You can’t replace ChatGPT for a few million dollars

When I asked the AI tool whether it was better than DeepSeek, its response was 'It depends!'

January 29, 2025, 3:55 PM

It is believed DeepSeek used ChatGPT data, raising questions about its true development cost

On Tuesday US chipmaker Nvidia’s value plunged by $600 billion (it has since recovered by around 50 percent), reflecting investors’ fears about Chinese AI tool DeepSeek and giving rise to a question.

Can you replace OpenAI’s Chat GPT with DeepSeek, an open-source platform, for only $6 million? The answer is no you can’t.

When I asked ChatGPT which was better, ChatGPT or DeepSeek, its response was “It depends!”.

It added: “DeepSeek may have advantages in efficiency, cost and targeted performance for certain technical tasks. However, ChatGPT remains a strong contender due to its versatility, robust ecosystem and established reputation. The ‘better’ model will depend on the specific use case, goals and resources available.”

Think of the AI journey as a chain in which one block can help derive the next block. No one needs to go back to the beginning and start over; rather, they should improve on what has already been built.

According to the paper released by DeepSeek, this is exactly what it has done, and it should be congratulated.

If one puts aside the potential IP infringement issues, DeepSeek will certainly open the minds of the smartest people in the room

Sam Altman, CEO of OpenAI and creator of Chat GPT, called it “an impressive model, particularly around what they’re able to deliver for the price,” though he added that OpenAI would “obviously deliver much better models” moving forward.

It is believed that DeepSeek started with Llama (Meta’s open-source AI platform) and, after adding some guard rails for the Chinese market, applied the model to data from ChatGPT (OpenAI’s proprietary platform).

In which case, the $6 million sum – as claimed by DeepSeek – probably ignored the platform cost incurred by others on which DeepSeek is built.

Ultimately, someone needs to pay for the technology and its upkeep. If Meta stops funding the development of Llama, then DeepSeek could not be developed.

On the other hand, proprietary models such as OpenAI have high entry costs and restrict innovation outside their walls. Co-existing can accelerate innovation, as we can see with DeepSeek.

The debate about which is superior, open source (code that is publicly available) or proprietary (code owned by an organisation), is a red herring.

We need both proprietary models, such as OpenAI, Anthropic and Gemini (Google), which set the direction of travel and can drive investment in infrastructure from data centres to algorithms, and open-source alternatives to encourage the wider developer community.

If one puts aside the potential IP infringement issues, DeepSeek will certainly open the minds of the smartest people in the room.

Marc Andreessen, general partner of the venture capital firm Andreessen Horowitz, described DeepSeek’s R1 on social media platform X as “one of the most amazing and impressive breakthroughs I’ve ever seen,” and praised the company’s decision to open source the model as “a profound gift to the world.”

Yann LeCun, chief AI scientist at Meta said that “open-source models are surpassing proprietary ones”.

There are rumours that Meta had put together a “war room” of engineers to learn from DeepSeek, which may have been built on Llama, and so on.

However, even Meta can’t employ every AI expert or expect to own every iteration on the path of AI utopia: the first to reach AGI, or artificial general intelligence. Machines, which are generally as intelligent or smarter than humans, progress toward super-intelligent systems (far more intelligent than humans).

The timing of the DeepSeek release was unfortunate for President Trump, who described it as a “wake-up call for America’s AI industry”. Only five days ago, Trump announced Stargate, a programme to build “physical and virtual infrastructure to power the next generation of advancements in AI, including the construction of colossal data centres.”

The anchor investors are OpenAI’s Sam Altman and Masayoshi Son from Softbank. Altman has previously courted Middle Eastern funds and partners to build AI infrastructure, capacity, energy and data centres.

China, through DeepSeek, may have managed to score against America’s AI supremacy, but it’s possible it will be chalked up as an own goal. DeepSeek is an interesting application of US technology and will help drive demand and greater efficiency for AI investment.

DeepSeek will be the first of many to claim a faster, cheaper alternative, but when you lift the bonnet, you will find Trump’s “Made in America” stamped on the chipset.

After a torrid trading day yesterday, with some $1 trillion wiped from the market value of US tech stocks, investors returned today, driving Nvidia shares around 10 percent higher. This suggests that many realise that Nvidia’s importance will remain, or at least until earnings day in February!

For now, DeepSeek has announced China’s AI ambitions globally, which is a big gift for today’s Chinese New Year. Kung Hei Fat Choi.

Nish Kotecha is a serial tech entrepreneur and former investment banker. He is the chair and co-founder of Finboot, chair of Agam.ai and advisory board of Innoviti