How DeepSeek Built a ChatGPT Rival at a Fraction of the Cost?

DeepSeek, a little-known Chinese startup, just shook up the tech world with its new AI model, DeepSeek-R1, said to rival giants like Google and OpenAI. What’s really surprising?

The company claims it built it using fewer, less advanced chips than U.S. tech leaders and that it matches or even surpasses leading U.S. models, all at a fraction of the cost. This breakthrough has the potential to disrupt the global tech landscape.

The impact was huge, so much so that Nvidia, the top chipmaker, lost nearly $600 billion in market value as DeepSeek’s breakthrough raised questions about the future of AI dominance.

What is DeepSeek?

DeepSeek is an AI company based in Hangzhou, China. It was founded in May 2023 by Liang Wenfeng, a Zhejiang University graduate. Wenfeng is also a co-founder of High-Flyer, a quantitative hedge fund that owns DeepSeek. Despite this connection, DeepSeek operates as an independent AI research lab.

The company focuses on developing open-source large language models (LLMs). It introduced its first model in November 2023 and has continued to refine its technology through multiple iterations. However, it wasn’t until January 2025, when DeepSeek launched its R1 reasoning model, that it gained widespread global attention.

Users can interact with DeepSeek’s models in different ways, including through a web platform, a mobile app, and API access.

DeepSeek is unique because the company claims it was developed at a much lower cost than top models like OpenAI by using fewer advanced chips.

DeepSeek focuses on developing open-source LLMs. The company’s first model was released in November 2023. The company has iterated on its core LLM multiple times and built several variations. However, it wasn’t until January 2025, after the release of its R1 reasoning model, that the company became globally famous.

How Did DeepSeek Build a ChatGPT Rival for Just $6M?

Imagine building a powerful AI model that can take on ChatGPT, but at a fraction of the cost. That’s precisely what the Chinese startup DeepSeek has done with its latest language model, DeepSeek R1. While OpenAI’s models reportedly needed around 10,000 GPUs to train in 2023, DeepSeek pulled it off with just 2,000 Nvidia chips.

And the best part? They did it all for just $6 million. (though that $6 million covered only the final training run, not the underlying infrastructure). Compare that to the massive $500 billion Stargate project, and it raises a big question: do we really need such massive AI investments to stay competitive?

Affordable and Accessible AI

One of the biggest reasons DeepSeek R1 is turning heads is its affordability. You can try it out for free on their web app (chat.deepseek.com), and if you need an API, it’s way cheaper than OpenAI’s.

To put it in perspective, DeepSeek charges just $0.14 per million cached input tokens, compared to OpenAI’s hefty $7.50 per million. No wonder developers are rushing to check it out!

The Secret Sauce: How DeepSeek R1 Was Trained

So, how did DeepSeek pull this off? It all comes down to their unique approach to training AI. Instead of relying on expensive, high-powered AI accelerators and long training times, they used:

• Reinforcement Learning – Optimised for reasoning tasks to improve problem-solving skills.
• Reward Engineering – Instead of traditional neural reward models, they used a rule-based system to enhance learning efficiency.
• Distillation – They compressed AI knowledge into smaller models (as few as 1.5 billion parameters) without sacrificing performance.
• Emergent Behaviour Networks – A game-changing discovery that allows the AI to develop complex reasoning patterns naturally, without being explicitly programmed.

Why is DeepSeek Disrupting the AI Industry?

DeepSeek’s breakthrough isn’t just making waves in AI development; it’s shaking up the entire tech industry. NVIDIA’s stock took a 13% hit after reports surfaced about DeepSeek R1’s efficiency.

This has sparked a heated debate: are we moving toward a future where low-cost AI models replace the need for massive, resource-hungry systems? It seems you don’t need a billion-dollar budget or the most powerful chips to push the boundaries of innovation. And that’s making some big players nervous.

Meanwhile, OpenAI, valued at $157 billion back in early 2025, is under growing pressure to prove it can stay ahead of the game. With sky-high spending and questions about its actual returns, investors are starting to wonder if the hype is justified.

The shockwaves from DeepSeek’s cost-efficient approach hit financial markets hard on January 27, sending the Nasdaq down over 3% in a massive sell-off that rattled chipmakers and data centres worldwide.

No one felt the impact more than Nvidia. Its stock nosedived 17% on Monday before clawing back about 4% by midday Tuesday. The chip giant, once the world’s most valuable company by market cap, slipped to third place behind Apple and Microsoft, with its valuation dropping from $3.5 trillion to $2.9 trillion, according to Forbes.

As for DeepSeek? It’s a privately owned company, meaning you won’t find its stock on any major exchange. However, with moves like these, the entire industry is paying attention.

DeepSeek’s success is turning the AI world on its head, making both Silicon Valley and the US government question their entire approach.

