The Future Has a
Carrying Cost.

A company can be right about the future and still lose to a Treasury bond.

That sounds ridiculous until the Treasury bond pays almost 5%.

Then it becomes math.

Oracle spent this week proving something I already believed: AI infrastructure demand is enormous.

Cloud infrastructure revenue rose 121%. Remaining performance obligations reached $664 billion. Oracle added another 850 megawatts of data-center capacity and says it delivered more than 300,000 GPUs to AI-cloud customers since the end of the prior quarter.

The machine is hungry.

Then Oracle showed us something almost as interesting as the demand.

Oracle sold $20 billion of common stock during the quarter.

That landed while the 10-year Treasury was flirting with 5%, producer prices were running 5.4% above last year, headline CPI sat at 3.4%, oil was climbing through triple digits, and markets were increasingly convinced the Fed would hike on Wednesday.

The AI build is real.
The space build is real.
The power bottleneck is real.
Defense demand is real.

But reality does not get a free multiple.

Issue #17 was called Proof Is Not a Multiple.

Proof still has to survive the carrying cost.

Time just became a fundamental.

For years, frontier investing rewarded the longest imagination in the room.

If a company could describe a sufficiently enormous future, investors were willing to travel a fairly impressive distance to collect the eventual earnings. Near-zero interest rates helped. Money waiting five years did not have many attractive alternatives.

That world is not this world.

The 10-year Treasury reached about 4.98% this week before finishing Friday around 4.93%.

Every investment in my portfolio therefore acquired an annoying new competitor:

A company promising enormous cash flow in 2030 is no longer competing only with other frontier companies. It is competing with an asset that pays while I wait.

This does not mean I suddenly want a portfolio of Treasury bills and orthopedic footwear.

It means the hurdle changed.

The further away the economics are, the better those economics eventually need to become.

Time is no longer a chart footnote.

Time is part of the position.

AI is becoming a capital trade.

The AI thesis inside this journal keeps moving down the stack.

GPUs.
Then memory.
Then electricity, cooling, transformers, nuclear and grid.
Now: capital.

Oracle's quarter was magnificent on demand. It was also a reminder that artificial intelligence does not materialize because somebody says “agentic” four times on an earnings call.

Factories get built. Data centers get built. Transmission gets expanded. Power gets contracted. GPUs get purchased by the hundreds of thousands.

And somebody writes the check.

Oracle's $20 billion equity sale makes the distinction unusually visible.

The useful question is no longer simply:

Practically everybody standing near a server can create a slide answering that.

The harder question is:

Nvidia sells scarce compute.
Micron sells memory.
Cooling and electrical suppliers sell infrastructure.
Nuclear manufacturers monetize power scarcity.
The hyperscaler writes the enormous check.

Those may all belong to the same AI build. They do not occupy the same economic position inside it.

The future is being constructed.

I increasingly want to own the companies getting paid while the scaffolding is still up.

The grid keeps wandering into the AI story.

Google announced another roughly $15 billion of AI infrastructure investment this week—this time in Finland.

Three new data centers. Grid upgrades. Battery storage. And a 22-year agreement to purchase as much as half the output of Finland's Loviisa nuclear plant.

There is something almost funny about watching artificial intelligence repeatedly rediscover civilization's oldest questions:

Where does the energy come from?
How do we move it?
Who owns the infrastructure?
How fast can we build more?

The model may live in the cloud.

Its electric bill does not.

My portfolio already owns this thesis through BWXT.

Which is exactly why I sold one share this week.

Not because BWXT became less attractive.

Because it became too successful at occupying my portfolio.

There is a psychological trap in investing where a winning position starts feeling safer precisely while the financial consequence of being wrong keeps getting larger.

That is how conviction quietly puts on avoidance's jacket.

I did not need less conviction.

I needed slightly less concentration.

Then I bought one share of a problem.

The proceeds did not remain cash for very long.

I bought one share of Honeywell Aerospace at $155.42.

HONA is almost the opposite of my normal speculative position.

The speculation is not whether aerospace exists.

Honeywell Aerospace did roughly $4.5 billion of sales last quarter. Its backlog reached $18.2 billion. Commercial aftermarket demand is real. Defense and space demand are real.

The problem is making enough of what customers already want.

Management cut expected full-year organic growth from 7–9% to 4–5% and reduced adjusted EBIT guidance from $4.65–$4.75 billion to $4.35–$4.45 billion.

Supply constraints remain part of the story.

So does execution.

That is a specific kind of broken.

And specific brokenness interests me considerably more than vague cheapness.

So it gets:

One share.

Not five.

Not “the market doesn't understand this.”

Not averaging down because I enjoy my own thesis.

One.

What the trade actually means:

I did not trade a better company for a worse company.

I traded a small amount of concentration for a small amount of curiosity.

If HONA improves output, supply-chain performance and margins, it can earn another dollar.

