Market calls / Timestamped · scored on resolution · never deleted
Directional calls with a date on them, graded when they resolve.
Conviction, timeframe and a public grade on every one.
- Calls
- 13
- Live
- 12
- Graded
- 1
All time
0 validated
The constellation
The Book as one argumentThe four things the calls are about (memory, silicon, power, models) are floors; each call is a light above its floor, higher for more conviction. The ring round each light is its falsifier, one arc per leg: red has fired, copper has partly fired, dashed can’t be checked, grey is quiet. Dotted lines show which other floors a call leans on, and the calls tagged with the house thesis thread to the centre. Lights dim when a call has gone a month without a review. Drag to turn; hover a light; click to open its call.
Hover or tap
Each light is a call; its ring is the legs that would break it
Higher is more conviction. Click a light to open its call.
- KV cache before HBM4 · THEME · Memory · conviction medium · legs: HBM4 ships at scale, and cheaply, before compression does (partly fired); Labs stop shipping aggressive KV quantisation on quality (not read yet) · last reviewed 7 Sept 2026
- Lumentum · LITE · Silicon · conviction medium · thesis broke · legs: Co-packaged optics misses production at scale inside the window (fired); The copper runway extends and the urgency goes (partly fired) · last reviewed 9 Oct 2026
- Hybrid architectures · THEME · Models · conviction medium · legs: No hybrid or state-space model reaches frontier quality (not fired); A pure state-space model replaces the Transformer outright (not fired) · last reviewed 7 Sept 2026
- Nuclear scales · THEME · Power · conviction medium · legs: Most firm datacenter baseload since April is not nuclear, by end-2028 (partly fired); The Crane (TMI-1) restart is abandoned or slips past 2028 (not fired) · last reviewed 11 Sept 2026
- Energy is the moat · THEME · Power · conviction high · legs: Power stops being scarce: PPAs sign at ordinary rates (not fired); Efficiency cuts demand faster than supply tightens (not fired) · last reviewed 7 Sept 2026
- Miners turn to AI · THEME · Power · conviction high · legs: The miners still earn most of their revenue from mining (not fired); They win hosting contracts and cannot hold them (can't be checked) · last reviewed 7 Sept 2026
- Inference costs −90% · THEME · Models · conviction medium · legs: One constraint gates two or more of the four layers (partly fired); Providers keep the savings instead of passing them on (partly fired) · last reviewed 7 Sept 2026
- Past parameter scaling · THEME · Models · conviction high · legs: A step change comes from a bigger dense model (not fired); Inference-time compute stops paying for its memory (not fired) · last reviewed 7 Sept 2026
- NVIDIA · NVDA · Memory · conviction medium · legs: The financing circle closes: marks, privates and neoclouds crack together (not fired); Two of the top three customers pause or build their own (not fired) · last reviewed 9 Oct 2026
- Micron · MU · Memory · conviction medium · legs: The contracts stop being contracts into 2027 (not fired) · last reviewed 9 Oct 2026
- SK Hynix · SK HYNIX · Memory · conviction medium · legs: The HBM rent is competed away (Samsung yields, NVIDIA allocation) (not fired); The Nasdaq listing is shelved or withdrawn (can't happen now) · last reviewed 9 Oct 2026
- Broadcom · AVGO · Silicon · conviction medium · legs: The premium never compresses while the thesis compounds (not fired); A hyperscaler re-bids or in-sources an XPU programme (not fired) · last reviewed 9 Oct 2026
- TSMC · TSM · Silicon · conviction high · legs: Leading-edge or CoWoS capacity moves into surplus (not fired); The discount closes because its customers de-rate (can't be checked) · last reviewed 9 Oct 2026
The instrument wall
One small instrument per callThe fastest way to scan the whole Book. Each instrument shows its falsifier lamps, its conviction since it was published (copper dots are changes, hollow dots are reviews that held it), its weekly line if it has a ticker, how much of its window has gone, and how long since anyone looked. Sort it to ask a question: which calls are under pressure, which are overdue a review, which close first.
Sort
- LITEbullish · mediumLumentumfired · partly firedConviction medium: 2 changes and 4 reviews that held it since 13 Apr 2026.17 days since a reviewbroke 9 Oct
+24.6% weekly close vs the call-day price, USD, at resolution
- THEMEbullish · mediumInference costs −90%partly fired · partly firedConviction medium: 1 change and 1 review that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2029 - THEMEbullish · mediumKV cache before HBM4partly fired · not read yetConviction medium: 1 change and 1 review that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2027 - THEMEbullish · mediumNuclear scalespartly fired · not firedConviction medium: 0 changes and 3 reviews that held it since 13 Apr 2026.
No price. Graded on its legs alone.
45 days since a reviewends 13 Apr 2029 - THEMEbullish · highMiners turn to AInot fired · can't be checkedConviction high: 2 changes and 1 review that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2027 - TSMbullish · highTSMCnot fired · can't be checkedConviction high: 0 changes and 4 reviews that held it since 9 Jun 2026.17 days since a reviewends 9 Jun 2029
−2.2% weekly close vs the call-day price, USD
- THEMEwatching · mediumHybrid architecturesnot fired · not firedConviction medium: 1 change and 1 review that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2029 - THEMEbullish · highEnergy is the moatnot fired · not firedConviction high: 1 change and 2 reviews that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2029 - THEMEbullish · highPast parameter scalingnot fired · not firedConviction high: 1 change and 2 reviews that held it since 13 Apr 2026.
