Leonardo Zepeda

[Leonardo Zepeda @leozgUS]

My AI infrastructure thesis was right.
My returns don't show it.

Back in spring I went long the upstream suppliers. Power, cooling, EPC, semi equipment. The ones with actual leverage over hyperscalers.

The fundamentals played out. Capex guidance for 2026 sits around $690-760 billion. Backlogs grew. Bottlenecks got worse, which was the whole point.

Then July took the theme down mid-teens in a month.

So what happened. Three things, and only one of them was in my model.

1. Rates stayed at 3.50-3.75% with hike odds priced in. Sticky rates crush the multiple on cash flows that arrive in 2029. Doesn't matter how good the backlog is.

2. The Hormuz situation. Brent near $100, helium spiking, shipping costs up. I honestly did not have a Gulf conflict hitting semiconductor inputs on my list. Nobody's list, probably mine least.

3. The market changed the question. It stopped asking "who's spending" and started asking "where's the free cash flow." Hyperscalers leaning on debt to fund capex made that worse.

What I got wrong wasn't the thesis. It was expecting a smooth ride on names that had gone parabolic. Mean reversion after a move like H1 isn't a signal, it's gravity.

Actually, one thing in the thesis got stronger. Energy security. Constellation, Vistra, Talen. A world where oil flirts with $100 makes contracted domestic power more valuable, not less. That's where my conviction went up.

Cooling names like Vertiv still off their highs. Demand intact. Cheaper entry than June.

NVIDIA and Broadcom I'm lighter on. Priced for perfection in a market that stopped paying for perfection.

The bottleneck is still the bottleneck. I just have to hold it through rate scares and a war premium now.

read more here: https://t.co/d0QgiHIpG0

[Mario Nawfal @MarioNawfal]

🇦🇷 The hate against Messi came out of nowhere, and it has nothing to do with football.

Here's what's actually going on...

Messi lifted the World Cup in December 2022 and the entire planet lost its mind celebrating.

Three and a half years later, a chunk of the internet suddenly decided he's a villain.

Nothing on the pitch changed. Same guy, same career, twenty years of footage anyone can check in under a minute.

What changed is the feed. Villain edits started outperforming legend edits and the algorithm noticed.

A 2018 MIT study tracked 126,000 claims across 3 million X users. False stories spread six times faster than true ones and got retweeted 70% more often. Outrage simply beats accuracy at getting attention.

None of this needs a single big lie either. Illusory truth effect: see a claim enough times and it starts feeling true even when you know it's false, and the effect survives fact checks.

Football is basically the most documented thing on earth. Every match, every angle, every ref decision, all archived. The narrative still won anyway.

If the algorithm can flip public opinion on the most watched athlete alive, it can flip it on anyone with no footage to fall back on.

Source: @LeoMessiFanZone, @therosieum / Writer: Sol

[Leonardo Zepeda @leozgUS]

Senior cloud architect.

20 years building infrastructure.

Replaced by AI in 90 days.

His daughter's cancer diagnosis came first.

The layoff notice came ten days later.

When he called his financial advisor panicked, silence.

Two weeks of nothing.

Then a product pitch with a 3-hour deadline.

I'm not the advisor in this story.

But I've watched this happen too many times.

Income gone.

Medical bills piling up.

Years of planning crumbling.

What broke him wasn't the job loss.

It was the silence from his financial advisor.

David worked with Henry for three years.
Months before the crisis, he raised concerns about his tech-heavy portfolio.

"Markets go up and down. We'll monitor it."

When the crisis hit, he called immediately.
"Henry, my daughter has cancer. I lost my job. I need to talk about cash flow, taxes, protecting what we have. Please call me back."

Two weeks of silence.
Just automated portal alerts.

Finally, Henry called back late Friday.
"Sorry for the delay. There's a structured note closing at 5 p.m. We should move thirty-five percent into it."

"Did you hear my voicemail? My daughter has cancer. I lost my job. I don't need a product."

"The window closes today. Let's get this done, we'll talk next week about the other stuff."

Pressured and drained, David agreed to end the call.

He made panicked decisions alone for weeks.

Word spread in his tech network.

Opportunities vanished.

A good advisor would've called the same day.

Listened first, then built a plan.

David found out later. Big commission tied to end of day Friday.

His crisis was a sales opportunity.

Your advisor should show up when you're scared.

If they don't, you have a salesperson with a fancy title.

visit   https://t.co/rWL42MLahP to learn more.

[Leonardo Zepeda @leozgUS]

Grid queues stretch 7 years.

Nobody's talking about it.

