The AI Bottleneck Trade Is Moving Beyond GPUs: $GEV $VRT $TT $ANET
One of the most interesting AI infrastructure themes right now has little to do with the GPUs themselves.
Elon Musk recently argued that roughly 15 GW of AI compute capacity built in 2027 may not be able to turn on that same year.
The reason is simple: having GPUs is not the same thing as having a functioning AI data center.
You still need transformers, switchgear, wiring, cooling systems, chillers and high-speed networking.
And those components are increasingly becoming the bottleneck. ⚡
🔌 Power Infrastructure
One of the broadest ways to play the electrification bottleneck, spanning turbines, grid equipment and switchgear.
The company booked about $2.4B of data-center electrification orders in Q1 2026, according to the figures cited here, already exceeding its full-year 2025 total.
Hitachi Energy is a major global transformer supplier, putting the company directly into one of the most constrained parts of the grid buildout.
A more concentrated switchgear play, with roughly $1.8B in backlog, including a data-center order worth more than $400M.
Prysmian
The world's largest cable maker is expanding across the infrastructure stack through acquisitions, including Encore Wire and Channell, while also pursuing opportunities in data-center optical connectivity.
This one needs a different lens.
Prysmian agreed to acquire Atkore for $95 per share in cash, so the current opportunity is more of a deal-spread situation than a pure AI infrastructure growth story. 👀
For European grid infrastructure, Nexans and NKT are also names worth watching, particularly as high-voltage interconnects have to be completed before new data centers can ultimately be energized.
💧 Liquid Cooling
AI racks are becoming dramatically more power-dense, making thermal management another critical bottleneck.
Probably the cleanest pure-play exposure to this theme.
Roughly three-quarters of Vertiv's revenue comes from data centers, while its Q1 2026 revenue reached $2.65B and adjusted EPS increased 83% YoY.
The major risk is customer concentration.
Eaton expanded its cooling exposure through the acquisition of Boyd Thermal, which closed in March 2026.
The deal adds significant revenue potential, although the $9.5B purchase price means valuation and integration deserve attention.
Another way to play power and thermal management inside AI data centers.
The company has a reported $2.6B backlog, with Nvidia among its customers, while roughly one-third of sales are tied to AI data centers.
The catch: higher raw-material costs are expected to weigh on this year's earnings.
❄️ Chillers
Cooling doesn't stop at the rack.
The facility itself needs increasingly sophisticated chiller infrastructure.
This may be one of the cleanest ways to express the data-center cooling thesis.
Trane has a reported $12.1B backlog, up 70% YoY, while applied bookings — including large engineered systems used in data centers — have grown sharply.
Johnson Controls is developing high-density cooling platforms and Silent-Aire CDUs for AI infrastructure, while also investing in direct-to-chip cooling technology.
A smaller-cap way to access the cooling theme.
The company is benefiting from demand for specialized HVAC and liquid-cooling infrastructure, although manufacturing scale remains an important constraint.
🌐 Networking
Once the power and cooling are solved, the GPUs still need to communicate.
That's where networking becomes another potential bottleneck.
Arista has become one of the leading data-center switching vendors and continues to benefit from AI cluster networking demand.
$Credo Technology Group Holding Ltd(CRDO)$
Credo sits closer to the high-speed interconnect layer, particularly copper connectivity inside AI systems.
Its recent growth has been extremely strong, although customer concentration remains something investors need to monitor.
Astera provides connectivity and semiconductor infrastructure inside AI racks.
The opportunity is substantial, but concentration is also significant: its three largest customers accounted for roughly 86% of 2025 revenue.
For investors looking for somewhat broader exposure, $AVGO, $MRVL and $COHR are additional names worth keeping on the radar.
And for Canadian investors, Celestica ($CLS) offers another angle on the AI infrastructure supply chain. 🇨🇦
🧠 The Bigger Picture
The important takeaway isn't that GPUs are becoming less important.
It's that GPU availability alone doesn't determine how quickly AI compute can actually come online.
The real bottleneck is increasingly the entire infrastructure chain:
Power → transformers → switchgear → cables → networking → liquid cooling → chillers → operational data center.
If AI capex continues accelerating, companies controlling these constrained components could capture a growing share of the spending wave.
That's why I'm paying increasingly close attention to the picks-and-shovels behind the AI boom, not just the companies designing the chips. 🔥
The next phase of the AI trade may be less about who has the most GPUs and more about who can actually power, cool and connect them. ⚡💧🌐
What do you guys think? Which bottleneck is likely to become the biggest constraint first — power, cooling or networking? 👇
Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.
