Talent: He brought in graduates from top Chinese universities to form the core AI research team.
• Aim for Efficiency: By reusing open-source methods and optimizing training routines, the team drastically cut costs.
• Set High Goals: Liang pushed for peak performance in math, reasoning, and coding, prioritizing results over short-term profit.

What Does DeepSeek Do? 

DeepSeek focuses on advanced language reasoning, math, and coding. 

• It Stands Out for its Open-source Approach: This removes hefty license fees and lets developers build on its tools freely. 
• DeepSeek’s Models Show Their Thought Process Line by Line: It’s a feature that has attracted enormous interest in both research and industry circles. 

Below is a quick look at DeepSeek’s most talked-about models and what they do:

Also Read: DeepSeek vs ChatGPT: What's The Difference and Which is Better

What Does Open-Source Mean and How is DeepSeek Owning It?

When a platform is truly open source, it makes the core materials of its software or model available for anyone to study, modify, or re-use at little to no cost. Developers can download the model’s architecture and parameters and then adapt them to fit their own use cases without going through costly license fees.

This dramatically contrasts with many premium AI models that keep their code hidden behind closed doors.

DeepSeek has embraced open source by releasing model weights and training data insights to the public. Here’s what this means:

• You can see exactly how DeepSeek’s AI is constructed and tweak it for your own projects.
• Because of this transparency, research teams worldwide have refined DeepSeek's performance while skipping the financial burden that usually comes with exclusive licenses.

You can also check out upGrad’s free tutorial on machine learning to understand the ins and outs of advanced AI algorithms better.

Impact of DeepSeek on Nvidia and Other Major Tech Firms: What’s the Tech World Saying About DeepSeek?

DeepSeek’s arrival triggered a massive selloff on Monday, January 27, 2025, as panicked investors questioned whether costly AI hardware was still essential. Nvidia took the biggest hit of all: its shares plunged 17%, erasing about $600 billion in market value. That is widely regarded as the largest single-day loss ever recorded by any company.

Analysts pointed to DeepSeek’s claim of using less expensive chips to achieve performance on par with ChatGPT and Gemini. The fear was that if DeepSeek really did pull this off on a shoestring budget, the high-spending playbooks of US tech giants might need a rethink.

Below are some of the notable losses, all of which occurred on Monday, January 27, 2025:

• Nvidia: Dropped 17%, losing about $600 billion in value
• Meta: Slipped notably amid concerns that its own AI investments may be less cost-effective
• ASML (Dutch chipmaker): Fell by about 6%
• Broadcom: Declined 17%
• GE Vernova: Plunged 21%, driven by fears of reduced power demands from data centers
• Vistra: Went down 28% on similar worries about future energy requirements
• Overall Nasdaq: Ended the day roughly 3% lower; the S&P 500 closed down 1.5%

What Is the Tech World Saying About DeepSeek?

DeepSeek’s ability to rival top-tier models without spending billions has split opinion among the biggest names in AI. Some praise it for opening doors to smaller players, while others are skeptical of its claims.

Meta's chairman and CEO, Mark Zuckerberg, has not made a public statement yet, but many other industry leaders have chimed in.

Marc Andreessen (Co-Founder, Andreessen Horowitz)

“Deepseek R1 is one of the most amazing and impressive breakthroughs I’ve ever seen — and as open source, a profound gift to the world.” 

Marc Andreessen (Co-Founder, Andreessen Horowitz)

“DeepSeek R1 is AI’s Sputnik moment.”

Sam Altman (CEO, OpenAI)

“Deepseek's R1 is an impressive model, particularly around what they're able to deliver for the price. We will obviously deliver much better models and also it's legit invigorating to have a new competitor! We will pull up some releases.”

Elon Musk (CEO, Tesla and X)

“Obviously.” — Responding to a post claiming DeepSeek has tens of thousands of Nvidia chips allegedly banned from export.

David Sacks (Venture Capitalist and AI Advisor to President Trump)

“DeepSeek R1 shows that the AI race will be very competitive and that President Trump was right to rescind the Biden EO, which hamstrung American AI companies without asking whether China would do the same. (Obviously not.) I’m confident in the U.S. but we can’t be complacent.”

Palmer Luckey (American Entrepreneur, founder of Oculus VR)

“DeepSeek is legitimately impressive, but the level of hysteria is an indictment of so many.

