
DeepSeek vs Stargate Project: How AI spending is evolving, in charts
Summary
The massive AI infrastructure project announced by US President Donald Trump raised questions about overspending. The launch of an AI app by Chinese company DeepSeek has intensified them.Last month, US President Donald Trump announced the Stargate Project—a venture between OpenAI, SoftBank, Oracle and MGX that plans to invest up to $500 billion over four years to build new artificial intelligence (AI) infrastructure in the US. The project, which includes setting up several data centres, key to AI, aims to create jobs, boost the economy and secure American leadership in AI amid an ongoing race with China. Stargate is now caught in the crosshairs of a Chinese move.
The launch of an AI app by DeepSeek, a Chinese AI company matches the performance of its rivals from OpenAI and Anthropic, at a fraction of the cost. These companies have received financial support from tech majors, sitting on ample cash. DeepSeek said it spent only $5.6 million on computing power for its base model. User adoption was quick. DeepSeek replaced ChatGPT as the top app on the Apple app store in several countries, including the US, pushing tech investors in the US to reexamine their assumptions about the demand for computer chips. Tech stocks dropped last Monday, led by Nvidia, best known for its AI chips.
Read more: Mint Primer | Stargate Wars: Where is India in the AI infra race?
Even before DeepSeek, the Stargate project was questioned by Elon Musk, Tesla CEO and a key player in the Trump administration, on whether the project's backers, especially SoftBank, had secured funding for it. Now, there are broader questions about its viability due to its massive scale, uncertain returns and the rapidly evolving nature of the AI landscape.
The Cost Factor
The tech industry was bracing itself for high computing costs. As per the Stanford AI Index Report 2024, OpenAI spent an estimated $78 million worth of compute resources to train GPT-4, while Google spent $191 million for Gemini Ultra. Earlier, training models using datasets like SQuAD and GLUE cost around $160,000, the report said. Now, DeepSeek has cut costs by using fewer chips, optimising its training process and reconfiguring its server farms. It has passed on the benefits of such cost savings to users.
Also Read: Mint Primer | DeepSeek: A Chinese marvel or OpenAI copy?
DeepSeek has also open-sourced its model to an extent, allowing others to build on its work. As a result, future large language models (LLMs) could be built with smaller budgets. However, this does not imply supply will necessarily exceed demand. The tech industry has many examples confirming Jevon's paradox—when technology lowers costs or increases efficiency, usage expands. DeepSeek's breakthroughs might simply lead to greater usage, offsetting any potential reduction in the demand for chips and power.
Investment Boom
AI has been attracting huge strategic attention and corporate and venture investments ever since OpenAI launched ChatGPT in November 2022. Last year, Google said it was preparing to invest over $100 billion in AI technology development. Microsoft, which has been betting big on AI, including a partnership with OpenAI, said it is investing $80 billion this year on developing AI-enabled data centres. Similarly, Meta, which renewed its focus on AI, away from metaverse, said it will spend up to $65 billion this year to expand its AI infrastructure. Amazon has invested nearly $7 billion in Anthropic.
AI is also dominating the VC funding landscape. AI accounted for 44% of all investments in US VC-backed companies, according to an Ernst & Young report this month. It said the trend is expected to continue this year since the infrastructure for AI is still relatively immature. It's in this context that Stargate plans to make its $500 billion investment.
Bubble Question
When a huge investment is justified by investments that others have made in the segment, it could be a sign of a bubble. Last June, a Goldman Sachs report carried an interview with Jim Covello, its head of global equity research. He said, "Over-building things the world doesn’t have use for, or is not ready for, typically ends badly," pointing out that the Nasdaq declined around 70% between the highs of the dot-com boom and the founding of Uber. He expressed concerns about the risks of low adoption. Others have been skeptical about the returns on AI investments.
Also Read: DeepSeek’s breakthrough is a pivotal moment for the democratization of AI
However, some other measures still show sanity. A September report from Goldman Sachs said the price-to-earnings ratio of the largest technology companies today is roughly half that of the dominant companies at the peak of the technology bubble in 2000, a sign that tech sector is not in one now. If the investment by Stargate was a sign of its overspending, the launch of DeepSeek signalled a course correction in the sector.
www.howindialives.com is a database and search engine for public data