ChipStack tracks the physical and silicon chokepoints behind trillion-dollar AI clusters: compute, HBM, custom silicon, foundry capacity, grid buildout, power/cooling, and network fabric.
What moved?
Only thesis-relevant tape makes the cut.
What broke?
Separate price damage from bottleneck damage.
What would change our mind?
Every brief ends with triggers, not vibes.
Free gets the signal. Alpha gets the decisions: daily briefings, simulated model portfolio stance, closest-to-action queue, and position movement notes.
Recent read
Power stack was the live bottleneck
GEV / ETN / PWR weakness was treated as evidence to investigate, not a forced action.
Recent read
GEV moved into the watch queue
Nearest evidence gate: orders, backlog, or hyperscaler capex commentary confirming power demand.
Recent read
HBM stayed evidence-gated
Memory weakness needed pricing, yield, or demand confirmation before changing the bottleneck read.
Ignored noise
ANET did not change the map
A small networking move was ignored until supported by cluster-utilization or switching evidence.
LIVE INTELLIGENCE
The homepage pulls from the same morning brief, 13F monitor, and weekly digest that feed ChipStack. If the snapshot is stale, the page labels it as latest available instead of pretending it is live.
$263.51
$903.61
$1806.37
$471.19
Grid and power is the bottleneck setting the tone today. The market is again paying up for the physical side of AI factories: turbines, switchgear, transmission, cooling, and the contractors that turn hyperscaler capex into energized megawatts.
Comparing filing windows: 5 Jun 2026 → 7 Jun 2026
Notable exits
ABERDEEN_GROUP_PLC appears across 11 tracked AI infrastructure names.
WEEKLY DIGEST
Portfolio value
-8.54% on the week
Latest weekly digest summarizes portfolio performance, tier rotation, and AI infrastructure thesis signals.
Start here
You get one clean AI infrastructure read each day and a weekly stack review. If it sounds like market wallpaper, unsubscribe. If it helps you see where the bottleneck is moving, Alpha gives you the full decision layer.
What lands free
Daily signal
The one thing in the stack worth noticing today.
Weekly read
What led, what lagged, and whether it mattered.
Alpha preview
Enough to judge the work before paying for it.
What stays paid
Get the free ChipStack note
No broad AI takes. No magic-model performance claims. Just the daily bottleneck signal and the weekly evidence trail.
Want to see what paid members get first? View a sample Alpha brief before you subscribe.
Who should read it
ChipStack is for investors who want to know which part of the AI buildout is becoming scarce, mispriced, or no longer special.
Good fit
Bad fit
AI BOTTLENECK MAP
Seven places the AI buildout can jam. Free gets the public read; Alpha gets the stance, triggers, and portfolio consequences.
Where demand first shows up. If GPU availability loosens faster than workloads arrive, the whole trade changes.
Compute: demand evidence is still better than the tape, so the public read is opportunity-with-confirmation-needed rather than panic.
Flow: Quiet
The memory gate. Watch pricing, yields, and guidance before treating a selloff as harmless tape noise.
HBM: keep the layer on watch; price weakness needs pricing, yield, or demand confirmation before it becomes thesis damage.
Flow: Quiet
The cost curve layer. Custom silicon matters when hyperscalers need cheaper inference, not better slide decks.
Custom ASIC: weak tape against intact evidence keeps MRVL/AVGO in the inference-cost follow-up lane.
Flow: Quiet
The fabrication and packaging choke. If capacity cannot clear, demand migrates into pricing power.
Foundry: mixed evidence makes capacity commentary more important than one-day price action.
Flow: Quiet
The least fashionable constraint and often the hardest to solve: turbines, transformers, substations, permits, and crews.
Grid: physical buildout stress is the signal to verify through orders, backlog, and capex language.
Flow: Quiet
Dense clusters fail without power distribution and thermal control. This is where capex becomes plumbing.
Datacenter power: weak tape plus strong evidence keeps cooling and electrical gear in the opportunity-to-verify bucket.
Flow: Quiet
The utilization layer. Compute is worth less when the fabric cannot keep the cluster fed.
Networking: do not over-read small moves until cluster-utilization or switching commentary confirms a bottleneck shift.
Flow: Quiet
If the free note earns trust, upgrade.
Free signal first. Alpha when you want the triggers, stance, and watchlist.
Pricing
Free is the public signal. Alpha is the decision layer: daily briefings, simulated portfolio stance, closest-to-action queue, position movement notes, and event-driven thesis updates.
Public signal layer
A daily read and weekly recap. Enough to judge whether ChipStack thinks clearly before you pay.
Founding decision layer
For investors who want the whole AI bottleneck map, not just the public read. Founding access keeps the lower rate as the product expands.
Founding tier closes after the first 50 members
Later price
Same product after founding pricing closes. Nothing extra, just less early-user risk discount.
FAQ