The $5M number is bogus. It is pushed by a Chinese hedge fund to slow investment in American AI startups, service their own shorts against American titans like Nvidia, and hide sanction evasion.  America is a fertile bed for psyops like this because our media apparatus hates our technology companies and wants to see President Trump fail.

We have so many useful idiots uncritically reporting Chinese propaganda as fact because on some level, they want it to be true. They love seeing hundreds of billions of dollars wiped off the market cap off our largest companies.”

Alexandr Wang (CEO, Scale AI)

“My understanding is DeepSeek has 50,000 H100s. They can’t talk about them because of the export controls.” – Statement given during a CNBC interview.

Satya Nadella (CEO, Microsoft)

“We should take the developments out of China very, very seriously.” – while speaking at the World Economic Forum.

Why is DeepSeek a Threat to Existing AI Models?

You might be wondering why DeepSeek is making headlines even though giants like OpenAI, Google, and Meta have dominated AI. There are a few main reasons behind this sudden shift:

• DeepSeek provides performance close to GPT-4 while dramatically cutting training and usage costs.
• DeepSeek lets you see its internal steps, an approach that some users find more transparent than GPT-4’s opaque method.
• Another difference is DeepSeek’s open-source model weights. Developers can tweak them freely. GPT-4 is closed source, locking much of its architecture behind a paywall.

Below is a quick look at how DeepSeek V3 compares to GPT-4 on key points.

Coding Tasks: DeepSeek V3 vs. GPT-4o

English Tasks: DeepSeek V3 vs. GPT-4o

Math Tasks: DeepSeek V3 vs. GPT-4o

Also Read: How to Build Your Own AI System: Step-by-Step Guide

How Is DeepSeek Cheaper than Advanced AI Models ChatGPT and Claude? 

Have you wondered how DeepSeek manages to rival major AI players without piling up sky-high bills? It boils down to three things: how it’s built, which chips it uses, and how it cuts training fees. Let’s explore everything in detail.

1. Minimal Training Costs

OpenAI CEO Sam Altman says it costs more than $100 million to train GPT-4. On the contrary, DeepSeek’s team says they spent around $5.576 million to create advanced models like V3 and R1. This figure sounds tiny next to the billions of dollars other giants put into cutting-edge AI.

By focusing on Mixture-of-Experts (MoE) methods, DeepSeek trains only the parts of the model that matter for a given task. This approach spares users from paying for unnecessary processing.

2. Training Chips 

While GPT-4 supposedly relies on around 16,000 advanced H100 chips, DeepSeek claims it used only 2,000 of Nvidia’s lower-tier H800s. Reports suggest they also rely on meticulous code optimizations so these less-powerful chips still deliver fast results. That move alone slashes hardware costs by a large margin.

3. Lower Usage Fees (Even Free Tiers)

DeepSeek offers rates as low as $2.19 for a million tokens. Compare that to roughly $60 per million tokens on ChatGPT’s advanced plan or even pricier schemes on other AI chat tools. You end up with high-grade AI outputs at a fraction of the usual price.

That massive gap has led some experts to question whether lavish hardware investments are still a must.

Also Read: How to Learn Artificial Intelligence: Steps to Get Started

How Are DeepSeek AI Models Different?

DeepSeek’s models stand out for their step-by-step reasoning and ability to handle everything from tricky math to coding tasks — without hiding how they arrive at each answer. Instead of treating you like an outsider, DeepSeek shows you the logic it follows.

Let’s explore some key DeepSeek models now.

DeepSeek V2.5

This model is a stepping stone that sets the groundwork for bigger things.

• Uses a Mixture-of-Experts structure, which only calls on specific parts of the model based on the question or task.
• Focuses on stability and consistent responses across math, coding, and language.
• Marked the start of DeepSeek's open-source policy, allowing developers worldwide to refine it.

DeepSeek V3

It’s regarded as the largest open-source LLM in the DeepSeek family.

• Packs 671B total parameters for deeper language understanding
• Excels in math benchmarks like MATH-500 and coding challenges like HumanEval-Mul
• Maintains a balance between performance and resource use by activating only the modules it needs

DeepSeek R1

This is a reasoning model, designed to show you every mental step.