Hyperscalers are spending $650B to $720B on capex this year.

Everyone talks about the model.

Nobody talks about the cable.

The real constraint isn't compute. It's infrastructure that takes a decade to build.

Here's what's happening.

Power grid interconnection queues in key markets? Three to seven years.

Transformer lead times? Over two years.

30% to 50% of planned 2026 data center projects are delayed or cancelled.

This isn't a software problem. It's physics.

When demand is inelastic and supply is constrained by regulation, physical timelines, and engineering bottlenecks... the marginal supplier captures disproportionate value.

Always.

Basic microeconomics applied to an unprecedented moment.

Where's the concentrated value?

PwC's April 2026 study of 1,200+ executives across 25 sectors found this:

74% of all measurable AI-driven economic value is captured by 20% of organizations.

The other 80%?

Still in pilot mode.

The productivity paradox is real at the macro level.

But at the micro level, in the upstream supply chain, monetization is happening today.

Here's what most investors are missing.

You're holding NVDA, MSFT, GOOGL thinking you have AI exposure.

You're watching your tech-heavy portfolio and feeling positioned.

But the suppliers selling them power, cooling, and physical infrastructure are posting margin expansion at twice the rate.

The hyperscalers are price takers in this equation.

The infrastructure suppliers are price makers.

That's where the asymmetric returns are hiding.

Five suppliers hold real bargaining power.

→ Power generation

Hyperscalers don't scale without reliable, dispatchable electricity.

Suppliers with signed PPAs, nuclear assets, or owned generation have revenue visibility measured in decades.

A single large language model training run consumes energy equivalent to powering thousands of homes for a year.

Demand isn't elastic.

They pay the scarcity premium.

→ Thermal management and liquid cooling

Rack densities went from 40 kW to over 100 kW.

Air systems don't work anymore.

Liquid cooling is seeing backlog growth over 100% year-over-year.

Specialized providers pass through cost increases because hyperscalers prioritize uptime over negotiation.

→ EPC firms for data centers

Those delivering turnkey, custom-designed facilities operate with order books 10 to 12 months out or more.

Delivery speed equals higher margins equals repeat contracts.

→ Semiconductor equipment and foundry capacity

Advanced lithography tools and specialized packaging processes are essential for next-gen AI accelerators.

With global capacity utilization near peaks and fab expansions taking years to complete, these suppliers operate with near-monopoly pricing leverage at critical nodes.

→ Interconnect infrastructure

High-density optical fiber and connectivity moving data at terabit-per-second speeds between thousands of processors.

Capacity reservations and multi-year supply agreements lock in supplier advantage.

Morgan Stanley's March 2026 earnings review found that S&P 500 companies citing verifiable AI benefits are generating cash-flow margin expansion at roughly twice the global corporate average.

Not hype.

It's in the earnings.

Why now and not later?

The cloud computing cycle of the 2010s generated sustained returns for those who identified the supply chain inflection point early.

The current cycle compresses that timeline.

Bottlenecks are sharper.

The capital wave is larger.

And suppliers already contracted and scaled capture the scarcity premium first.

Those who wait watch value creation unfold without participation.

You've seen this movie before.

Early 2010s cloud infrastructure.

2020 pandemic digitization wave.

The pattern is identical: physical constraints create temporary monopolies, early allocators capture disproportionate returns, late movers pay premium valuations for diminished upside.

Our position.

We focus exclusively on upstream suppliers with structural bargaining power over the hyperscalers themselves.

Concentrated long positioning.

No exposure to model developers.

No broad market beta.

Selection criteria: locked PPAs, record backlogs, margin resilience tied directly to AI demand.

No single position exceeds 6%.

Quarterly rebalancing tied to earnings signals: backlog growth, PPA announcements, capacity utilization metrics.

The result is asymmetric participation: we capture the concentrated 20% value creation documented by PwC while avoiding the 80% still stuck in unproven pilots.

Most allocators won't move until this thesis becomes consensus.

By then, supplier valuations will reflect the scarcity premium.

PPAs will be locked for decades.

Order books will be closed to new entrants.

You'll be buying earnings multiples that already price in the next five years of growth.

We're positioning now, while the market is still pricing these suppliers as cyclical industrials instead of monopolistic infrastructure plays with decade-long revenue visibility.

Wharton's Penn Budget Model projects AI will lift GDP by 1.5% by 2035 and nearly 3% by 2055.

That's the long term.

Today, gains are front-loaded in the physical supply chain that makes AI computation possible.

Not in the software.

Not in the models.

In the cable.

In the transformer.

