The need to compute is likely to grow exponentially in the coming years and support greater artificial intelligence (AI) monetization opportunities, explains analyst Angelo Zino in CFRA Research's flagship newsletter, The Outlook.
Although LLMs (large language models) come in all shapes and sizes, the largest models are expected to continue to push the needle toward greater compute. These include the likes of OpenAI GPT, Gemini, Claude, and Llama, which should drive higher investments from the major hyperscalers well into 2026 and beyond.
Ultimately, greater compute will allow developers to create better AI agents, as well as other tools that could help the broader technology space better monetize upcoming AI investments.
Lack of monetization won’t keep hyperscalers from investing in AI; rather, healthy cloud and digital ad spend should keep them spending. In our view, as long as the biggest drivers of top-line growth for the hyperscalers continue to remain healthy — higher cloud and digital ad spend — then investments toward greater compute and AI will stay elevated.
We have many reasons to believe this, but note that in an environment where mega caps can’t look at M&A, they will spend aggressively on growth and return more cash back to shareholders.
While macroeconomic factors are a risk for any company, hyperscalers are more inclined to adjust AI investments during tougher times than cut them as the focus remains on reaching artificial general intelligence to open new AI monetization opportunities.
Hyperscalers represent the biggest and most important component of Data Center spend. We estimate that about 40% to 45% of investments tied to greater compute will come from hyperscalers in the next three years.
Not to be forgotten, spending tied to AI is also growing among Tier 2 Cloud providers as well as Enterprise customers. Sovereign AI, although small, was a non-existent market a year ago and has high growth potential attached to it in the coming years.
We think all categories are poised to increase spending in the next two to three years, with hyperscalers increasing spend by nearly 40% this year and another 15% to 20% in 2025. We also think that hyperscalers are best positioned to sustain elevated spend among the four categories should macroeconomic conditions sharply deteriorate.
Looking at our top compute-exposed companies, Marvell Technology (MRVL) is our top pick. Our Strong Buy rating on Marvel reflects our outlook for improving trends in cyclical markets and AI opportunities (+20% long-term EPS growth rate).
The primary reason that MRVL is our top pick is that the relative performance based on a cyclical trough coupled with the AI opportunity has the opportunity to allow MRVL to outgrow its peers through CY 27 and drive upside to consensus expectations.
Our Buy rating on NVIDIA (NVDA) reflects our view of NVDA’s expanding total addressable market (TAM), driven by content gains from higher system sales, penetration into edge devices, and software opportunities. NVDA’s greater software capabilities across the data center stack and plans to accelerate the cadence of new chip designs should support its massive competitive moat.
Our Buy rating on Broadcom (AVGO) reflects our view that it will be a major winner from an AI infrastructure boom, driven by its ASIC and networking/switcher businesses, as well as VMware cost synergies. We believe the VMware integration is progressing ahead of schedule, given sharply lower spending, streamlining SKUs, and good traction migrating customers to subscriptions.
Our Buy on Advanced Micro Devices (AMD) rating reflects our growth outlook for AMD’s CPU data center servers from the ramp of its next- generation EPYC processor, prospects for GPU offerings, and our expectation for balance sheet improvement. AMD’s AI opportunities stem much further than the cloud market given its wide variety of offerings, with its TAM expansion supporting growth for years.
Several risks could impact our outlook on the semiconductor industry. This includes lower-than-expected data center spending should negative economic trends cause us to ratchet down our capital spending forecast for certain providers.
A slower-than-expected ramp in AI adoption by enterprise companies could also negatively weigh on our assumptions should the timetable for certain investments by these companies be pushed out by an extended period of time. In addition, should a recovery in more cyclical end markets occur at a slower pace than we anticipate, that could cause multiples to compress and shares to underperform.
