Deepseek and Destroy – By Ramsey Crookall
If, like me, you have had the pleasure of reading earnings report after earnings report over the past two years, you would have noticed a growing number of companies outlining plans to integrate Artificial Intelligence (AI) into their businesses. Satya Nadella, CEO of Microsoft, has described AI as “perhaps the most transformational technology of our time.” For example, within the healthcare sector, AI can enhance drug discovery and facilitate more efficient patient diagnoses.
To achieve this ‘transformational’ leap, the leading tech companies in the AI revolution are committing hundreds of billions of dollars not only to establish the necessary infrastructure capable of handling the vast amounts of data required but also to invest similar sums in developing the Large Language Models (LLMs) that serve as the brain of AI. In simple terms, an LLM is a machine-learning model that can process, understand, and generate human language. Probably the most renowned of these companies is OpenAI, whose ChatGPT programme, upon its release in November 2022, helped to ignite the current tech market bull run. Other notable names in the tech sector currently involved in developing their own AI models include Google with PaLM 2, Meta with LLaMA 2, Anthropic, and Microsoft, to name a few.
The development of these LLMs has primarily been confined to the mega-cap US tech firms that can afford to spend hundreds of millions of Dollars. In the case of the latest ChatGPT-4 model, OpenAI has invested upwards of $100 million in pursuit of artificial perfection.
The excitement surrounding AI has propelled a select number of companies to astronomical valuations and solidified the US tech sector as the prime destination for investors looking to make profits. The top five companies in the S&P 500, all of which are expanding their AI capabilities, have surfed the wave of enthusiasm and account for 25% of the index.
However, the dominance of US tech giants may be under threat, with repercussions possibly being felt more broadly in the global AI landscape. Stocks such as TSMC in Taiwan and the Dutch-based ASML are affected. Towards the end of January, a challenger to ChatGPT emerged in the form of the Chinese-based DeepSeek. While its American counterparts are sinking hundreds of millions of pounds into building models and tens of billions into the required hardware, the much smaller Chinese startup claims to have developed its latest version, DeepSeek V3, for a relatively modest sum of $5.5 million. Reports suggest that the DeepSeek V3 model has outperformed its competitors, including Meta, OpenAI, and Alibaba, across various third-party benchmarks. In a further blow to Silicon Valley, DeepSeek is open-source, allowing anyone to use and modify it freely.
In an ironic twist to the US government restrictions on advanced semiconductor sales to China, the development of this reportedly more efficient and cheaper alternative has arisen out of the necessity to utilise older generation chipsets. Nvidia has been one of the biggest beneficiaries of the AI boom, with its market cap rising from $145 billion in 2020 to nearly $3 trillion today. The company has increased revenues to nearly $61 billion annually, driven by the development and supply of the most advanced GPUs to power LLMs. However, the promise of greater power may now be unnecessary if DeepSeek can fulfil the recent claims.
As you would expect, the impact on US companies hasn’t been for the faint-hearted – in today’s market, Nvidia’s share price has dropped by over 16%, amounting to nearly $600 billion in lost market cap. Questions will arise regarding the necessity of spending vast sums of money on the latest technology when an older generation of chipsets can perform similarly. Microsoft, Meta, and Alphabet have all experienced sharp declines in the market; whether this will develop into a sustained sell-off and capital reallocation from US tech remains to be seen.