You are currently viewing DeepSeek AI: A Game-Changer or the Next Tech Mirage?

DeepSeek AI: A Game-Changer or the Next Tech Mirage?

Thomas Edison, the self-taught telegraph operator turned entrepreneur, is widely regarded as the greatest inventor of all time, whereas Nikola Tesla, who worked for an Edison company in Paris before immigrating to the United States, is virtually unknown, except through Elon Musk’s electric vehicle company. Tesla’s innovation with alternating current (AC), rather than Edison’s direct current (DC) technique, made mass electrification cheap. The high expenses of direct current would have made Edison’s urban electrification, like many of his other discoveries, a luxury for the wealthy.

Could Chinese investor Liang Wenfeng’s DeepSeek AI models be a comparable AI breakthrough, or are they frauds like cold fusion and room-temperature superconductivity? And if they are verified, should the United States regard them as a threat or as a gift to the world? AI, like many breakthrough technologies, had progressed over many decades before OpenAI’s release of ChatGPT in late 2022, which sparked the present hype. Better algorithms, supplementary devices like mobile phones, and lower-cost, more powerful cloud computing have made the technology widely used yet hardly recognised. Trial and error have demonstrated where AI can and cannot beat human effort and judgement.

The miraculous glibness of ChatGPT and other large language models (LLMs) gave the impression that generative AI was a novel breakthrough. ChatGPT had a million members within five days of its launch and 300 million weekly users two years later. High-tech behemoths such as Microsoft, Meta, and Alphabet made multibillion-dollar bets on AI technologies and data centres, swiftly forgetting their initial enthusiasm for virtual and augmented reality. In 2024, Nvidia, which had invested $2 billion in its Blackwell AI chip, became the world’s most valuable firm, with its market capitalisation increasing ninefold in two years. Jensen Huang, the company’s CEO, anticipated that $1 trillion would be invested in data centres employing such chips over the next five years. All of this made Apple’s cautious, wait-and-see attitude to AI appear quaintly outdated.

Never mind that the new AI provided little benefit to end users in comparison to the massive expenditure. Investments continued to rise on the expectation that hyper-scaled data centres would lower AI costs and that more usage would make the models smarter. However, under their shiny new hoods, LLMs, like many decades-old AI models, continue to create their output through pattern recognition and statistical forecasts, implying that their dependability is dependent on the future being similar to the past. This is a significant constraint. Humans may imaginatively interpret past facts to forecast what could happen differently in the future, and they can enhance their predictions by engaging in imaginative discourse with one another. Artificial intelligence algorithms cannot.

But this defect isn’t deadly. Because natural processes are inherently stable, the future is similar to the past in many respects. AI models can be trained to be more trustworthy when given unambiguous input, and even if the underlying process is unstable or the feedback is ambiguous, statistical predictions can be less expensive than human judgment. Even wildly off-the-mark ads offered out by Google’s or Meta’s algorithms are superior to advertising blindly. Dictating texts to a cell phone may cause howls, but it is still faster and more convenient than typing on a small screen.

By 2022, smart inventors had uncovered several scenarios in which statistically based AI was as good as or superior to alternatives that depended on human judgement. As computer technology and software developed, cost-effective applications were destined to grow. But it was naive to believe that LLMs were a huge step forward merely because they could communicate like humans. In my experience, LLM programs have been completely useless for conducting research, making summaries, or creating visualisations.

Nonetheless, claims of DeepSeek’s capabilities have sent shockwaves across the financial markets. DeepSeek claims to have reached OpenAI and Google-level AI performance with only low-end Nvidia CPUs and a fraction of the training and running expenditures. If this is true, the demand for high-end AI processors will be smaller than expected. That is why the DeepSeek revelation wiped off almost $600 billion from Nvidia’s market capitalisation in a single day, as well as crushing the stocks of other semiconductor firms and corporations that have invested in or sold energy to data centres.

To be sure, DeepSeek’s statements may prove to be incorrect. Many of Tesla’s boasts about his innovations following his AC breakthrough were greatly inflated, if not false, and the Soviet propaganda machine frequently invented scientific and technical triumphs alongside genuine advancements. However, cost-effective, unconventional inventions can have a revolutionary impact. Just look at Musk’s low-cost, reusable rockets. India’s successful Mars mission cost about $73 million, less than the budget for the Hollywood sci-fi film “Gravity”.

If proven correct, DeepSeek’s technology might revolutionise LLMs in the same way that Tesla’s AC breakthroughs revolutionised electric vehicles. While it cannot overcome the inherent limits of backwards-looking statistical models, it may improve their pricing performance sufficiently for widespread adoption. Those building LLM models will no longer need to rely on subsidies from major operators that are interested in locking them in. Less resource-intensive models may lower demand for data centres or redirect their capacity to other economically viable uses.

What about geopolitics? Last March, the Bipartisan Senate AI Working Group recommended $32 billion in yearly “emergency” expenditure on non-defense AI, ostensibly to compete with China. Marc Andreessen, a venture entrepreneur, described DeepSeek’s entrance as “AI’s Sputnik moment.” US President Donald Trump views the Chinese AI model as a “wake-up call for US industries,” which should be “laser-focussed on competing to win.” He has declared intentions to put fresh taxes on Chinese semiconductor imports, while his predecessor placed export limits on high-end AI processors.

DeepSeek’s successes call into question the notion that larger expenditures and top-tier CPUs are the only options to advance AI, raising concerns about the future of high-performance computers. “DeepSeek has proven that cutting-edge AI models can be developed with limited compute resources,” adds Wei Sun, chief AI analyst at Counterpoint Research. “In contrast, OpenAI, valued at $157 billion, faces scrutiny over its ability to maintain a dominant edge in innovation or justify its massive valuation and expenditures without delivering significant returns.”

DeepSeek’s lower prices roiled financial markets on January 27, causing the tech-heavy Nasdaq to plummet more than 3% as part of a global sell-off that included chip makers and data centres. Nvidia’s stock price fell 17% on Monday before beginning to rebound on Tuesday. The chipmaker was the world’s most valuable corporation in terms of market capitalisation. However, on Monday, it plummeted to third position behind Apple and Microsoft, with its market value falling to $2.9 trillion from $3.5 trillion, according to Forbes.

DeepSeek’s success is a significant boost for the Chinese government, which has been working to develop technology independent of the West. While the Communist Party has yet to respond, Chinese state media was quick to point out that Silicon Valley and Wall Street titans were “losing sleep” over DeepSeek, which was “overturning” the American stock market. “In China, DeepSeek’s advances are being celebrated as a testament to the country’s growing technological prowess and self-reliance,” says Marina Zhang, an associate professor at the University of Technology Sydney. “The company’s success is seen as a validation of China’s Innovation 2.0, a new era of homegrown technological leadership driven by a younger generation of entrepreneurs.” However, she cautioned that this mentality may lead to “tech isolationism”.

Author: Amar Chowdhury

Leave a Reply