According to JPMorgan analyst Harlan Sur's interpretation of the latest earnings reports from Microsoft and Meta Platforms, Inc., the intensity of spending on AI infrastructure has clearly entered a new expansion cycle.
Both tech giants indicated in their latest reports that the tight supply of AI computing power is expected to persist through 2026. Their fourth-quarter capital expenditures surpassed market expectations: Microsoft at $37.5 billion and Meta at $22.137 billion. Furthermore, Meta raised its full-year 2026 capital expenditure forecast to $125 billion, a 73% year-over-year increase.
This suggests that, driven by the accelerated deployment of foundational models, AI agents, and commercial applications, the demand for computing power continues to outpace supply capacity, prompting cloud computing and hyperscale enterprises to ramp up investments. JPMorgan believes that under the current supply-demand dynamics, major tech companies' capital expenditures still have room for upward revision, with investments primarily focused on data centers, servers, and network infrastructure, thereby boosting the performance of the related semiconductor industry chain. This investment trend is expected to continue into 2027.
Supply constraints have become the norm: The demand gap will persist. Supply tightness has become a core bottleneck in current AI infrastructure development. Both Microsoft and Meta highlighted in their latest earnings calls that the demand for computing power continues to exceed supply capacity.
The structural reason for this supply-demand imbalance lies in the accelerated deployment of foundational models, AI agents, and commercial applications, which is driving an exponential increase in computational intensity. Meta revealed that the scale of its GPU clusters used for training generative advertising models has already doubled and is being further expanded to support the training of its next-generation GEM models in 2026.
This persistent supply tightness is expected to continue supporting high-intensity investment in data centers, servers, and network infrastructure between 2026 and 2027.
Custom chip development has become a strategic priority. In addition to continuing to procure GPUs from AMD and NVIDIA, Microsoft and Meta are accelerating their strategies for custom ASIC chips to improve energy efficiency and expand application scenarios.
For Meta, its in-house chip project, MTIA, is undergoing continuous iterations and currently supports inference for its search engine, with plans to extend to core ranking, recommendation training, and inference workloads by the first quarter of 2026. JPMorgan points out that Meta's chip design partner, Broadcom, will benefit from this, with revenue from Meta expected to see a significant jump in 2026.
Microsoft is focusing on optimizing the energy efficiency and economics of token processing, emphasizing the critical role of custom chips. According to JPMorgan analysis, although Marvell Technology was not involved in Microsoft's previously announced MAIA 200 chip, it is progressing on schedule with the development of the next-generation MAIA 300 chip, expected to enter mass production in the second half of 2026.
As model scale and complexity increase, the demand for computing power grows exponentially. The capital expenditure guidance from cloud service giants remains focused on AI infrastructure investment. The latest plans from Meta and Microsoft further corroborate JPMorgan's assessment: investments related to networking, custom chips (ASICs), and GPUs for accelerating computing and storage will maintain strong momentum in the medium to long term.
Comments