If not, I bought an education for roughly $155.

I can live with that tuition.

CEO Jim Currier presents at Morgan Stanley Laguna on Tuesday morning. I want to hear about throughput, aftermarket, supply constraints and confidence in the reduced guide.

Conference commentary is information. Not permission.

Rocket Lab is shortening its own future.

Rocket Lab launched another Electron on Friday.

Its 16th launch of 2026.

Its 95th launch overall.

Successful deployment. Another customer served. Another receipt.

Then came the announcement I may actually like more.

Rocket Lab released IMM Apex, a production solar cell with 31.5% beginning-of-life efficiency and roughly 40% lower mass. No germanium substrate. Available now.

A solar cell does not livestream quite as well as a rocket launch.

Economically, I like it.

Because the best Rocket Lab thesis was never:

It is increasingly:

Launch.
Spacecraft.
Components.
Solar.
Defense.
Eventually communications infrastructure.

Neutron still matters enormously.

But every dollar Rocket Lab can earn before Neutron has to carry the commercial thesis reduces how much of the valuation depends on one future event.

Duration compression.

Take a future thesis.

Monetize pieces of it today.

Turn tomorrow's possibility into today's revenue stream.

That is how a Growth company begins earning the right to become Core.

Rocket Lab's stock finished the week slightly lower anyway.

Good.

A rocket did its job and the quote did not throw a parade.

Thesis confidence: up. Position size: unchanged.

If a fresh Growth dollar eventually appears—after the Fed and after I rebuild some cash—RKLB has moved near the top of the underwriting queue.

Not because the stock fell.

Because the business broadened.

The market ran the experiment for us.

Friday bounced.

The week still lost.

That last number matters.

Smaller businesses generally depend more heavily on future capital. Their profits tend to live farther away. Their refinancing choices matter more.

So when Treasury yields march toward 5%, the market does not need a conspiracy to punish them.

It needs a calculator.

My portfolio showed the same pattern. Good operating evidence did not automatically produce good stock performance.

That is useful.

Not every red candle means somebody discovered your thesis was wrong.

Sometimes the market simply increases the admission price for waiting.

Then Sunday moved the goalposts again.

Friday's oil relief did not survive the weekend.

Saudi Arabia's East-West pipeline remained offline after drone strikes hit the route. That pipeline had become a critical bypass around the already disrupted Strait of Hormuz.

Reuters estimated that a prolonged outage could threaten roughly 4% of global oil supply.

By Sunday evening Brent had moved above $107.

That is no longer merely an oil story.

A diesel story.
A transportation story.
A manufacturing-input story.
An inflation-expectations story.
And therefore a Fed story.

I have no idea where crude trades Monday afternoon.

That is precisely why I do not need a clever oil trade Monday morning.

Sometimes respecting uncertainty is the trade.

The portfolio, without the theater.

One new question goes into the playbook.

Every frontier thesis now gets a duration test.

I already ask:

Can they build it?
Will somebody buy it?
Does scale improve the economics?

Now I add:

Rocket Lab becomes more interesting because it is shortening its own clock.
Oracle becomes more complicated because extraordinary demand arrives beside extraordinary capital requirements.
BWXT already gets paid for scarce capacity.
HONA already has customers; the problem is converting demand into enough output at the right margins.
SERV still lives much farther away from self-financing economics.

Same frontier.

Very different clocks.

No pre-Fed shopping trip.

Not buying Nvidia because it had a bad week.
Not chasing Micron before September 30.
Not buying another HONA because I enjoyed writing the HONA section.
Not rescuing Planet.
Not opening Bloom, Vertiv or Eaton because “time-to-power” is a good thesis.
Not promoting Oklo because Google signed another nuclear agreement.
Not inventing another Moonshot because rising rates made the existing Moonshots cheaper.

The book owns enough future.

It needs more optionality.

The cash problem remains boring enough to matter.

Selling one BWXT and buying one HONA improved concentration.

It did not improve liquidity.

Different job.

Cash remains too thin.

I still want roughly a 3–5% operating reserve—preferably rebuilt with fresh money rather than amputating the Core simply to satisfy a spreadsheet.

Cash does not need to outperform Nvidia.

Its job is to prevent me from selling Nvidia when something else becomes irresistible.

Sometimes the smartest new ticker really is USD.

Still annoying.

Still true.

Next receipts.

Those two company-specific events matter more to this portfolio than discovering another ticker between now and then.

Receipts.

Oracle — Q1 FY27 financial results

Reuters — Google Finland AI infrastructure and nuclear agreement

Rocket Lab — Mission success / launch #16 of 2026

Rocket Lab — IMM Apex solar cell

Honeywell Aerospace — Q2 2026 results and outlook

Reuters — August CPI

Reuters — 10-year Treasury approaches 5%

Reuters — Saudi pipeline outage and global oil-supply risk

Associated Press — weekly U.S. market performance