Engineer in Dublin building a concentrated US growth portfolio.
AI infrastructure • Space • Growth.
I share the thesis, numbers and what I’m getting wrong.
Core book, 2 Nov. Weights move with price. These are the trades.
$TSLA 34%. Cost $179.
$PLTR 13%. Cost $25.
$SOFI 7%. Cost $11.
$RKLB 7%. Cost $34.
$NET 7%. Cost $99.
$TEM 6%. Cost $55.
$AMZN 5%. Cost $171.
$MSFT 5%. Cost $510.
$CBRS 3.8%. Cost $141. Added lower.
$META 3.7%. Cost $624. Added 4 at $622.
$NVTS 2.2%. Cost $8.
$DOCN 2%. Cost $93. New. 15 shares at $96.
$SPCX 1.9%. Cost $134.
$UPST 1.7%. Cost $29. Still the one I’m least happy with.
Cash is thin. Not adding again just because the screen is red.
Rockets book. Smaller book.
$NBIS 26%. Cost $215. Last $156.
$IREN 20%. Cost $51. Last $24.
$ASTS 19%. Cost $78. Last $41.
$OSCR 14%. Cost $23.
$ONDS 11%. Cost $6.7.
$CIFR 10%. Cost $18. Last $7.
High-beta. They’ll move more than the index both ways. The cut is still whether the megawatts and the network turn into revenue. Not the de-grossing print.
https://t.co/n966L47HaM
Core book portfolio update.
$TSLA 34%. Cost $179.
$PLTR 14%. Cost $25.
$NET 9%. Cost $99.
$RKLB 7%. Cost $34. Added under the last print, not under the old average.
$SOFI 7%. Cost $11.
$TEM 6%. Cost $55.
$AMZN 6%. Cost $171.
$MSFT 5%. Cost $510.
$CBRS 3.6%. Cost $153. Added lower.
$NVTS 2.4%. Cost $8. Added.
$META 2%. Cost $626.
$UPST 2%. Cost $29. Still the one I’m least happy with.
$SPCX 0.9%. Cost $129.
Trimmed a bit of $COIN. No other exits.
Almost no cash. Not adding again just because the screen is red.
https://t.co/mAhjEnhWXj
Colette Kress said leading software and security companies announced products built on post-trained open models this quarter. Microsoft $MSFT, Salesforce $CRM, Palantir $PLTR, CrowdStrike $CRWD, Cadence $CDNS and Synopsys $SNPS. Training partners like Scale, Prime Intellect and Applied Compute are helping enterprises tailor open models to their own data. Abridge is using them in healthcare.
I own $PLTR. Cost $25. Last $157. The open-model list is the distribution. The part I’m watching is whether that turns into paid usage on the platforms that already sit in the workflow, or whether it stays a feature that gets bundled in.
The GPU print is the demand. The open-model adoption is the next layer. That’s the part I’d want to hear more on.
Nvidia Chair said the board does not intend to reinstate Huang. Data center lease commitments and cloud service agreements are now $102B, up from $74B last quarter. Upcoming leases run up to 20 years. Still no China Data Center compute revenue in the outlook. Commerce Department change would block the chips from being rented to Chinese companies entirely.
I don’t own $NVDA. The $102B in lease commitments is the capacity still being locked in. The China restriction is the part that didn’t change with the print. That’s the overhang I’m still watching.
$NVDA call is on. Colette Kress is interim CEO. From the release and early comments: roadmap is on track. Vera Rubin NVL72 has shipped. “Our momentum is real.” Gross margin dip sequential is from ramping Rubin production. Q4 guide $130B, no China Data Center compute revenue assumed.
I don’t own the stock. The print is the demand. The numbers show the spend is still happening.
$CBRS just printed 70% core revenue growth and a Microsoft deal. $NBIS and $IREN are in the rockets book. The GPU is the capacity. The part I’m watching is who gets paid for the power and the inference once it has to run.
Holding the names I already own. Not adding $NVDA on the beat. I’ll update if something specific comes off the call.
