Worth separating the two capex stories inside AI.
The compute buildout is corporate and discretionary. It is funded off hyperscaler balance sheets and re-rates on every model release, which is why the multiples there swing.
The grid buildout is a different animal. Most of this $1.1 trillion is regulated, rate-based utility capital, underwritten by reliability mandates and load growth that does not pause for the cycle. It is visible years out and close to non-discretionary.
That changes the quality of the backlog sitting under the equipment makers. Much of the market still prices them as late-cycle industrials. The order books look more like multi-year secular annuities.
The re-rating from the first to the second is where the work is.
The detail everyone is skipping: this is equity, not debt.
The 2001 telecom bust was lethal because the buildout was levered. Vendor financing, junk bonds, balance sheets that couldn't take a miss. Alphabet is funding compute with stock sales to the strongest balance sheet on earth. If returns disappoint, the loss sits where it can be absorbed.
That's the difference between a capex cycle and a credit cycle. So far, AI is the former.
Berkshire Hathaway just anchored Alphabet's roughly $80 billion equity raise with a $10 billion check. Greg Abel signed off on a weekend call, at a 5.5 to 6.5% discount to market.
Two signals matter more than the headline:
1. Hyperscaler AI capex has outgrown operating cash flow. When Alphabet sells stock to fund compute, the buildout has entered a new financing regime: equity raises, private placements, anchor investors. This is how railroads and telecom networks got built.
2. The most conservative balance sheet in America is now funding it. Buffett avoided tech for decades because he couldn't underwrite the winners. Abel just made Alphabet a top-4 Berkshire holding, rivaling Coca-Cola. With nearly $400 billion in cash to deploy, this is where he chose to put it.
When value capital starts financing the AI buildout on negotiated terms, it stops being a momentum trade and becomes a capital cycle.
Follow the capex. $AIS $POW $OMAH
https://t.co/HHnhNKfHql
$OMAH ETF holds Berkshire’s top names while seeking consistent income through options.
Learn More: https://t.co/fYfhe1nFti
Investing involves risk - Loss of Principal is Possible
Distributed by Foresides Fund Services
Worth separating the two capex stories inside AI.
The compute buildout is corporate and discretionary. It is funded off hyperscaler balance sheets and re-rates on every model release, which is why the multiples there swing.
The grid buildout is a different animal. Most of this $1.1 trillion is regulated, rate-based utility capital, underwritten by reliability mandates and load growth that does not pause for the cycle. It is visible years out and close to non-discretionary.
That changes the quality of the backlog sitting under the equipment makers. Much of the market still prices them as late-cycle industrials. The order books look more like multi-year secular annuities.
The re-rating from the first to the second is where the work is.
A decade of grid spending. Now in five years.
U.S. utilities will invest more than $1.1 trillion upgrading the grid through 2029, nearly double the annual pace of the entire prior decade.
That capital does not accrue to the AI models everyone is watching. It accrues to the companies that make the transformers, switchgear, and substations the buildout cannot happen without.
The opportunity was never the demand. Everyone sees the demand. It is owning the picks and shovels that collect the check, before the rest of the market does.
This is the supercycle we have been mapping for $POW at @vistasharesX.
AI runs on power and the grid is racing to keep up. $POW invests in the infrastructure enabling datacenters, energy storage, and smart distribution networks that power the AI economy.
Learn More: https://t.co/8jIqNbtlfL
#electrification
Investing involves risk. Principal loss is possible. Distributed by Foreside Fund Services, LLC.
The part of this most investors are getting backwards:
AI has two halves. The compute half is brilliant and deflating. A better model or a cheaper chip can erase a pricing assumption overnight, and the whole complex re-rates on every headline.
The grid half is boring and inflating. Multi-year backlogs, contracted pricing, and supply that physically cannot ramp on the same timeline as demand.
The market is paying the higher multiple for the volatile, deflating half, and the lower multiple for the durable, inflating one.
That gap is the opportunity. Own the cash flows that get more certain as the buildout scales, not less.
The US is trying to win the AI race on a power grid it can no longer build for itself.
Roughly 80 percent of its large power transformers are imported. The wait for a new one has gone from six weeks to as long as three years.
The constraint on AI is no longer the chip. It is the hardware that moves electricity, and a short list of companies controls it.
