Analysis: Why has DeepSeek rocked AI in the US?

US tech giants were stopped in their tracks this week following the release of China-based DeepSeek’s AI model, which has threatened to upend the industry as we know it.

Indeed, the US has had a certain swagger about it when it comes to AI, with powerhouses including Microsoft and Meta Platforms announcing major investment plans at the start of 2025. It was further bolstered by President Donald Trump last week, after he unveiled the Stargate consortium, with Oracle, OpenAI and SoftBank Group pledging to invest around $500 billion into AI infrastructure over the next four years.

With this influx of funding, in addition to the impact of DeepSeek which has apparently been created at a fraction of the cost its rivals have spent on competing models, Mobile World Live (MWL) reached out to industry analysts to find out what this all means for the state of AI play in the US.

On Monday (27 January), the tech-heavy Nasdaq stock exchange was down over 3 per cent while chip powerhouse Nvidia saw its stock tumble by nearly 17 per cent due to the release of the DeepSeek-R1 AI app, which was the top iPhone download in the US the same day.

DeepSeek claims it took two months and around $5.6 million to build an AI model using Nvidia’s less-advanced H800 chips.

Roy Chua, founder and principal analyst at research outfit AvidThink, told MWL the DeepSeekV3 model, which R-1 is based on, was trained on just 2,000 Nvidia H800 chips.

He said when comparing DeepSeek’s 2.7 million GPU hours on Nvidia H800 GPUs versus Meta Platforms’ Llama 3.1 model on 30.8 million GPU hours using Nvidia’s more advanced, export restricted  H100s “the outcome, it’s impressive”.

“Some analysts have expressed scepticism and indicate that DeepSeek has access to tens of thousands of more capable GPUs, though not confirmed,” he added.

Chua noted DeepSeek, which is backed by Chinese hedge fund High-Flyer, releases its models under the MIT licence model with no restrictions unlike Meta Platforms’ open-source Llama models that block its competitors from using it without a licence.

Scott Raynovich, founder and principal analyst at research company Futuriom, told MWL DeepSeek is causing anxiety “because it is leading people to question if we are overpaying for AI infrastructure and about to enter a phase of rapid cost reduction and commodification”.

“This fear may be a bit premature, but it’s likely that the costs of AI infrastructure are reduced over time,” Raynovich explained. “That will be good for enterprise adoption.”

Chua stated the success of DeepSeek’s low-cost models show that tough US export laws against China are not working, that innovation can thrive under constraints and that Chinese companies are adaptable, the latter of which is also borne out by Huawei’s continued growth.

“I believe the US is still in the lead but the gap between the US and the West and China is rapidly closing.”

Matt Walker, chief analyst at MTN Consulting, said that competition between the US and China is the central motivator behind the Stargate funding announcement.

“China has shown it is willing to throw enormous sums of money into R&D initiatives that give it a leg up on other countries, but China’s government can control its private sector,” he said.

He explained both US political parties have been too soft on China policies for decades because “commercial interests have outweighed everything else”.

Both Chua and Raynovich estimate the money needed to develop AI and the surrounding infrastructure in the US is trillions of dollars.

“We are all trying to figure out how much money needs to be spent on AI infrastructure, and how it will be adopted by enterprises,” Raynovich stated.

“We currently know that a trillion dollars or more will be spent on this infrastructure. What we don’t know is how much enterprise demand will be created from this investment or how cost effective it will be.”

Indeed, Trump’s Stargate consortium could help significantly in hitting that target.

Chua stated Trump, SoftBank CEO Masayoshi Son, OpenAI CEO Sam Altman and Oracle executive chair Larry Ellison “tend to paint in broad strokes, leaving the details to be worked out later”.

Aside of the massive investments, Chua noted AI needs to solve a powering problem, which could upend some of the companies’ previous sustainability pledges.

The tech giants also need to prove the technology will continue to scale and that “other software companies can package that technology into consumable services and products”.

“So far, we’re a little behind on value realisation and driven more by the desire to win the arms race, but that won’t last.”

Walker stated the potential $500 billion Stargate investment is “mostly big private companies investing what they already would have,” albeit with subsidised help from the government.

“A good chunk of the promises will never materialise, but the US government could play a useful role in stimulating faster development of renewable energy resources, which is one impediment to making progress in AI,” Walker said.

Trump has said DeepSeek’s release should serve as a “wake-up call” for US tech companies.

GOT A STORY?

Contact: +233243201960 or email: manuelnkansah33@gmail.com