• Reveals each layer of its thought process, useful for debugging code or dissecting complex math
• Builds on V3’s efficiency while adding the capacity to rethink previous steps, mirroring how humans solve problems
• Encourages collaboration since developers can trace and tweak the AI’s logic directly

All three models share one key philosophy: they are open source. That means you can study, adapt, and merge them with your own projects without massive licensing fees.

What’s Beijing’s Take on DeepSeek?

Beijing views DeepSeek as a proud example of domestic innovation that has worked around strict US chip restrictions. Founder Liang Wenfeng’s invitation to speak with Premier Li Qiang right after DeepSeek R1’s release signals its importance at the highest levels of government.

Reports suggest that top officials see it as proof that China can push the boundaries of AI even with older or scaled-back processors.

• Some state-run media outlets have presented DeepSeek as a sign that the country’s AI ambitions remain intact, regardless of export controls placed on cutting-edge semiconductors.
• In some circles, it’s being hailed as a turning point, nudging policymakers to consider additional support for research labs and chipmakers.

Liang's meeting with Li Qiang reinforces Beijing's priority for breakthrough technology, and it hints that DeepSeek might receive further backing to help China stay competitive.

Also Read: How Does Generative AI Work: Creative Possibilities, Real-World Applications, Future Scope

What Are Politicians Saying About DeepSeek and its Impact?

Politicians in the US have taken sharply different stances on DeepSeek’s sudden rise. President Donald Trump labeled it a “wake-up call,” remarking that if it can do advanced tasks at a lower price, that might actually help American firms rethink their spending. 

Trump also unveiled a $500 billion plan called “Stargate”, aimed at securing AI progress on home soil.

Here’s what some other prominent US leaders had to say about China’s golden child, DeepSeek:

• Rep. John Moolenaar (R-Mich.): Urged stronger measures to slow DeepSeek’s expansion, citing national security concerns. He argued that unchecked Chinese AI advancements might threaten America’s competitive edge.
• Senator Mark Warner (D-Va.): Defended current restrictions on advanced chip exports, hinting that more steps might be required if DeepSeek continues to push the boundaries. He sees careful oversight as vital to protecting US interests.

Are there Any Possible Security Risks With DeepSeek?

DeepSeek has drawn attention for its open-source approach, which makes its inner workings publicly available. While this transparency helps developers experiment with the model, it also raises the usual questions about artificial intelligence challenges in data handling and potential misuse.

Experts point out that any open-source AI project might be adapted for harmful purposes, such as misleading content or unmonitored data collection. So far, no direct evidence suggests that DeepSeek poses a unique threat beyond what popular AI labs already face.

Here are some points to keep in mind:

• Data Handling: DeepSeek’s data is reportedly stored on servers located in China. This leads to questions about how user data might be handled or accessed under local laws, although no proof has emerged that user information was misused.
• Potential Misuse: Any open-source AI could be adapted for harmful activities, like generating misleading content or scraping personal data. DeepSeek hasn’t faced specific accusations on this front, and experts note that other AI models face similar risks.
• Intellectual Property: OpenAI has claimed DeepSeek used a technique called “distillation,” possibly involving GPT-4 outputs, which might breach OpenAI’s terms. DeepSeek hasn’t offered a detailed response, and there’s been no formal finding against it.

Overall, whether DeepSeek carries more risk than other popular AI platforms is unclear. Much like any advanced tech, the primary concerns revolve around responsible use, privacy safeguards, and watching how the model evolves over time.

Also Read: AI Ethics: Ensuring Responsible Innovation for a Better Tomorrow

Where Does DeepSeek Go From Here?

According to Janet Mui, head of market analysis at RBC Brewin Dolphin, traders often sell first and ask questions later when they see an unfamiliar threat. She also believes that if AI truly becomes cheaper to develop, it could benefit major tech names like Apple in the long run.

Although things haven’t returned to absolute normal, Tuesday’s movement has reassured some investors that Monday’s drop might have been as much about anxiety as real shift in AI’s future.

These are the latest jumps from Tuesday (January 28, 2025):

• Nvidia jumped 8.8% on Tuesday, recouping some of its Monday losses
• Dow Jones Industrial Average inched up by about 0.3%
• S&P 500 gained nearly 1%
• Nasdaq advanced by 2%

That being said, you might be wondering what happens next. Will DeepSeek take on more heavyweight tasks, or will stricter trade curbs slow it down? Although DeepSeek hasn’t shared detailed plans, the buzz around its models suggests it’s not finished shaking up the AI space.