No price. Graded on its legs alone.
49 days since a reviewends 13 Apr 2027 - NVDAbullish · mediumNVIDIAnot fired · not firedConviction medium: 0 changes and 3 reviews that held it since 9 Jun 2026.17 days since a reviewends 9 Jun 2027
+1.3% weekly close vs the call-day price, USD
- MUbullish · mediumMicronnot firedConviction medium: 0 changes and 3 reviews that held it since 9 Jun 2026.17 days since a reviewends 9 Jun 2027
−6.2% weekly close vs the call-day price, USD
- SK HYNIXbullish · mediumSK Hynixnot fired · can't happen nowConviction medium: 0 changes and 4 reviews that held it since 9 Jun 2026.17 days since a reviewends 9 Jun 2027
−42.3% weekly close vs the call-day price, KRW
- AVGOwatching · mediumBroadcomnot fired · not firedConviction medium: 0 changes and 3 reviews that held it since 9 Jun 2026.17 days since a reviewends 9 Jun 2027
−18.0% weekly close vs the call-day price, USD
These screens never show a price target, an entry, a size or a position. A move is the weekly close against the price on the day the call was published, in that ticker’s own currency, shown for the record and not as the grade. Nothing here is an instruction to buy or sell.
| Stance | Thesis |
|---|---|
| LONG | TSMC: own the bottleneck, not the brandOwn the bottleneck, not the brand: the chokepoint every AI chip depends on, at a discount to all of them.TSM · High · 3YActive · opened Jun 9−2.2% to last Friday close |
| WATCH | Broadcom: the right custom-silicon thesis at the wrong priceThe right custom-silicon thesis at the wrong price — watching the most expensive, lowest-quality way to play it.AVGO · Med · 1YWatching · since Jun 9 |
| LONG | SK Hynix: the largest HBM rent, valued like a Korean cyclicalThe largest slice of the HBM scarcity rent, valued like a Korean cyclical — with a Nasdaq listing as the re-rating catalyst.000660.KS · Med · 1YActive · opened Jun 9−42.3% to last Friday close |
| LONG | Micron: priced like a cyclical at the top, contracted like a utilityPriced like a peak-cycle memory-cyclical, contracted like a utility — a re-rating bet, not an earnings bet.MU · Med · 1YActive · opened Jun 9−6.2% to last Friday close |
| LONG | NVIDIA: paying a market multiple for a franchise financing its own demandStill constructive on the best franchise in AI infrastructure — but the risk has migrated from the income statement to the balance sheet.NVDA · Med · 1YActive · opened Jun 9+1.3% to last Friday close |
| LONG | Parameter Scaling Is Dead. Multi-Dimensional Scaling Is the New Paradigm.AI scaling has moved from raw parameter count to four memory-hungry dimensions — inference-time compute, MoE, data curation, architecture — making memory bandwidth the universal bottleneck.High · 1YActive · opened Apr 13 |
| LONG | Inference Costs Will Fall 90% by 2028 — The Memory Stack Is the MechanismFour independent memory-stack improvements — HBM4, KV-cache compression, optical interconnect, and architecture — could compound to a ~10x fall in per-token inference cost by 2028.Med · 3YActive · opened Apr 13 |
| LONG | BTC Miners Pivoting to AI Infrastructure: The Power ArbitrageBitcoin miners already hold the power, cooling, and space AI needs — mining revenue is expected to fall from 85% to under 20% of total as they convert to AI hosting.High · 1YActive · opened Apr 13 |
| LONG | Energy Ownership Is the New AI MoatPower, not GPUs or algorithms, is AI's real moat — a location-specific power contract is scarce and non-fungible in a way chip access never is.High · 3YActive · opened Apr 13 |
| LONG | Nuclear Is the Only Power Source That Scales for AI Data CentersAI data centers need reliable, carbon-free baseload at massive scale — nuclear is the only source that delivers all three, and owners can charge a premium for it.Med · 3YActive · opened Apr 13 |
| WATCH | Transformer Successors: State-Space Models and Linear Attention Are Worth WatchingTransformers' quadratic memory scaling is the KV-cache problem. State-space and linear-attention models scale linearly — hybrids blending both look like the likely path.Med · 3YWatching · since Apr 13 |
| LONG | KV Cache Compression Will Cut Inference Costs 50% Before HBM4 Ships at ScaleKV caches are AI inference's biggest bottleneck. Software compression (6x cited) is shipping faster than HBM4 hardware — and will drive the cost collapse of 2026-2027.Med · 1YActive · opened Apr 13 |
Page 1 / 2Older calls
How calls are graded
A call locks at publication — direction, conviction, timeframe and target are timestamped and can’t be edited. Resolution is mechanical: the stated direction against the realized move. Accuracy breakdowns, calibration and the resolved feed live on the track record. Method details: methodology · the open book.