This page covers the memory bandwidth and persistent data capacity that make AI clusters usable. High-bandwidth memory, or HBM, is stacked DRAM mounted next to an accelerator so the chip can read model weights and activations quickly. DRAM feeds CPUs and servers, NAND flash and solid-state drives move hot datasets, and nearline hard disk drives store large model, training, backup, and inference data. These parts sit inside the accelerator package, on server boards, in storage appliances, and across cloud storage fleets. The node converts AI capex into company economics only when hyperscalers, AI-cloud providers, server OEMs, storage buyers, and device makers pay for capacity while suppliers show revenue, gross margin, inventory discipline, and free cash flow. The current basket read is supplier-led but timing-sensitive: MU and SNDK are the cleanest memory and flash capture routes, STX and WDC are mass-capacity storage routes, and reviewed tails such as RMBS, NTAP, SIMO, FORM, and PSTG/P stay outside the ranked basket until local coverage, node ownership, or source depth improves.
What the stack is: memory and storage are the cluster capacity layers that hold model weights, activations, training data, checkpoints, logs, backups, and inference data close enough to compute for AI workloads to run.
What it does: HBM and DRAM feed processors during training and inference; NAND and SSDs serve fast reads and writes for active data; nearline HDDs keep large datasets available at lower cost per terabyte.
Main pieces: HBM stacks, DRAM modules, NAND flash, enterprise SSDs, storage controllers, nearline HDDs, drive firmware, storage enclosures, cloud storage software, and inventory tied to wafer starts and capacity agreements.
Where it sits: HBM sits inside the accelerator package beside the compute die; DRAM sits on the server board; SSDs sit in servers or storage shelves; HDDs sit in scale-out storage racks and cloud data centers.
How the parent theme uses it: AI infrastructure sends data from storage into server memory and accelerator HBM, writes checkpoints back to storage, and keeps inference data available across racks and regions.
Terms used later: ASP means average selling price; bit shipments measure memory capacity shipped; exabytes measure large storage capacity; nearline HDD means high-capacity hard drives used in data centers; free cash flow tests whether higher revenue becomes cash after capex and working capital.
Report boundary: this node is a tactical report layer. It ranks the current basket, static setup labels, confirmation triggers, invalidation levels, and chart provenance. Durable research, claim IDs, raw source registries, and company underwriting stay in the linked knowledge pages. Setup levels use completed weekly bars from read-only discovery daily_ohlc through 2026-06-05; the local table has daily rows through 2026-06-12.
Current Setup
Memory capacity gateAI clusters need bandwidth first and cheap capacity next.
HBM and DRAM decide compute utilization, while NAND, SSDs, and nearline HDDs decide whether large datasets, checkpoints, and inference records can move and persist economically.
Positive proofMU and SNDK have the cleanest supplier capture.
Micron shows direct memory economics; Sandisk gives a cleaner NAND and flash route after the WDC separation.
Conversion gatePricing must become margin and cash.
Watch bit shipments, ASPs, exabytes, gross margin, inventory, capex, and free cash flow.
Primary constraintDownstream hardware has to absorb the bill.
Server OEM, EMS, AI-cloud, and device margins can weaken if component cost rises faster than revenue.
Demand sourceAI capex + cloud storage
ComponentsHBM, DRAM, NAND, SSD, HDD
Accounting proofRevenue, GM, FCF
Live watchASP, exabytes, inventory
Memory bandwidth and storage capacity decide whether purchased accelerators become usable AI clusters. HBM and DRAM feed compute, NAND and SSDs move hot data, and nearline HDDs hold large model, training, backup, and inference datasets. The theme benefits when supplier bit shipments, ASPs, exabytes, gross margin, inventory discipline, and free cash flow improve while server OEMs, EMS providers, cloud buyers, and device makers can still absorb the bill of materials.
