The Markets Desk — Week 31 · Tue 28 Jul 2026

Markets

NVDA 196.98 +0.24%·AVGO 379.25 −1.04%·MU 814.51 −9.52%·AMAT 473.50 −8.39%·LNZA 5.67 +2.90%·ASTS 56.29 −3.43%·FN 443.75 −5.75%·MDT 86.84 +3.11%·TSHA 5.50 −0.72%·USAR 13.81 −7.50%·WULF 16.51 −8.18%·MNTS 4.56 −4.80%·LTBR 7.84 −2.24%·A 140.32 +1.76%·ADI 368.11 −1.02%·CRCL 64.27 −2.13%·DFDV 2.68 −1.83%·MAZE 28.23 −1.60%·PONY 7.46 +1.08%·BIDU 104.63 −0.63%·LUNR 12.38 −6.99%·CRWV 66.26 −6.40%·OKLO 39.13 −6.41%·NAKA 4.51 −3.84%·SNPS 384.14 −1.25%·XPEV 12.85 +0.55%·ARAY 0.25 +11.91%·CSCO 114.79 +0.19%·KEYS 302.52 −3.56%·MGX 1.23 +1.65%·

The Record14 callsFull track record →

Market Calls

EVERY THESIS TIMESTAMPED · SCORED WHEN RESOLVED · NO DELETIONS

Directional calls with conviction, timeframe, and a public grade. Analytics live on the Track Record — this page is the ledger itself.

14 total14 active0 validated0 invalidated

▲ LONG · —

Parameter Scaling Is Dead. Multi-Dimensional Scaling Is the New Paradigm.

Parameter scaling hit diminishing returns — GPT-4 to GPT-5 gap smaller than GPT-3 to GPT-4

●●●○ HIGH · 1YACTIVE · opened Apr 13

▲ LONG · —

Inference Costs Will Fall 90% by 2028 — The Memory Stack Is the Mechanism

Four independent memory stack improvements compound: HBM4 (2x) × KV compression (4-6x) × optical (70% energy) × architecture (2-10x)

●●●○ HIGH · 3YACTIVE · opened Apr 13

▲ LONG · —

BTC Miners Pivoting to AI Infrastructure: The Power Arbitrage

BTC miners own power-ready facilities — the scarcest asset in AI infrastructure

●●○○ MEDIUM · 1YACTIVE · opened Apr 13

◇ WATCH · —

Transformer Successors: State-Space Models and Linear Attention Are Worth Watching

Transformer attention scales O(n²) in memory — the fundamental cause of KV cache and HBM bottlenecks

●○○○ LOW · 3YWATCHING · since Apr 13

▲ LONG · LITE

Lumentum: The Optics Play That Bridges Power and Memory

60% of data center energy is data movement — Lumentum's optics solve this directly

●●●○ HIGH · 1YACTIVE · opened Apr 13 · −26.4%

▲ LONG · 000660.KS

HBM Suppliers Are the Picks-and-Shovels of AI: SK Hynix Leads

HBM4 entering mass production 2026 with 2TB/s bandwidth — doubling interface width to 2048-bit

●●○○ MEDIUM · 1YACTIVE · opened Apr 13

▲ LONG · —

KV Cache Compression Will Cut Inference Costs 50% Before HBM4 Ships at Scale

KV cache is THE bottleneck: 128K prompt on Llama 3.1-70B = 40GB HBM just for key-value storage

●●●○ HIGH · 1YACTIVE · opened Apr 13

The full reasoning behind every call — conviction, catalysts, price targets — is available to Pro members →

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 · Backtest.