Reporting quarterly earnings Wednesday afternoon, it will be tough for Microsoft to avoid the issue of the AI breakthroughs from China’s DeepSeek that are shaking up the tech world and the stock market this week.
DeepSeek’s dramatic cost efficiencies and novel AI training techniques are prompting new scrutiny of the record $80 billion in capital expenditures that Microsoft has projected to build out its AI infrastructure this fiscal year.
The buzz over DeepSeek is so strong that the company’s response may rival its actual results in Wall Street’s assessment of the stock after its earnings report. For the record, analysts expect revenue of $68.9 billion for the December quarter, up 11%; and earnings of 3.11 per share, up 6%.
Regarding DeepSeek, look for Microsoft to make the case that these developments are in line with where it expected things to go, if much faster than it thought they would happen.
Microsoft has been saying since last year that large language models were becoming a commodity, with the real differentiation to come from how LLMs are applied.
“This means companies should focus more on how they integrate these models with their own data and workflows, rather than seeing the models themselves as a unique competitive advantage,” wrote Jared Spataro, chief marketing officer for Microsoft’s AI @ Work initiatives, in a post foreshadowing the trend in October.
In that context, the company is positioning its infrastructure buildout as critical not just for training AI models but also to allow its business customers to take full advantage of them.
Microsoft CEO Satya Nadella previewed the company’s response on Monday, posting on Twitter and LinkedIn that it was an example of Jevons Paradox, in which cost efficiencies lead to increased in overall demand.
“As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of,” Nadella wrote.
Further supporting this position is Microsoft’s decision to partner on technology but not contribute financially to the $500 billion Stargate project from OpenAI, Oracle, and Softbank — despite its role as OpenAI’s partner and investor.
“By electing not to fund OpenAI’s ambitious GPU compute targets, and by being the most vocal CEO about ‘LLMs commoditizing,’ Microsoft’s CEO has been signaling a desire to pivot away from material training GPU commitments to a single [large language model] and instead to scale an inference infrastructure for large enterprise customers,” wrote UBS analyst Karl Kierstead, as quoted by TheStreet.
After dropping Monday along with other tech stocks, Microsoft shares recovered Tuesday, closing up 2.87% on the day at more than $447/share, about $4/share higher than the closing price last week.