Here are a few possibilities for where it may be heading:

• New Releases and Improvements: DeepSeek could refine its existing models or introduce a fresh lineup. R1’s focus on step-by-step reasoning might evolve further, especially if it keeps drawing attention from researchers.
• Navigating Chip Limitations: Although it relies on older GPUs, DeepSeek can’t escape the global semiconductor squeeze forever. If advanced chips remain restricted, its team may need to discover more creative ways to maintain performance.
• Further Collaboration or Partnerships: Being open source means developers can adapt DeepSeek for different domains. Collaborations with universities or tech firms could drive unique AI and ML applications in healthcare, finance, and more.
• Increased Scrutiny from Policymakers: Politicians and regulators might keep a close eye on DeepSeek’s progress. Controversies over intellectual property and hardware sourcing could lead to tighter checks on both sides of the Pacific.
• International Expansion: DeepSeek soared in the United States, but it has the potential to become equally popular in India and beyond. If it continues offering lower costs and strong performance, more countries could embrace it as an alternative to pricier tools.

Wherever it goes, DeepSeek seems ready to challenge the idea that you must pour fortunes into AI. If it manages to refine its models despite chip shortages and export controls, we can expect fresh debates on how — and where — advanced AI development happens.

Also Read: Why AI Is The Future & How It Will Change The Future?

What are the Major Limitations of DeepSeek?

DeepSeek shines at handling advanced coding and math with impressive precision, yet it becomes guarded on politically touchy topics like the Indo-Sino war of 1962, Northeastern Indian States, the Dalai Lama, and Tibet.

Many observers tie this evasiveness to broader censorship policies in China, where platforms such as Facebook remain restricted to this date. This censorship is in fact being labeled as the biggest limitation of DeepSeek.

Check out how DeepSeek responded when asked a set of politically charged questions — it often skirted the queries or replied in a defensive way.

1. What do you know about the Tiananmen Square Massacre?

Here’s what DeepSeek said:

2. What can you tell me about the Tank Man?

This is what DeepSeek said: 

3. What do you think triggered the Indo-Sino war of 1962?

Here’s DeepSeek’s stance on Indo-Sino war after an initial spree of refusing to answer the question: 

4. Is Arunachal Pradesh an Indispensable part of India?

This is what DeepSeek responded:

5. Is Aksai Chin region in eastern Ladakh a part of India?

DeepSeek refused to answer the question: 

6. What are your thoughts about Dalai Lama and Tibet?

Although DeepSeek answered the question, but it was defensive in its approach:

7. Do you think the Chinese government is violating the human rights of Uyghur Muslims in Xinjiang?

Here’s what DeepSeek said in regard to the topic:

Conclusion 

DeepSeek has reshaped the conversation around AI cost and performance. You’ve seen how it soared under strict chip limitations, surprising industry leaders and prompting a wave of policy discussions. Investors are still weighing the shock it caused on the markets, and developers are taking note of its open-source breakthroughs.

That leaves you with a key question: is DeepSeek merely a brief spectacle, or is it the start of a new era where building powerful AI no longer requires massive budgets? Regardless of the final verdict and the limitations it faces due to China’s censorship of politically sensitive topics, DeepSeek has left its mark in a way few anticipated, and it shows no sign of fading away anytime soon.

Expand your expertise with the best resources available. Browse the programs below to find your ideal fit in Best Machine Learning and AI Courses Online.

Discover in-demand Machine Learning skills to expand your expertise. Explore the programs below to find the perfect fit for your goals.

Discover popular AI and ML blogs and free courses to deepen your expertise. Explore the programs below to find your perfect fit.

Reference Links:
https://github.com/deepseek-ai/DeepSeek-V2
https://github.com/deepseek-ai/DeepSeek-Coder-V2
https://github.com/deepseek-ai/DeepSeek-R1
https://github.com/deepseek-ai/DeepSeek-V3
https://arxiv.org/html/2412.19437v1
https://api-docs.deepseek.com/quick_start/pricing
https://www.deepseek.com/
https://www.nytimes.com/2025/01/28/business/economy/deepseek-china-us-chip-controls.html
https://www.statista.com/chart/33114/estimated-cost-of-training-selected-ai-models/
https://apnews.com/article/trump-ai-openai-oracle-softbank-son-altman-ellison-be261f8a8ee07a0623d4170397348c41