AI computing power delivery bottleneck emerges? Microsoft cuts $5 billion in capital expenditure due to data center delays, Supermicro raises revenue forecast.

BlockBeats news, November 7 - According to The Information, Microsoft's capital expenditure in the third quarter was reduced by approximately $5 billion compared to previous plans due to delays in the delivery of third-party data centers it planned to lease. Sources familiar with the matter said that Iren's data center project in Childress, Texas, USA, experienced delays, causing Microsoft to postpone some expenditures. Microsoft had previously announced plans to invest $9.7 billion over five years to lease Iren's servers equipped with Nvidia GPUs.

Power supply and grid access have also become bottlenecks for computing power delivery. Last month, Microsoft withdrew from a planned UK data center project leased from Nscale due to grid access delays. As self-built data centers struggle to meet demand in a timely manner, Microsoft is simultaneously leasing computing power from multiple cloud service providers.

Meanwhile, AI server supplier Super Micro Computer (SMCI) continues to raise its performance expectations. The company's first-quarter revenue reached $19.95 billion, exceeding analysts' expectations of $16.39 billion, and it expects second-quarter revenue of $19.5 billion to $21.5 billion. Its fiscal year 2027 revenue forecast was also raised from $65 billion to $72 billion to $72 billion to $82 billion.

However, growth in server sales does not mean computing power has actually been put into use. Bernstein pointed out that assessing AI computing power demand cannot simply rely on contract amounts or megawatt scale; it is also necessary to consider contract cancellability, duration, and actual deployment timing. Some data center projects may be delivered in phases spanning months or even years; in addition, chip procurement agreements and data center lease contracts may correspond to the same batch of computing power, and if counted repeatedly, this could overestimate actual new computing power demand.

This means that the bottleneck in the AI industry chain is extending from chip supply to data center construction, power supply, and computing power deployment, and the time gap between server order growth and final computing power delivery deserves attention.