DeepSeek Is Opening Doors For Advancement Of AI At A Much Lesser Cost

The arrival of the Chinese artificial intelligence (AI) model DeepSeek has given Western tech giants a run for their money. Within a week of its launch on January 20, it has not only gotten the attention of AI enthusiasts but also the tech world.

The efficiency of DeepSeek, made in China at a fraction of the cost of its peers like OpenAI’s ChatGPT, has surprised the world. DeepSeek’s R-1 has zoomed to the global top 10 in performance, according to a popular ranking. It is now the most downloaded app on the Apple store.

Was this really a surprise? Jaspreet Bindra, founder and MD of Tech Whisperer Ltd and co-founder and CEO of Ai&Beyond told The Core that it wasn’t surprising that China wasn’t too far behind the US in terms of AI tech. “But the fact that they were this far ahead and looking at costs which were less than 3% of the costs that the Valley companies are building models at was a super big surprise. I think you see the effect of it when the markets opened in a few hours time.”

Stock prices of almost all tech majors started spiralling downwards on Monday, including AI chip giant NVidia. The slide wiped off $1 trillion from the NASDAQ’s market value. It also raised questions about the sustainability of Western-led tech advancement.

DeepSeek’s foray into the world of tech is not only a threat to Western tech giants, but also opens doors to smaller companies, including in India.

Does AI Need That Much Investment?

Among the concerns after the arrival of DeepSeek was the huge spending by US tech giants on leading-edge semiconductors and other AI infrastructure. Was it really justified?

“I think in many ways the US's efforts to keep China a distant second in the race by the whole chips ban, the semiconductors ban and so on has come back to bite it,” Bindra said.

Being starved of resources, the Chinese engineers were able to find a way to build this on the cheap. “ As we all know well in India —necessity is the mother of invention. And so I think they just approached this entirely fresh, built upon some of the stuff that already had come out in the valley,” Bindra said.

Now, the giant data centres that big companies, including India’s Reliance Industries, have been spending money on, are also being questioned. Big tech giants like Google and Microsoft have committed billions to AI. “They're not as much in the AI game as they are in delivering AI-infused services to their end customers. So in the medium to long term, this could actually be a gain for them, “ Bindra said.

However, things are different for AI infrastructure companies such as NVIDIA. They may fall the hardest now, but it isn’t all doom and gloom. “If suddenly because of this AI model production gets democratised, and 100,000 people can build small models, they will still need GPUs, semiconductors, chips from companies like NVIDIA, " Bindra said.

What Made DeepSeek Cheaper?

Unlike ChatGPT, DeepSeek did not use raw data from the internet to train its AI. The training of Chat GPT or Gemini needed significant computing power, billions of dollars and a lot of time because they were being trained on raw data. These guys (the makers of DeepSeek) used a new technique which people are discovering called distillation, which basically takes ChaTGPT and the results that gives and builds on top of that,” Bindra said.

Bindra called this a part of the “secret sauce” to building such apps and believes that there will be other such technologies that will emerge that we don’t yet know about.

“You know, the fact that they've been able to juice out more power from cheaper last-generation chips, the fact that they've used young, very bright, but very hungry engineers to make this happen are all parts of the mix. I'm sure there's other stuff, too,” Bindra said.

Does India Stand To Gain?

As in other parts of the world, DeepSeek has created quite a stir in India as well. Bindra believes that the success of DeepSeek should make Indians happy despite our diplomatic beef with China.

While cost has been on top of the list of concerns when it came to building large language models in India, DeepSeek's feat now eases that concern.

“It gives an opportunity for India to build its own not one, but multiple large language models. Secondly, it's great for startups and developers, if they were paying what they were paying an open AI or equivalent for their APIs, now they can potentially get it for cents to the dollar, and therefore create much cheaper products that have lower costs. So I think for both these reasons, it's good for India if we grasp the opportunity.” Bindra said.