The Information Technology lane still shows an AI-infrastructure-led cycle, with data-center systems, memory, semiconductors, servers, storage, optical networking, and implementation services carrying the strongest evidence. Bounded official checks show MU Q2 FY2026 revenue of $23.86B, SNDK Q3 FY2026 revenue of $5.95B, STX Q3 FY2026 free cash flow of $953M, and WDC Q3 FY2026 free cash flow of $978M. The next positive proof is fresh HBM allocation, DRAM/NAND ASP, datacenter NAND, nearline exabyte, gross-margin, inventory, and free-cash-flow evidence from the next earnings and filing cycle.
The setup weakens if memory and storage pricing rolls over before volume broadens, or if tight supply lifts supplier revenue while downstream server, AI-cloud, EMS, and device margins compress. The four core tickers have no maintained local security lanes, SNDK has only short post-separation chart history, and reviewed tails either belong to other nodes or lack local chart coverage. Static setup thresholds use completed weekly bars through 2026-06-05, while local daily OHLC now has rows through 2026-06-12, so refresh setup levels before using thresholds as current trading evidence.
Static setup labels use completed weekly bars aggregated from discovery daily_ohlc through 2026-06-05. Local daily_ohlc rows now run through 2026-06-12 for MU, SNDK, STX, and WDC, and selected-security right-rail charts request as_of=latest; refresh setup levels before using the older thresholds as current trading evidence.
Basket
This ranked basket is inherited from the AI Capex parent report, the memory/component theme route, official company checks, read-only discovery coverage, and a 10-primary plus merge subagent ranking pass. Ranking uses node exposure and source-backed business mechanism first, then technical timing. MU ranks first because it is the broadest direct memory route. SNDK ranks second because the separation created a cleaner NAND/flash route but the trading history is short. STX and WDC rank third and fourth because they are storage routes with stronger nearline HDD exposure than memory exposure. RMBS, NTAP, SIMO, FORM, and PSTG/P were reviewed as tails; they stay outside the ranked basket because they are interface IP, storage systems, SSD controllers without local OHLC coverage, semicap/probe exposure, or a ticker/coverage mismatch.
MUMicron Technology
Broad direct memory supplier across HBM, DRAM, NAND, and data-center SSDs.
Market cap$1.107T
Next earningsJun 24, 2026
Latest qtr revenue$23.86B
Role in stack
Micron sells HBM, DRAM, NAND, and SSD products to cloud, AI-server, mobile, client, automotive, and embedded customers. Theme pressure becomes economics through bit shipments, ASPs, HBM qualification, gross margin, inventory control, capex discipline, and customer commitments.
Revenue mix
Q2 FY2026 Cloud Memory revenue was $7.749B and Core Data Center revenue was $5.687B, about 56% of total revenue together. Local knowledge does not maintain a MU security lane or HBM mix table.
Latest qtr revenue
Q2 FY2026 total revenue was $23.86B for the quarter ended February 26, 2026, from Micron's March 18, 2026 earnings release. The next check is Q3 revenue, HBM allocation, DRAM/NAND ASPs, and FCF.
SNDKSandisk
Cleaner NAND and flash route after the WDC separation, with short trading history.
Market cap$293.2B
Next earningsNot confirmed
Latest qtr revenue$5.950B
Role in stack
Sandisk sells NAND flash, data-center SSD, edge, embedded, and consumer flash products. AI and device storage demand becomes economics through NAND ASPs, exabytes, datacenter mix, multiyear customer agreements, gross margin, and cash conversion.
Revenue mix
Q3 FY2026 revenue was Datacenter $1.467B, Edge $3.663B, and Consumer $820M. Datacenter was about 25% of revenue, so the node route is cleaner than WDC for flash but still not pure data-center exposure.
Latest qtr revenue
Q3 FY2026 total revenue was $5.950B, from Sandisk's April 30, 2026 earnings release. The next earnings date was not confirmed in bounded company/Nasdaq checks on June 13, 2026.