The $18.6B add is the scale. One quarter now bigger than the whole Data Center business was three years ago.
I don’t own $NVDA. The print is clean. The part I’m still watching is the lag. How fast does this kind of growth show up in the inference contracts the hyperscalers and neo-clouds already have, or does the capacity just get built further ahead of the actual usage.
$CBRS and $NBIS are the names I’m in. The GPU bar is the demand signal. The timing of when that turns into their revenue is the part that still has to show up. That’s what I’d want to hear more on.
$NVDA just printed $115.8B revenue, up 103%. Q4 guide $130B. No China compute revenue assumed.
$NBIS, $CBRS and $IREN are up after hours. The group is trading the print.
I own all three. $CBRS cost $141, last $103. $NBIS and $IREN are in the rockets book.
The demand signal is still there. The cut is the same one I’ve been using. Turn the megawatts into revenue before dilution or the cost of capital eats the upside. A strong GPU print can lift the names for a day. It doesn’t change which ones actually get paid.
Holding. Not adding on the bounce.
$NVDA revenue $115.8B, up 103%. Non-GAAP EPS $2.68. Data Center $107.6B, up 110%. Gross margin about 75%. Q4 guide $130B. No China Data Center compute revenue assumed. Vera Rubin NVL72 has shipped. Colette Kress is interim CEO.
I don’t own the stock. The print is the demand. The cut is the same one I’ve been using on the book. The leadership change and the China restriction are still the overhang. The numbers show the spend is still happening.
$CBRS just printed 70% core revenue growth and a Microsoft deal. $NBIS and $IREN are in the rockets book. The GPU is the capacity. The part I’m watching is who gets paid for the power and the inference once it has to run. Holding the names I already own. Not adding $NVDA on the beat.
$NBIS is in the rockets book.
The demand signal is still there. GPU pricing on new contracts is up more than 20% sequentially. Prepayments on new deals now cover more than 100% of the associated capex.
The selloff is about delivery and capital. Vineland delays. The ~$45B capex need for 2027. The market went from rewarding every new contract to asking whether the capacity shows up on time without constant dilution.
Holding. Not averaging down just because the price is lower. The cut is the same one. Turn the megawatts into revenue before the cost of capital eats the upside.
$CBRS is in the core book. Cost $141. Last $103.
Core revenue up 70% to $231M. Cloud up 272%. Q4 guide $310–330M. Three hyperscalers in six months.
Cloud revenue is guided flat in Q4. Not because demand slowed. Because data center capacity is the limit.
Holding. Not adding. The growth is real. The cut is whether the contracted megawatts turn into revenue without the buildout eating the cash flow.
$RKLB is in the core book. Cost $37. Last $55.
100th Electron is done. Fastest commercially developed rocket to the milestone. The next 100 are already booked.
Space Systems carried the last quarter. Neutron is still 2027.
Holding. Not adding on the launch count. The cut is whether the cadence turns into the backlog and the margin once Neutron is flying. Not the milestone by itself.
$IREN is deploying Lenovo ThinkSystem servers with $NVDA B200 and B300 GPUs in its Canadian data centers. Part of the AI cloud buildout. Renewable power behind it.
I own it in the rockets book. The hardware is the capacity. The cut is the same one. Turn the megawatts into revenue before dilution or the cost of capital eats the upside.
A server order is the build. It doesn’t answer the cash flow yet. That’s the number I’m still watching. Holding. Not adding on the announcement.📣
The $NVDA story is the governance and the China risk.
I don’t own the stock.
Chinese firms are already trying to lock in capacity outside China. Grey-market demand for restricted chips is up.
That demand doesn’t disappear. It just has to find somewhere else to run.
$CBRS just printed 70% core revenue growth and a Microsoft deal. $NBIS and $IREN are in the rockets book.
The cut is the same one I’ve been using. Turn the megawatts into revenue before the next raise or the next rule change.
A headline can reprice the GPU. It doesn’t rewrite the power and inference bill.
$CBRS core revenue up 70% to $231M. Cloud up 272%. Q4 guide $310–330M. Three hyperscalers in six months.
