![\"artificial](\"https:\/\/www.europesays.com\/uk\/wp-content\/uploads\/2025\/04\/AI-1200x675.jpg\")
<\/p>\n<\/p>\n
Image source: Getty Images.<\/p>\n
Since the Trump administration paused the introduction of tariffs on most countries for 90 days and subsequently exempted semiconductor equipment from tariffs on Chinese products, Nvidia<\/strong> (NASDAQ: NVDA<\/a>) stock has soared<\/a> 17%. But I remain to be convinced that now is the right time to buy.<\/p>\n DeepSeek<\/p>\n Despite surging over the past few days, the stock still trades below the carnage following the release of DeepSeek. Remember that? In one trading day, $600bn was wiped off the value of the company. That\u2019s twice the market cap of AstraZeneca<\/strong>, the FTSE 100<\/strong>\u2018s largest constituent.<\/p>\n The reason why I pose it like that is because it seems to me that event has been airbrushed out of history by many. The mantra of buy the dip is so deeply rooted in the psyche of investors that they couldn\u2019t resist running headlong into buying. No thought was given to even trying to understand the consequences of a new way of building a large language model (LLM).<\/p>\n Data centre growth<\/p>\n Early this year, the Wall Street Journal reported that Microsoft<\/strong>\u2018s total capital expenditure in 2025 would top $90bn. However, only last month, analysts at TD<\/strong> Cowen reported that the company had decided to cancel several leases for new data centres in the US, totalling a couple of hundred MWs.<\/p>\n Its report suggested that Microsoft used facility and power delays as a justification for cancelling. They likened such a tactic to one employed by Meta<\/strong> in 2022 following its cancellation of multiple data centre leases related to the metaverse.<\/p>\n Microsoft did respond to these claims with a rather bland statement, but which didn\u2019t deny the cancellations.<\/p>\n Whatever the truth here, a company doesn\u2019t cancel leases for no good reason. Building data centres is like building a mine, it doesn\u2019t happen overnight. It takes years of careful planning. Could it be that Microsoft doesn\u2019t believe that the capacity is there to the extent that it was just a few months ago?<\/p>\n Major breakthrough<\/p>\n The hyperscalers have collectively spent in excess of $100bn building out their LLMs. But no killer app has yet emerged. At the moment, no path to profitability and a return on their investments are in sight. The more time that passes without any significant breakthrough, the more inpatient investors will be become.<\/p>\n I would postulate that most investors don\u2019t really understand the underlying technology. They certainly don\u2019t care about the future. What they do care about is what moves their portfolios higher. And one thing momentum investors certainly don\u2019t have is patience.<\/p>\n I don\u2019t dispute that LLMs can do some very interesting things. But will it ever be a multi-trillion dollar opportunity? I doubt whether this innovation will be on the same scale as the internet.<\/p>\n To me, the technology is nothing more than an evolutionary step in the AI space, and a small one at that. Many will argue that Nvidia is a lot more than generative AI. That is certainly true. But it\u2019s foolish to believe that its fate is not inextricably linked to the continued mass investment by the hyperscalers.<\/p>\n