STXSeagate Technology
Mass-capacity HDD route for cloud and enterprise data retention.
Market cap$208.8B
Next earningsNot confirmed
Latest qtr revenue$3.112B
Role in stack
Seagate sells mass-capacity hard drives and related storage products to cloud, enterprise, and OEM buyers. AI model storage, training data, inference records, and backups become economics through nearline exabytes, high-capacity drive pricing, HAMR/Mozaic execution, gross margin, and FCF.
Revenue mix
Seagate's Q3 FY2026 10-Q showed Data Center at 80% of revenue and nearline HDD shipments of 175.4 exabytes out of 199.4 total exabytes. There is no maintained local STX security lane.
Latest qtr revenue
Q3 FY2026 total revenue was $3.112B for the quarter ended April 3, 2026, from Seagate's April 28, 2026 earnings release. The next earnings date was not confirmed in bounded company/Nasdaq checks on June 13, 2026.
WDCWestern Digital
Post-separation HDD and cloud-storage route, no longer the NAND/flash route.
Market cap$194.0B
Next earningsNot confirmed
Latest qtr revenue$3.337B
Role in stack
WD sells hard disk drives and storage infrastructure to cloud service providers, hyperscalers, enterprises, and channel buyers. AI data growth becomes economics through high-capacity HDD demand, cloud revenue, pricing, gross margin, free cash flow, and customer visibility.
Revenue mix
Western Digital's Q3 FY2026 10-Q showed Cloud revenue of $2.972B, or about 89% of total revenue. The Sandisk separation means WDC should be treated as the HDD/cloud storage route rather than the flash route.
Latest qtr revenue
Q3 FY2026 total revenue was $3.337B, from WD's April 30, 2026 earnings release. The next earnings date was not confirmed in bounded company/Nasdaq checks on June 13, 2026.
This ranked screen is an exposure map, not a recommendation list. Market caps use local discovery close times weighted shares outstanding on 2026-06-12. Company underwriting stays out of the report layer; no maintained local security lanes exist for MU, SNDK, STX, or WDC.
What Confirms Or Weakens
AreaWhat confirmsWhat weakens or invalidatesWatch next
01Node thesis
Capacity becomes usable AI infrastructure
What confirms
HBM, DRAM, NAND, SSD, and HDD suppliers show rising revenue, ASPs, exabytes, gross margin, and cash while AI capex and server shipments remain funded.
What weakens or invalidates
Memory or storage pricing rolls over before volume growth broadens, or shortages delay AI cluster delivery and transfer too much cost into downstream hardware margins.
Watch next
Micron HBM and data-center mix
Sandisk Datacenter and Edge
Seagate nearline exabytes
WD Cloud revenue
02Supplier economics
Pricing becomes margin and cash
What confirms
Official releases and filings show higher bit shipments, higher ASP per bit or exabyte, firm customer agreements, controlled inventory, operating cash flow, and free cash flow.
What weakens or invalidates
Inventory builds, receivable stress, price cuts, capex that outruns demand, customer deferrals, or margin guidance cuts reduce supplier capture.
Watch next
ASP and bit shipments
Exabytes shipped
Gross margin
Inventory days
FCF after capex
03Customer absorption
Downstream margins hold
What confirms
Server OEMs, EMS providers, hyperscalers, AI-cloud buyers, and device makers keep shipping systems while hardware gross margin and cash conversion hold.
What weakens or invalidates
OEMs or EMS providers report bill-of-materials pressure, storage delays, backlog conversion issues, or customer pushouts tied to memory and storage costs.
Watch next
DELL and SMCI server margin
CLS and JBL conversion
AI server pricing
Device unit commentary
04Funding and supply
Capacity does not overbuild
What confirms
Customer commitments, multiyear agreements, prepayments, disciplined wafer starts, and capex plans line up with memory and storage demand.
What weakens or invalidates
Suppliers add capacity ahead of binding demand, customers delay commitments, export controls disrupt shipments, or financing and working capital absorb the revenue uplift.