In the cooling system.

In the substation that takes seven years to connect to the grid.

The window is narrowing.

As hyperscaler urgency intensifies and more projects hit grid or cooling limits, the bargaining premium accrues first to those already contracted and scaled.

The broader market keeps debating productivity timelines.

We extract returns from the foundational layer while that debate continues.

We're opening a limited number of allocations this quarter.

If you're holding traditional AI exposure through software plays or broad tech indices, you're likely positioned in the 80% that PwC identified as capturing negligible value.

If your portfolio lacks direct exposure to the physical bottlenecks, you're watching $650B to $720B in annual capex flow past you.

We're available for a confidential portfolio review to show you exactly where your current holdings sit relative to the supply chain chokepoints.

The data is unambiguous.

The bottlenecks are verifiable.

The microeconomic incentives are measurable in quarterly earnings.

What's negotiable is your timing.

The question isn't whether the AI supply chain realignment will deliver outsized returns to early participants.

That's already happening in the earnings Morgan Stanley documented.

The question is whether you'll secure allocation before the next wave of institutional capital closes the window.

📩 DM "INFRASTRUCTURE" for a confidential analysis of your current AI exposure and how it maps to the actual value capture we're seeing in Q1 2026 earnings.

Limited availability this quarter.

read more at: https://t.co/rnSzbVkBz5

[Leonardo Zepeda @leozgUS]

Mexico City. April 2026.

The scent of lilies and copal incense filled the air.

Before me sat the mahogany casket of my friend and client for 16 years: media mogul, tech investor, real estate titan.

Cancer took him at 74.

His name was Alejandro Burillo.

Standing there, I finally understood what he'd been trying to teach me.

2008: The First Warning

Mid-2007. Alejandro felt something was wrong. "The market feels like a reckless party," he told his advisors.

"We need to trim equities. Now."

They shut him down.

Their models said otherwise. The data looked fine.

When 2008 hit, the damage went beyond numbers.

He watched his employees suffer while feeling completely powerless.

Technical certainty had silenced his market instinct.

2015: When I Made the Same Error

Alejandro wanted to move into live sports assets.

Soccer teams. Tennis tournaments. Live sports and lifestyle.

He saw steady cash flow and crowds who showed up rain or shine.

I pushed back.

Too much concentration risk, I said. Too uncertain for a developing economy.

When markets crashed again, those sports investments?

They held steady.

People kept filling stadiums even when times got tough.

"You saw numbers," he told me later. "I saw people."

2018: The Legacy We Let Fear Kill

He wanted to build a sustainable hospitality community.

High-end. Family legacy. Something for Mexico.

The models flashed red. Too risky. Too uncertain.

We talked him out of it. He scaled back.

That scaled-back project in tulum?

It roared.

A symbol of the impact we let spreadsheets steal from him.

The regret in his voice still haunts me.

What He Made Me Promise

Near the end, Alejandro looked at me. "Leo, promise you'll never let the charts silence a client's voice the way mine was silenced."

Every missed opportunity sat in those words.

Here's what I learned:

→ When clients say "this doesn't feel right," that's where analysis starts

→ Models without context are dangerous

→ The times we honored his instinct delivered returns AND peace of mind

→ Technical advice means nothing without the wisdom across the table

Alejandro's chair sits empty.

But his lesson is clear: intuition and data need each other.

One without the other destroys value. And sometimes, it destroys legacy.

to read the full article:

https://t.co/uRIkd7WsKJ

[Leonardo Zepeda @leozgUS]

Learn more of my opinion at : https://t.co/vwDqLlL6J0

[Leonardo Zepeda @leozgUS]

The top 10% live in a different economic reality.
We control 67% of America's wealth.
The bottom half owns just 2.5%.
But that's not the shocking part...

[Leonardo Zepeda @leozgUS]

We can use wealth to ensure AI becomes a bridge, not a barrier.
Fund education over automation.
Choose empowerment over efficiency.
Build systems that lift everyone up.
What kind of world do we want our children to inherit?

[Leonardo Zepeda @leozgUS]

learn more at https://t.co/vwDqLlL6J0

[Leonardo Zepeda @leozgUS]

🚨 Your $2M portfolio is BLEEDING $23,000 yearly in hidden taxes
Your investment advisor never mentioned it
This isn't a horror story - it's happening right now
And it's completely preventable

[Leonardo Zepeda @leozgUS]

The solution: WEALTH ENGINEERING
New minimum threshold: $10M portfolio
Professional guidance isn't optional anymore
Like/RT if you want to learn more about protecting your wealth from tax destruction