Skip to main content
The Frontier File · Issue #15 · August 23 2026 · Into Nvidia / Jackson Hole

The Market
Raised The Rent.

One concentrated frontier-tech portfolio, documented in public. The long bond hit 2007 territory, semiconductors cracked, and the portfolio’s latest moves made one thing clear: the future still works — the market is just charging more to wait for it.

Aggressive growth · asymmetric upside · no fake certainty · research notes, not financial advice
5.34%
30-year Treasury high
time got expensive
−2.05%
Nasdaq · week
duration got repriced
~18%
BWXT · portfolio
core ceiling reached
Aug 26
Nvidia reports
the next AI receipt
Cold Open

The future did not break. It got a carrying cost.

There are weeks when a company changes the story, and there are weeks when the price of money changes every story at once. This was the second kind. The 30-year Treasury yield pushed to 5.34% — the highest level since 2007 — while the Nasdaq lost 2.05%, the S&P 500 fell 1.43%, and the semiconductor index dropped about 5%. Friday bounced. The week still ended with the market sending one very clear invoice to anyone whose investment case contains the phrase “by 2030.”

Nothing about artificial intelligence suddenly stopped working. Satellites did not fall out of orbit. Nuclear reactors did not forget how to split atoms. Archer did not wake up and discover that aircraft are a fad. What changed was the hurdle rate. A company asking me to wait years for its cash flow is now competing with a government bond offering more than 5% while I do absolutely nothing except own it.

That matters to this portfolio because I own companies scattered all along the timeline. Microsoft, Alphabet, Amazon, Nvidia, Micron, Seagate and BWXT already generate real cash from real demand. Rocket Lab, Planet, MDA, AST SpaceMobile and Archer are farther out on the curve: increasingly real businesses, but still businesses where execution and capital requirements matter enormously. Then there are the seeds — Oklo, Serve, Aeva, D-Wave — where the future is the whole point and therefore the price of waiting matters most.

And right in the middle of that rate reset, I made four portfolio decisions that looked unrelated on the screen: I sized BWXT up to the top of the book, started a small Reddit position, trimmed Archer, and gave newly independent Honeywell Aerospace a tiny surveillance seat. The common thread is more useful than any of the individual trades.

How much uncertainty am I willing to finance before the business starts paying me back?

That is the question this week. Not whether the frontier is real. It is. The question is which parts deserve expensive capital now, which parts still need proof, and which parts should stay seasoning no matter how good the story sounds.

The Thesis Board

Where my head is at

The map before the positions. firing   mixed / watching   next-wave / emerging   active headwind.

Power & Nuclear BWXT · OKLO · CENX
The AI build keeps descending into the physical layer: generation, fuel, components, transmission and manufacturing capacity. BWXT is now the clearest example in the book of a bottleneck that already gets paid. The speculative nuclear names still have to earn their way there.
AI Compute NVDA · MSFT · GOOGL · AMZN
Still firing, but Wednesday is a serious receipt. Nvidia has become a read-through not just on chips but on the hundreds of billions being financed around data-center construction. Demand can stay exceptional while the financing math gets harder.
Memory & Storage MU · STX
The clean scarcity-to-cash layer remains intact. These businesses sit close enough to the AI spend to monetize the build now, not someday. This week’s semiconductor selloff is a valuation and rates event until the order books say otherwise.
Human Data / Provenance RDDT
New this issue, but not the whole issue. Generative AI makes content abundant; authentic human experience may become relatively scarcer. Reddit has the scale and financial profile to make the idea investable. It still has to prove that the platform captures the value instead of merely supplying the raw material.
Space Systems RKLB · PL · MDA · ASTS · SPCX
Operational proof keeps improving across the cluster, but the long bond is a reminder that rockets, constellations and manufacturing lines consume capital before they become annuities. I still like the destination. I am less interested in financing every mile of the trip at any price.
Physical AI / Air Mobility ACHR · JOBY · SERV · AEVA
Archer’s Boeing transaction materially upgraded the capability stack — autonomy, defense drones, airspace software and real revenue — while simultaneously raising integration complexity. Better company, higher burden of proof. That tension is the theme.
Aerospace Turnaround HONA
Excellent assets, deep backlog, ugly execution bottleneck. Honeywell Aerospace does not need more demand; it needs suppliers and output to catch up with the demand already sitting there. A surveillance position is enough until that happens.
Cost of Capital the whole book
The 30-year yield is the new pressure gauge. Core can survive expensive time. Growth has to convert milestones into economics. Speculative and Moonshot names become less forgiving every basis point higher.
The Week in One Tape

The landlord sent the invoice.

MarkerLevelRead
S&P 500 Friday rebound, weaker week
7,674.37
−1.43% wk
Nasdaq duration / AI sensitivity
26,180.46
−2.05% wk
30Y Treasury long-end stress
5.34% high
highest since 2007
Philadelphia Semis AI bellwether group
~−5% wk
Brent crude Iran / inflation pressure
near $95
+6.39% wk
RDDT index pop fully tested
$178.09 → $153
−14.1%

The easy interpretation is “risk-off.” The more useful one is that the market spent the week repricing duration. Oil kept inflation pressure alive, the long end refused to calm down, and the most expensive future cash flows got marked down first. That is uncomfortable in a frontier portfolio, but it is also clarifying. When the tide of cheap capital goes out, I get a much cleaner look at which businesses are actually carrying their own weight.