For a long time, US tech giants were thought to have an unbeatable edge, massive budgets, access to the best talent worldwide, and the ability to build huge data centres packed with high-end chips. But now, that assumption isn’t looking so certain.

DeepSeek’s ultimate vision is to push AI closer to Artificial General Intelligence (AGI). With their focus on affordability, efficiency, and advanced reasoning capabilities, they’re proving that cutting-edge AI doesn’t have to come with a sky-high price tag. The question is, will the rest of the industry follow the same pattern?

Top Challenges Facing DeepSeek

With 3 million downloads and skyrocketing popularity, DeepSeek is outpacing competitors, yet the U.S. Navy has banned it, and the White House is investigating security implications. DeepSeek’s sudden rise is shaking up the AI market, but a major security flaw has raised alarms; researchers found it exposed over a million user records, including API tokens, to the open internet.

The app collects vast user data stored in China, including chat histories and keystroke patterns. While its privacy policy mirrors competitors like ChatGPT, experts urge caution.

Let’s break down the top concerns:

1. Privacy and Data Security

What started as a US Navy ban has snowballed into a global pushback. Italy was first to pull DeepSeek over GDPR concerns, and it has since been restricted by governments in Australia, Taiwan, South Korea, India, the Czech Republic and the Netherlands, alongside a string of US states such as Texas, New York, Virginia and Florida, and federal bodies including NASA, the Pentagon and the Department of Commerce. In Congress, lawmakers have advanced a bipartisan “No DeepSeek on Government Devices Act”, while companies including Microsoft, Toyota and News Corp have barred staff from using it. The common thread is the worry you opened with: where your data goes, and who can legally demand it under China’s National Intelligence Law once it lands on servers in China.

2. Censorship and Content Control

Users have noticed that DeepSeek avoids discussing topics sensitive to the Chinese government, such as the Tiananmen Square protests and Taiwan’s political status. The AI often responds with messages like, “Sorry, that’s beyond my current scope. Let’s talk about something else.” This built-in censorship raises concerns about freedom of information and potential bias in responses. 

3. Geopolitical Implications

DeepSeek’s rapid advancement in AI without reliance on high-performance chips has caught global attention. This development challenges existing tech dynamics and has prompted discussions about the need for the U.S. to bolster its AI capabilities to remain competitive.

While DeepSeek offers innovative AI interactions, users and policymakers need to weigh these benefits against the potential risks. As with any emerging technology, a balanced approach is crucial to harness its advantages while safeguarding against possible pitfalls.

Is DeepSeek the Future of AI?

DeepSeek isn’t just another AI model. It’s redefining how artificial intelligence is built and deployed. Instead of relying on massive computational resources like its Western counterparts, DeepSeek embraces a leaner, more efficient approach, proving that innovation isn’t just about scale but about smarter strategies. This shift challenges the dominance of big tech players, opening doors for new players to enter the AI race.

What makes DeepSeek’s rise so significant is how it reshapes AI development. With the emergence of distinct AI pathways, open-source application developers, large research labs, and domain-specific experts, AI is becoming more accessible, competitive, and environmentally sustainable.

By breaking the traditional mould, DeepSeek could democratize AI, allowing smaller companies and startups to harness powerful models without massive budgets. If this trend continues, DeepSeek might just be the blueprint for the future of AI.

DeepSeek in 2026: What Happened Next

A year on, the DeepSeek story looks very different. The biggest surprise? That January 2025 panic never repeated itself. Nvidia, Broadcom and ASML didn’t just claw back their losses; they kept growing, with Nvidia going on to become a roughly $5 trillion company. Rather than slowing down, AI spending actually accelerated through 2025, and analysts now frame that first shock as a one-off repricing rather than the start of a cheap-AI revolution.

Through 2025, the company shipped a steady run of updates to its V3 and R1 models rather than a brand-new flagship, each squeezing out more efficiency. Its hotly anticipated R2 model, originally pencilled in for May 2025, slipped after the team hit problems training it on homegrown Huawei chips.

The headline release came in April 2026 with DeepSeek V4. Like everything before it, V4 is open source and brings significant jumps in reasoning and agentic ability, meaning it can act on your behalf, such as by writing code. The twist is what’s powering it: where R1 leaned on Nvidia hardware, V4 was built on domestic Chinese chips, Huawei’s Ascend line and Cambricon silicon. Some analysts reckon that shift could end up mattering more than R1 ever did, because it shows advanced models can be trained without Nvidia at all.

There’s a money story too. For most of its life, DeepSeek was bankrolled by founder Liang Wenfeng’s hedge fund, High-Flyer, and took no outside cash. That changed in 2026, when it opened its first external funding round, reportedly valuing the company at up to $50 billion with backing from Tencent and China’s state-linked chip fund. DeepSeek has gone from scrappy lab to national AI champion in record time.