Cloud revenue is guided flat in Q4. Not because demand slowed. Because data center capacity is the limit.
I own it. Cost about $141. The growth is real. The cut is whether the contracted megawatts turn into revenue without the buildout eating the cash flow.
A third hyperscaler is a step. Capacity is still the bottleneck. That’s the number I’m watching.
The $1.5T is the spend. Neoclouds as the biggest bucket is the shift.
The reply on profit and ROIC is the right question. Capex isn’t a margin.
$NBIS ROIC is still negative. Free cash flow is deeply negative while they build. $IREN is still posting losses. Revenue is growing. The return on the capital is not there yet.
I own both in the rockets book. The cut is the same one. Turn the megawatts into revenue before dilution or the cost of capital eats the upside. A bigger capex chart can lift the group for a day. It doesn’t change which names actually get paid.
The $900B is the spend. It doesn’t tell you who keeps the cash.
I don’t own $NVDA. The leadership noise is the tape. The part I’m watching is where the money has to land after the GPU. Power, cooling, grid, the inference capacity that has to be built before any of it runs.
$CBRS just printed 70% core revenue growth and a Microsoft deal. $NBIS and $IREN are in the rockets book. The cut is the same one I’ve been using. Turn the megawatts into revenue before the next raise or the next rule change.
A bigger capex number can lift every name for a day. It doesn’t change which ones actually get paid. That’s the part I’d want to hear more on.
The $6–7 per GB cost versus ~$20 selling price is the margin print. 65% if it holds is the number that doesn’t look like a normal memory cycle.
The part that stood out more is the split. On the 32GB stack, base die, packaging and testing are about half the cost. The memory dies aren’t the whole bill. That’s where $TSM and $TER sit next to $MU.
I don’t own the memory name. The cut for me is whether the packaging and test content stays scarce as the stacks go from 8 to 16 layers, or whether that piece gets competed away once the capacity is there. Pricing can hold. The margin still depends on who owns the non-die cost. That’s the part I’d want to see in the next print.
$SPCX caught Starship with the tower arms after it returned from orbit. First time on the ship, not just the booster.
I own a small line. Cost $134. The catch is the reusability step. The booster missed this flight.
Holding. Not adding on the video. The cut is whether the catch becomes routine and the cost per flight actually drops. Not one successful recovery.
$RKLB just flew its 100th Electron. Fastest commercially developed rocket to the milestone. From New Zealand, for iQPS. The next 100 are already booked.
I own it. Cost $34 plus the 50 at $46. Electron is the cadence that already works. Space Systems carried the last quarter. Neutron is still 2027.
100 flights is the track record. The cut is whether that cadence turns into the backlog and the margin once Neutron is flying. Not the launch count by itself. Holding. Not adding on the milestone.
$CBRS is in the core book. Cost about $141.
Core revenue $231M, up 70%. Cloud $145M, up 272%. Q4 guide about $320M. They expect revenue to triple in 2027. Core gross margin 41%, cloud margin 47%. Q4 core margin guided near 48%.
Three hyperscalers in six months. AWS, Oracle, Microsoft. RPO $26.1B, mostly OpenAI. Oracle and Microsoft aren’t in that number yet.
The part I’m watching is the capacity. OpenAI deliveries are behind on land, power and construction. No contracts canceled. Cloud revenue is guided flat in Q4 because capacity, not demand, is the limit.
Holding. Not adding on the print. The growth and the margin expansion are real. The cut is whether the contracted megawatts turn into revenue without the buildout eating the cash flow. A third hyperscaler is a step. It doesn’t answer that yet.
Three hyperscalers in six months is the list. AWS, Oracle, Microsoft. RPO at $26.1B before those two even count.
The flat Q4 cloud guide is the part that matters more. Demand isn’t the constraint. Data center capacity is. That’s the physical limit showing up in the number.
I own $CBRS. Cost about $141. The growth is real. The cut is whether they can turn the contracted megawatts into revenue without the margin getting crushed by the buildout. Capacity is the bottleneck. It can also be the cost. That’s what I’m still watching.