Skip to main content
The Frontier File · Issue #7 · July 5 2026 · H2 Begins

Chips Cooled.
The Grid Woke Up.

One retail account, one frontier-tech thesis, documented in public. The best quarter since 2020 is in the books. As I start the second half, the AI trade is quietly moving down the stack — from the chips to the electricity that runs them.

+65%
First quarter · on capital
est. · pending broker mark
$8,672
Book value · from $5,270
19 / 20
Positions green
Cold Open

The bottleneck moved

The first half ended with the best market quarter since 2020 — the S&P up 14%, the Nasdaq up 20% — and my little account up about 65% on capital in its first ninety-odd days. Then the last week did something more interesting than go up: it rotated. A soft jobs report cooled the Fed, the Dow hit a record, and semiconductors — the group that carried everything — sold off 4.5% as investors took profits in the most crowded winners. My Micron dropped 5.5% in a day.

Read the wrong way, that's scary. Read right, it's a map. Chips didn't break; they got repriced, while money hunted for the next leg of the same story. And the next leg was hiding in plain sight this week: on the hottest day of the summer, the largest U.S. grid operator braced for a record 166 gigawatts of demand, with data centers alone driving billions in new power costs. Hyperscalers are pouring $400-billion-plus into AI infrastructure this year — and a growing share of that isn't chips at all. It's power, grid, cooling, and the raw materials to build them.

That's the thesis evolving in real time. The AI trade isn't dying. It's descending the stack — from silicon to the electricity that makes silicon useful. And I already own the beachhead.

The Thesis Board

Where my head is at

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

Power & Grid BWXT · OKLO · CENX → grid/cooling/storage
UPGRADED this issue. PJM hit record 166GW demand as data centers strained the grid. AI's limiting factor is shifting from chips to electricity. My nuclear/materials names are the beachhead; transmission, cooling, and storage are the expansion I'm researching now.
AI Compute / Memory MU · STX · NVDA · ARM
Repriced, not rejected. The chip complex sold off on profit-taking, not a demand break — MU literally guided to $50B/quarter last week. Still the engine. Still, in MU's case, too big.
Concentration MU ~21% · main controllable risk
Both my research passes — and I mean both — flagged the same thing: trim MU toward 17–18%, rebuild cash. It bit 5.5% Thursday. The dovish tape is the calm window to finally act. This is Q3's first job.
Space Economy RKLB · PL · MDA · ASTS · SPCX
RKLB's $8B Iridium buy turned it into a consolidator; SPCX joins the Nasdaq-100 this week (mechanical buying). The market now rewards execution over narrative — good, because mine has contracts and backlog.
Rates hike odds falling
Soft jobs (57K) dropped July-hike odds below 20%. Yields eased. A tailwind for every long-duration name I own — space, eVTOL, quantum, nuclear. But CPI July 14 is the real referee.
Seeds SERV · QBTS · JOBY · ACHR
Planted small, earning promotion through milestones, not price action. The dovish shift helps the rate-sensitive ones — but lower rates don't certify aircraft or ship robots. Patience.
The Week in One Tape

Just the numbers

MarkerLevelRead
Dow record close
52,900
+1.14%
Nasdaq chips drag
25,833
-0.80%
SMH semis profit-taking
-4.5%
June jobs dovish miss
57K
vs 113K
PJM peak demand record
166.2GW
grid strain
WTI Crude Iran easing
~$67
Mar low

Q2 closed as the strongest quarter in years — S&P +14%, Nasdaq +20%. The last week's split (Dow record, Nasdaq down) is the tell for H2: leadership is broadening out of the crowded chip trade into the physical layer underneath it. Power demand hitting a record on the same week chips sold off isn't a coincidence. It's the story rotating one level down.

The Portfolio Report

The whole book — H2 opening marks

MU
+254%
~21% · trim me
RKLB
+83%
Bought Iridium
PL
+69%
Sat imagery
CENX
+48%
Power/materials ▲
STX
+47%
AI storage
BWXT
+41%
Nuclear/power ▲
AEVA
+38%
LiDAR
GOOGL
+33%
Dow member
MDA
+32%
Space systems
ASTS
+30%
Sat connect
ACHR
+27%
eVTOL
JOBY
+26%
eVTOL
NVDA
+21%
Repriced
SPCX
+19%
Nasdaq-100 add
AMZN
+19%
Cloud · robots
SOFI
+9%
Fintech rails
OKLO
+6%
Nuclear/power ▲
QBTS
+4%
Quantum seed
SERV
+3%
Delivery bots
ARM
-2%
Lone red

This issue's moves

Reframed: BWXT, OKLO, and CENX from "nuclear/materials" to the front edge of a power thesis — the AI bottleneck moving to electricity. Standing order, Q3 job #1: right-size MU toward 17–18%, then reassess RKLB after the Iridium financing and integration picture is clearer; rebuild cash deliberately. Researching: grid equipment, transmission, cooling, and storage — the next layer of the same buildout. All marks are estimates pending a broker refresh.