The Portfolio Report

The hierarchy finally had to earn its keep.

This was not a week for finding a fifth move. It was a week for understanding the four I had already made — and what they say about the kind of portfolio I am trying to build.

Core · BWXT

The paid bottleneck got the biggest seat.

BWXT reported a stronger-than-expected second quarter, raised full-year EBITDA, EPS and free-cash-flow guidance, completed the Precision Components acquisition, and agreed to sell its medical business for up to $800 million so more capital can go toward nuclear national security and commercial nuclear power. That is what I mean by a bottleneck already getting paid. The position is now roughly 18% of the book — which is also why the next action is to stop adding and let the thesis work.

Growth · AI Stack

Wednesday is an MRI for half the book.

Nvidia reports this week, but the read-through goes far beyond NVDA. Micron and Seagate need the AI build to keep consuming memory and storage. Microsoft, Alphabet and Amazon need the economics of their capex to keep making sense. Even the power thesis cares because every rack eventually becomes an electrical load. Nvidia’s latest financing partnerships target more than $500 billion of AI infrastructure — a staggering number and a reminder that AI demand now has a balance-sheet side.

Growth · Space

The destination can be right while the financing gets harder.

Rocket Lab, Planet, MDA and AST SpaceMobile are no longer just speculative stories; each has real contracts, hardware and operating proof. That is progress. It also does not repeal math. Launch vehicles, satellite constellations and manufacturing capacity are expensive things to build, and a 5%-plus long bond raises the value of every future dollar they have not earned yet. A red week in space is not an automatic invitation to average down. Backlog conversion, margins, deployment and cash use still decide who earns the next dollar.

Growth / Speculative · ACHR + HONA

One got more capable. One needs to get less messy.

Archer is buying Boeing’s Wisk, Insitu and SkyGrid, adding autonomy, airspace intelligence and a profitable defense business with more than $200 million in annual revenue. That makes the company strategically stronger and operationally harder — exactly why I trimmed rather than chased the excitement. HONA is the mirror image: $4.5 billion of quarterly sales and an $18.2 billion backlog prove the demand, while supply constraints forced management to cut 2026 growth and EBIT guidance. Tiny position, front-row seat, no pretending the repair is finished.

The new idea hiding inside the week

Reddit still belongs in this issue, just not at the expense of everything else. The company entered the S&P 500 on Tuesday, then spent the week giving back the index pop. Underneath the price action, the business is difficult to laugh off: second-quarter revenue reached $805 million, net income was $253 million, adjusted EBITDA was $343 million, and weekly active uniques crossed 514 million.

The reason I started a small position is not the index inclusion. It is the possibility that generative AI is quietly changing what the scarce asset on the internet actually is. When anyone can generate a polished answer, review, article or recommendation in seconds, the thing that becomes harder to manufacture is provenance — somebody who actually bought the product, drove the truck, stayed in the hotel, fixed the engine or worked inside the company.

That is an interesting thesis. It is not yet a Core thesis. Reddit still has to prove durable direct-user growth, advertising strength and the ability to capture value from the human conversation AI systems increasingly want to summarize. So the position stays small. A new idea should buy my attention before it buys a large piece of the portfolio.

The portfolio rule this week

Core owns bottlenecks that already get paid. Growth owns businesses proving that scale improves the economics. Speculative and Moonshot positions finance possibility — which means they should become smaller, not larger, when the price of waiting rises.

The Catalyst Calendar

Wednesday asks about demand. Friday asks about the price of money.

Wed · Aug 26
8:30 a.m. ET
July PCE + second-estimate Q2 GDP + preliminary corporate profits.The inflation print and growth revision arrive together. For this portfolio, watch the long end more than the headline reaction.
Macro
Wed · Aug 26
~4:20 p.m. ET
Nvidia fiscal Q2 FY2027 results.Conference call at 5:00 p.m. ET. This is the cleanest near-term receipt for AI compute, memory, storage, data-center financing and eventually power demand.
Portfolio
Fri · Aug 28
10:00 a.m. ET
Fed Chair Kevin Warsh — Jackson Hole keynote.His first Jackson Hole as chair lands after the PCE/GDP data. The market wants a framework for policy in a world where inflation, oil and long-term borrowing costs are all refusing to behave politely.
Rates
Next receipts
Space, aerospace and physical AI move back to execution.For RKLB / PL / MDA / ASTS: backlog conversion and margins. For ACHR: integration and certification. For HONA: output and supply-chain improvement. No invented dates — only evidence when it arrives.
Watch

That sequencing is almost too neat. Wednesday morning tells me what inflation and growth are doing. Wednesday afternoon tells me whether the AI spending machine is still accelerating. Friday tells me how the Fed chair thinks about the cost of capital wrapped around all of it. Demand, economics, financing — three different receipts for the same portfolio.

