#AI股持续上涨还有哪些投资机会
AI has been up for so long—what can you still buy next?
Everyone is currently competing for AI compute power chips, but simply looking at compute power makes it hard to capture excess returns anymore. The market logic is changing: this year, tech giants’ capital expenditures are still surging—these next few categories might be the new main drivers.
Electricity and water are the real bottlenecks for compute power.
AI data centers are power-guzzling beasts. In both Silicon Valley and China, people are aggressively upgrading the power grid. High-voltage power supplies, transformers, and even power infrastructure companies are seeing performance get realized faster than many software companies that are just waving the “AI” banner.
And then there’s liquid cooling.
Once chip power consumption breaks 1000W, air cooling simply can’t handle it. Rigid demand is emerging for components like CDU liquid-cooling distribution units, quick connectors, and cold plates. This year, penetration is rapidly jumping from 20% to 40%, which looks like a hard, certain growth cycle.
The replacement wave at the edge device layer.
Everyone is waiting for AI terminals to roll out—don’t just look at phone or PC brand names. Check the upgraded parts inside: high-frequency high-speed PCB boards, high-capacity batteries, thermal modules, and next-level HBM memory. If terminals are going to run large local models, all of these hardware components must be upgraded to the next generation.
From “buying tools” to Agents that can actually get work done.
On the software layer, the industry is speeding up its divergence. Companies that used to sell only API interfaces find it hard to make money. But AI Agent applications that can truly embed into specific scenarios—like healthcare, finance, or office workflows—and help businesses save labor costs are starting to see profit leverage from the second half of this year. The ones that plug into real business processes are the ones that will survive.
Next, “air stocks” that rely purely on hype will quickly face a major reshuffle. The market will transition from frenziedly building infrastructure to hard tests of performance and return on earnings. Compute-power stocks will churn in a high-range consolidation, and capital will spread toward energy and power, liquid-cooling key components, and downstream applications that have real profits after recurring non-deductible items (扣非利润). When picking, focus more on orders and cash flow—don’t just listen to the story.
DYOR
$NVDA $NVDAB
AI has been up for so long—what can you still buy next?
Everyone is currently competing for AI compute power chips, but simply looking at compute power makes it hard to capture excess returns anymore. The market logic is changing: this year, tech giants’ capital expenditures are still surging—these next few categories might be the new main drivers.
Electricity and water are the real bottlenecks for compute power.
AI data centers are power-guzzling beasts. In both Silicon Valley and China, people are aggressively upgrading the power grid. High-voltage power supplies, transformers, and even power infrastructure companies are seeing performance get realized faster than many software companies that are just waving the “AI” banner.
And then there’s liquid cooling.
Once chip power consumption breaks 1000W, air cooling simply can’t handle it. Rigid demand is emerging for components like CDU liquid-cooling distribution units, quick connectors, and cold plates. This year, penetration is rapidly jumping from 20% to 40%, which looks like a hard, certain growth cycle.
The replacement wave at the edge device layer.
Everyone is waiting for AI terminals to roll out—don’t just look at phone or PC brand names. Check the upgraded parts inside: high-frequency high-speed PCB boards, high-capacity batteries, thermal modules, and next-level HBM memory. If terminals are going to run large local models, all of these hardware components must be upgraded to the next generation.
From “buying tools” to Agents that can actually get work done.
On the software layer, the industry is speeding up its divergence. Companies that used to sell only API interfaces find it hard to make money. But AI Agent applications that can truly embed into specific scenarios—like healthcare, finance, or office workflows—and help businesses save labor costs are starting to see profit leverage from the second half of this year. The ones that plug into real business processes are the ones that will survive.
Next, “air stocks” that rely purely on hype will quickly face a major reshuffle. The market will transition from frenziedly building infrastructure to hard tests of performance and return on earnings. Compute-power stocks will churn in a high-range consolidation, and capital will spread toward energy and power, liquid-cooling key components, and downstream applications that have real profits after recurring non-deductible items (扣非利润). When picking, focus more on orders and cash flow—don’t just listen to the story.
DYOR
$NVDA $NVDAB