+65%
On capital · est.
$8,672
Book value · est.
19/20
Positions green
~3%
Cash — too low
~4×
vs S&P · 93 days
The Catalyst Calendar

We position around dates

Mon Jul 6/7
SPCX joins the Nasdaq-100 — mechanical index buying kicks in
Tailwind
Wed Jul 8
FOMC June minutes — first read behind Warsh's hawkish-hold curtain
High
This week
Right-size MU first; reassess RKLB after the Iridium deal settles
My move
Mon Jul 14
June CPI + Q2 bank earnings open — the real macro referee
Critical
Jul 28–29
FOMC decision — does the Warsh Fed hold or hike?
Critical
Q3 2026
Honeywell Aerospace (HONA) spin-off — the GE-style unlock
On deck
The Hot Take

"The AI trade isn't topping. It's descending the stack. First the chips, now the power that runs them — and eventually the dirt you dig to build the grid. Follow the electrons."

Every big technology wave gets priced in layers, top down. The market bid up the obvious layer first — the GPUs and the memory — until those names got crowded and started taking profits, which is exactly what happened this week. But the demand didn't evaporate. It moved to the constraint underneath. And right now the binding constraint on AI isn't chips. It's electricity. A grid operator just braced for record demand because data centers are eating power faster than anyone planned for, and hyperscalers are spending $400 billion-plus a year on infrastructure — a huge and growing slice of it on generation, transmission, and cooling.

That's why I'm not panicking about a chip pullback — I'm following the money one level down. I already own the beachhead: BWXT in nuclear, OKLO in next-gen reactors, CENX in the aluminum that builds the grid. The next research is the rest of the layer — transmission gear, cooling, storage, the unglamorous industrials that turn megawatts into uptime. The frontier people picture is a robot or a rocket. The frontier that actually gets built first is a substation. Own the electrons.

Buddy, Real Talk

Two things I owe you straight

First, the number at the top of this issue — +65%, ~$8,672 — is an estimate. I mark my book off index moves and holding-level reads, not a live brokerage feed, and this week's chip drop plus the Iridium pop in my space names means the real figure could sit a few points either side of that. I'm labeling it as an estimate everywhere in this issue, and before I ever put a hard performance number on the public site, I reconcile it against the actual account. A newsletter that shows returns has exactly one job it can't fumble: the returns have to be real. So I'd rather tell you "about 65%, pending confirmation" than print a clean-looking number I haven't checked.

Second — the Micron trim. I've now flagged it for five straight issues, and this week even a second, independent pass at my whole book came back with the identical verdict: MU is ~21%, it's the main risk I can actually control, trim it toward 17–18%. When every read points the same way and I still haven't moved, that's not conviction — it's avoidance wearing conviction's jacket. The chip rout gave me a 5.5% preview of what an oversized position does on a bad day. Q3's first act, before any new pick, is to right-size it into this dovish, calm tape and rebuild the cash cushion from a too-thin 3%. Next issue reports whether I did it. No more jacket.

The Bottom Line

Second half: follow the bottleneck

Ninety-three days, an estimated +65%, the best market quarter since 2020 behind me — and a clear map for what comes next. Chips cooled, the grid woke up, and the AI trade started doing what big trades always do: broadening out of the obvious winner into the layer underneath.

So H2 has a plan, not just a hope. Lean the power thesis up the board — nuclear, grid, cooling, materials — because electricity is becoming AI's binding constraint and I already own the beachhead. Right-size the known single-name risk, reassess the transformed Rocket Lab thesis, and rebuild the cash I've let run too thin. Respect the real referee, June CPI on the 14th, and the Warsh Fed at month-end. And keep every speculative seed small until it earns promotion with a milestone, not a price spike.

The first quarter was extraordinary and partly lucky — a war ending, a historic IPO, a memory blowout, a space consolidation, all breaking my way. The second half won't hand me that. What it'll reward is the boring stuff: sizing, cash, patience, and following the money down the stack instead of chasing it back up. The frontier is still the trade. The vessel just needs to be worthy of it.

See you next Sunday, buddy. — Alex

§ 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.