The last two weeks were critical for the future of the AI trade.
Google, Amazon and Microsoft all said their cloud businesses are capacity-constrained.
AI demand is well and alive.
Here are my top 10 AI infrastructure stock picks: 🧵
1. $Advanced Micro Devices(AMD)$
5-year Revenue CAGR: 27%
Gross Margin: 53%
Forward P/E: 22
It's second largest provider of data center GPUs after $NVDA.
Though its hardware was comparable to that of $NVIDIA(NVDA)$ , it was way behind in software stack.
Now it's catching up with ROCm updates. Its MI300x chips are already outperforming Nvidia H200 in inference performance.
The demand will explode for its next-gen chips.
Image
3-year Revenue CAGR: 194%
Gross Margin: 37%
Forward P/E: N/A
Nebius is an emerging dark horse in AI infrastructure.
It's building vertically integrated AI-native cloud platforms—from custom data centers to ML tooling.
Unlike hyperscalers, it’s designed from the ground up for AI workloads. Its focus on cost efficiency and control makes it a potential disruptor, especially in markets wary of US cloud dominance.
Demand is exploding as model providers try to diversify away from hyperscalers.
Image
5-year Revenue CAGR: 34%
Gross Margin: 64%
Forward P/E: 31
Arista is the networking backbone of AI data centers.
As AI workloads explode, so does demand for high-speed, low-latency Ethernet switches. $ANET dominates this space, especially among hyperscalers.
Its close ties to $Meta Platforms, Inc.(META)$ and $Microsoft(MSFT)$ , plus custom silicon and EOS software, give it both scale and stickiness.
A pure-play on AI-driven data traffic.
Image
4. $Taiwan Semiconductor Manufacturing(TSM)$
5-year Revenue CAGR: 17%
Gross Margin: 57%
Forward P/E: 15
Monopoly in cutting edge chip manufacturing.
Its 3nm process yields are still 20-25% higher than Samsung while its 2nm yields are 40% higher.
It's also the only foundry having started the development of 1nm process.
No matter who designs the best chips, TSM will be the manufacturer.
Image
5-year Revenue CAGR: 17%
Gross Margin: 52%
Forward P/E: 25
Monopoly in manufacturing EUVL machines that are indispensable for the cutting edge chip manufacturing.
As chips become more and more complex, share of EUVL machines in all revenue will increase, boosting the revenue and margins.
Attractively priced at 25 times forward earnings given its monopoly position.
Image
5-year Revenue CAGR: 67%
Gross Margin: 75%
Forward P/E: 25
It's dominating data center GPUs with over 90% market share.
Though it'll eventually lose some market share to AMD, the growing market will make up for the loss and revenue will keep climbing fast.
Attractively valued at 25 times forward earnings.
Image
5-year Revenue CAGR: 49%
Gross Margin: 50%
Forward P/E: 9
ACMR makes advanced cleaning tools for semiconductor fabs—critical for high-end AI chips.
As node sizes shrink and chip complexity grows, wafer cleaning becomes more essential than ever.
ACMR is gaining market share and revenues are set to explode as China is pressing for self-sufficiency in advanced chip manufacturing.
Extremely undervalued at 9 times earnings.
Image
3-year Revenue CAGR: 122%
Gross Margin: 76%
Forward P/E: 56
Astera Labs makes connectivity chips that solve one of AI’s biggest bottlenecks: moving data fast between GPUs, CPUs, and memory.
Its PCIe/CXL solutions are mission-critical for AI clusters—especially as systems scale to thousands of GPUs.
Demand is exploding as cluster sizes keep growing but the valuation is quite high, yet it still deserves to be on your watchlist.
Image
5-year Revenue CAGR: 4%
Gross Margin: 34%
Forward P/E: 9
Micron is a top supplier of DRAM and NAND—both are essential for training and running AI models.
AI GPUs are memory-hungry, and $MU’s high-bandwidth memory (HBM) is finally gaining traction.
Its HBM3E chips are now being validated by NVDA for use in next-gen AI accelerators. As AI model sizes balloon, so does demand for memory.
Its product is open to commoditization but the valuation is so cheap that it may be worth an investment.
Image
10. $Marvell Technology(MRVL)$
5-year Revenue CAGR: 18%
Gross Margin: 47%
Forward P/E: 22
Marvell builds the custom chips that move and process data inside AI data centers—think optical DSPs, networking gear, and custom accelerators.
It’s deeply embedded in hyperscaler infrastructure and has partnerships with AMZN, MSFT, and META.
It'll benefit from hyperscalers' shift to custom silicons as they don't want to depend on external suppliers forever.
Image
Open a CBA today and enjoy privileges of up to SGD 20,000 in trading limit with 0 commission. Trade SG, HK, US stocks as well as ETFs unlimitedly!
For whom haven't open CBA can know more from below:
Find out more here：
Comments