The Hot Take
The next phase of the frontier trade is not about finding the wildest technology. It is about finding the scarcity that can turn itself into cash before the market gets tired of financing the wait.

That is why I keep coming back to bottlenecks. Memory got paid because the supply was scarce. Storage got paid because AI made capacity valuable. BWXT gets paid because qualified nuclear manufacturing is not something a hyperscaler can spin up in a weekend. Reddit may get paid if authentic human provenance becomes scarce enough — but that conversion still has to be proven.

Meanwhile, some of the most visually futuristic companies I own can still be terrible investments at the wrong price or wrong size. A working aircraft is not automatically a good stock. A satellite in orbit is not automatically a margin. A reactor design is not automatically cash flow. Technology answers “can this exist?” Investing has the much ruder follow-up: “who gets paid, when, and how much capital disappears first?”

Higher yields make that second question impossible to avoid. I think that is healthy.

Buddy, Real Talk

A red screen is not a clearance rack.

This is exactly the kind of week that makes every position start whispering the same thing: I’m cheaper now. Some of them are. Some of them are simply less expensive than they were when I was more excited about them.

BWXT fell roughly ten percent this week after I intentionally made it the largest position in the book. The business did not suddenly get worse. That does not mean I need to reward the decline with even more concentration. RDDT’s thesis got cleaner as the index premium came out. That does not mean a starter position needs a second tranche five days after I bought it. Archer is a much more compelling strategic company after the Boeing deal. I still trimmed it because the company also became harder to execute. Those are not contradictions. That is position sizing doing its job.

The same goes for the speculative names. Oklo can fall a lot and still be a pre-commercial reactor developer. Aeva can fall a lot and still have to turn technical wins into revenue. Serve can trade under five dollars and remain seasoning. The first rule from Issue #1 survives every market regime because it is not really about price — it is about survivability.

Cheap and distressed are different. A better story and a better stock are different. A lower price and a better risk/reward are sometimes different too.

I already made the meaningful moves. I do not need Monday morning to become a shopping montage just because Friday left a lot of red numbers lying around.

The Bottom Line

The portfolio is getting older in a good way.

Not older as in boring. Older as in less willing to finance every exciting sentence equally.

Earlier in this journey, the frontier was mostly about identifying what could work: AI infrastructure, launch, satellite networks, eVTOL, nuclear, robotics, quantum. Fifteen issues in, more of those technologies are working. That moves the argument. The question is increasingly not whether the future arrives, but which companies own enough of a bottleneck to collect cash when it does.

That is why BWXT can be the largest position while Oklo stays small. Why Nvidia, Micron and Seagate can sit closer to the steak while the next generation of physical-AI names stays farther out. Why Archer can become strategically better and still deserve less capital. Why HONA can have an $18 billion backlog and still need to prove its supply chain. Why Reddit can introduce a genuinely interesting human-data thesis without taking over the whole portfolio — or this whole newsletter.

This week the market raised the rent on the future. Good. I would rather discover now which tenants can pay it.

Wednesday brings the AI receipt. Friday brings the rate framework. Between them, I get a cleaner answer to the two questions that matter most right now: is the build still accelerating, and what does it cost to finance?

Until then, the plan is gloriously unexciting: hold the Core, make Growth prove the economics, keep Speculative and Moonshot small enough to remain curious, and let the next dollar be earned instead of invented.

See you next Sunday, buddy. — Alex

Source Notes

Receipts, not vibes.

  1. Reuters · Aug. 21 market close — S&P 500 −1.43%, Nasdaq −2.05%, Dow −0.85% for the week; oil and bond-yield pressure.
  2. Reuters · Big, bad bond market — 30-year Treasury yield reached 5.34%, highest since 2007.
  3. Reuters · Week Ahead — Nvidia / Jackson Hole focus, semiconductor weakness, AI-infrastructure financing context.
  4. BWXT Q2 2026 results — raised EBITDA, EPS and free-cash-flow guidance.
  5. BWXT medical-business sale — transaction valued at up to $800 million.
  6. Reddit Q2 2026 results — $805M revenue, $253M net income, $343M adjusted EBITDA, 514.6M WAUq.
  7. S&P Dow Jones Indices — Reddit added to the S&P 500 effective Aug. 18.
  8. Archer + Boeing transaction — Wisk, Insitu, SkyGrid and >$200M annual revenue contribution from Insitu.
  9. Honeywell Aerospace Q2 2026 — $4.5B sales, $18.2B backlog, supply constraints and revised guidance.
  10. Nvidia earnings schedule — results Aug. 26 at approximately 1:20 p.m. PT; call at 2 p.m. PT / 5 p.m. ET.
  11. U.S. Bureau of Economic Analysis — July Personal Income & Outlays / PCE and Q2 GDP second estimate on Aug. 26 at 8:30 a.m. ET.
  12. Federal Reserve calendar — Chair Kevin Warsh Jackson Hole keynote Aug. 28 at 10 a.m.
§ Related reading
§ The frontier continues.

Get the next issue when it drops.

One sharp dispatch from the frontier: what changed, what matters, what I'm watching, what the evidence changed — and what the portfolio is doing about it.

No spam. No daily market sludge. Research notes only. Unsubscribe whenever.