Everyone is racing to own the intelligence. Far fewer own what it physically runs on.
Which side do you think has more pricing power over the next decade? $POW physical AI infrastructure or $AIS AI intelligence infrastructure?
Everyone’s modeling the demand curve. The harder question is the supply response, and that’s measured in interconnection queues, not quarters.
Note the y-axis is electricity, not capex. The dollars required to actually deliver these electrons are a multiple of the demand story itself.
The second-order trade isn’t the power. It’s everything that gets squeezed when AI outbids every other buyer for the same grid.
IEA data center electricity demand is set to rise from ~290 TWh today to ~1,200 TWh by 2035, a ~4x increase, with AI-optimized servers driving the majority of incremental load.
The investable implication: power generation, grid infrastructure, and cooling become the binding constraint on AI buildout, not chips. We view this as the early innings of a multi-decade #electrification capex cycle.
⚡ International Energy Agency
$AIS Outperforming $AIQ, $CHAT, $ARTY
Most AI ETFs own the same names.
$AIS doesn't - it focuses on the infrastructure companies powering AI growth - the "picks & shovels" behind the global AI value chain.
That difference is showing up clearly:
Notice the pattern in every AI infrastructure debate.
It was packaging. Then it was memory. Then it was power. Each time, the consensus "real bottleneck" was already changing by the time it became consensus.
The edge was never picking the chokepoint. It was seeing that there is always a chokepoint, the spend routes through it, and a short list of toll takers get paid on every reroute.
You don't have to call the next constraint. You have to own the road it runs on.
Everyone is hunting for the next bottleneck in AI.
Two years ago it was advanced packaging. $TSM CoWoS was sold out and the whole buildout waited on it.
Then it was memory. HBM became the tightest component in the stack, sold out through 2026.
Now it's power. You can get an AI chip faster than you can get a transformer, which runs three to four years out.
The bottleneck keeps moving. The capex keeps compounding. Stop guessing the chokepoint and follow the spend. $AIS & $POW follow this strategy.
What layer do you think binds next?
A year ago, Berkshire owned zero Alphabet.
This weekend it added another $10B, wired straight into a private placement funding Google's AI buildout.
Alphabet's 2026 capex guide: $180B to 190B.
Buffett avoided Big Tech for 60 years. Berkshire just became a direct financier of the AI infrastructure cycle.
What does the most disciplined allocator alive see here? $GOOGL $BRK.B $OMAH
$OMAH ETF holds Berkshire Hathaway's $BRK.B top names while seeking consistent monthly income through options.
Learn More: https://t.co/gRPafMl8If
Investing involves risk. Principal loss is possible. There is no assurance the Fund will achieve its investment objective or make distributions in any given month.
Distributed by Foreside Fund Services, LLC.
The AI capex bear case has always been one question: does the spend ever earn its return?
The most return-obsessed allocator alive just answered it with $10B.
Berkshire doesn't fund growth stories. It funds returns on capital. When that balance sheet underwrites your buildout, the ROIC question stops being open.
That's the tell.
A year ago, Berkshire owned zero Alphabet.
This weekend it added another $10B, wired straight into a private placement funding Google's AI buildout.
Alphabet's 2026 capex guide: $180B to 190B.
Buffett avoided Big Tech for 60 years. Berkshire just became a direct financier of the AI infrastructure cycle.
What does the most disciplined allocator alive see here? $GOOGL $BRK.B $OMAH
For two years, AI was a chip story. It just became a power story.
US data center electricity demand goes from ~62 GW today to 134 GW by 2030. By then data centers draw close to 9% of all US power, up from 4%.
The race was never just silicon. It's the grid behind it.
What caps the AI buildout first: chips, or megawatts?
AI demands massive new energy capacity. $POW invests in the global companies building the backbone of this next-generation grid, enabling data centers, energy storage, and smart distribution networks.
Learn More: https://t.co/8jIqNbtlfL
Investing involves risk. Principal loss is possible. Distributed by Foreside Fund Services, LLC.
Berkshire Hathaway $BRK.B bought home builder Taylor Morrison Homes $TMHC for $6.8 Billion
StockMKTNewz partner @adampatti went on CNBC to talk about it 👀