Goldman Sachs Sees $1T AI Spend, 3 Stocks to Buy

Goldman Sachs says AI infrastructure spending could top $1 trillion in 2027.

AI infrastructure spending is already expected to pass $700 billion this year, and Goldman Sachs says it could climb to between $920 billion and $1.4 trillion in 2027. That puts the AI buildout in the same conversation as some of history’s biggest capital projects, while still leaving room for more spending if demand keeps rising.

For investors, the question is simple: which companies actually capture that money? In The Motley Fool piece by Geoffrey Seiler, the answer centers on Nvidia, AMD, and Micron.

AI spend is moving from hype to hard budgets

The scale of the spending matters because it is no longer just about a few model launches or headline-grabbing product demos. Data centers need GPUs, CPUs, memory, networking gear, cooling, and a lot of power, and each layer creates a different profit pool.

Goldman Sachs’ midpoint estimate of $1.25 trillion in 2027 would equal about 3% of GDP, which is a staggering figure even before you compare it with older infrastructure booms. The Motley Fool article notes that this is still below some past technology buildouts, including the U.S. and U.K. railroad expansions in the late 1800s.

That matters because infrastructure cycles tend to reward the companies selling the picks and shovels, not just the companies building the final product. In this round, the beneficiaries are spread across compute, memory, and the software stacks that make the hardware useful.

• Current AI infrastructure capex is already above $700 billion.
• Goldman Sachs expects 2027 spending between $920 billion and $1.4 trillion.
• The midpoint estimate implies AI capex could equal about 3% of GDP.
• Demand is spreading across training, inference, and agentic AI.

Nvidia still owns the center of the market

Nvidia remains the biggest direct beneficiary of AI infrastructure spending. The company’s GPUs are still the default choice for training large models, and that gives it pricing power, scale, and a huge installed base of developers who already know the CUDA software stack.

Geoffrey Seiler points out that Nvidia posted 85% revenue growth in the latest quarter, which is the kind of number that tells you demand is still running hot. The stock also trades at a forward price-to-earnings ratio of 16 times fiscal 2028 estimates, which is lower than the valuation many investors would expect for a company tied to the biggest spending wave in tech.

Nvidia is also trying to expand beyond training. Its push into inference and agentic AI matters because those workloads will need different hardware mixes, including ARM-based CPUs and processors aimed at the decode phase of inference. That gives Nvidia more ways to stay embedded in the stack even if the market shifts away from pure training demand.

“The company has been seeing extraordinary revenue growth, including 85% last quarter,” Geoffrey Seiler wrote in The Motley Fool.

• Nvidia’s latest quarterly revenue growth: 85%
• Forward P/E: 16x fiscal 2028 estimates
• Primary strength: AI training GPUs
• Expansion area: inference and agentic AI hardware

AMD and Micron benefit from the next phase

AMD did not beat Nvidia in model training, but the article makes a strong case that it is better placed for inference and agentic AI. That is partly a software story, since AMD’s ROCm stack has improved enough to loosen Nvidia’s CUDA grip in some areas, and partly a hardware story, because AMD’s chiplet GPU design can pack in more memory.

Memory matters more when workloads become inference-heavy. As AI systems move from training giant models to answering questions and running agents, the bottleneck shifts toward memory bandwidth and system balance. AMD also has two $100 billion GPU commitments tied to inference, which is a very large demand signal for a company still trying to gain share.

Micron gets a different kind of tailwind. It sells DRAM and high-bandwidth memory, both of which are essential for modern AI chips. The article says Micron has been posting record revenue, gross margins, and profits as DRAM prices rise, and its forward P/E of 9 times fiscal 2027 estimates makes it look far cheaper than the most popular AI chip names.

• AMD has two $100 billion GPU commitments tied to inference
• AMD sees the GPU-to-CPU ratio shifting from 8:1 in training to 1:1 in agentic AI
• AMD says that CPU opportunity could become a $120 billion market
• Micron trades at 9x fiscal 2027 earnings estimates
• Micron benefits from DRAM and HBM demand

What investors should watch next

The big takeaway is that AI infrastructure spending is broadening, and that usually helps different companies at different points in the cycle. Nvidia is still the clearest winner in training, AMD has a plausible path in inference and agentic AI, and Micron benefits from the memory shortage that comes with both.

If Goldman Sachs is even close to right, the next year will be less about whether AI spending continues and more about which bottlenecks get relieved first. That means investors should watch GPU supply, memory pricing, and whether inference workloads grow fast enough to change buying patterns across data centers.

For readers tracking the broader AI trade, the smarter question may be which part of the stack is still underpriced. Right now, Goldman’s numbers suggest the answer may not be the most obvious mega-cap name, but the suppliers that keep every new model and agent running.

One thing is worth watching closely: if 2027 AI infrastructure spending lands near $1.25 trillion, the market will probably stop treating this as a short cycle and start pricing it as a multi-year industrial buildout.