Watch next
Long-term agreements
Customer concentration
Capex and wafer starts
China/export updates
05Stale condition
Refresh trigger
What confirms
Discovery daily_ohlc remains current, official releases and 10-Q source checks still cover the latest material company events, and no new memory pricing update changes the read.
What weakens or invalidates
A new earnings release, 10-Q, pricing update, customer contract disclosure, export-policy update, or supply-chain update arrives before this page is refreshed.
Watch next
Discovery coverage
MU and SNDK releases
STX and WDC filings
PSTG/P or SIMO coverage repair
Source Trail
Canonical Thesis
AI Capex Cycle for the cross-sector capex chain, proof burden, and supplier routing.
Micron's Investor Relations events page, checked June 13, 2026, confirms June 24, 2026 for Micron's Third Quarter 2026 Financial Call. Nasdaq's public earnings page returned no usable date in the bounded check.
Sandisk, Seagate, and Western Digital had no company-confirmed next earnings date found in bounded company IR, company release, and Nasdaq earnings-page checks on June 13, 2026. The Basket therefore says Not confirmed for SNDK, STX, and WDC.
Market caps use read-only discovery daily_ohlc latest close on 2026-06-12 multiplied by instruments.weighted_shares_outstanding: MU $1.107T, SNDK $293.2B, STX $208.8B, and WDC $194.0B.
Latest-quarter revenue sources: Micron Q2 FY2026 revenue $23.86B from its March 18, 2026 release; Sandisk Q3 FY2026 revenue $5.950B from its April 30, 2026 release; Seagate Q3 FY2026 revenue $3.112B from its April 28, 2026 release; WD Q3 FY2026 revenue $3.337B from its April 30, 2026 release.
Discovery And Chart Provenance
The chart API reads daily_ohlc from the local discovery store and returns weekly packages using first open, maximum high, minimum low, final close, and summed volume.
Charts show 20w_ema, weekly_volume, and 20w_average_volume. MU, STX, and WDC also show 100w_ema. EMAs are computed from weekly closes using full available weekly history before visible-window clipping.
Static setup labels use completed weekly bars through 2026-06-05. Local daily OHLC rows through 2026-06-12: MU 1,278 rows, SNDK 328 rows, STX 1,278 rows, and WDC 1,278 rows. Latest stored daily rows are newer than the static setup thresholds, so refresh setup levels before using the thresholds as current trading evidence.
Completed-week closes and EMAs through 2026-06-05: MU $864.01 / 20W EMA $560.35 / 100W EMA $272.82; SNDK $1,559.32 / 20W EMA $1,009.50 / no 100W EMA; STX $847.47 / 20W EMA $588.90 / 100W EMA $289.44; WDC $511.72 / 20W EMA $370.84 / 100W EMA $179.98.
Representative selected-security API check on June 13, 2026 for /api/securities/MU/chart?frequency=weekly&window=3y&as_of=latest resolved through the local report API with as_of_date and resolved_as_of of 2026-06-12, source table reports.weekly_ohlc+discovery.daily_ohlc, weekly rows, 20-week moving-average data, 100-week EMA data, and weekly volume fields.
Ticker-level daily OHLC lineage still shows May 11, 2026 per-ticker ingestion runs with source freshness of 2026-05-08, while the canonical daily_ohlc table has stored rows through 2026-06-12. Treat the stored table as the chart source and refresh lineage before using this page for a new market session.
Universe reference from the parent AI Capex report: 5,274 active common stocks and 4,991 with at least 64 daily bars and latest OHLC coverage through 2026-06-05.
Known Gaps
No maintained local security lanes exist for MU, SNDK, STX, or WDC, so this node does not store company underwriting.
SNDK has post-separation discovery history starting 2025-02-24; the page therefore excludes a 100W EMA for SNDK and treats the chart as a short-history signal.
RMBS, NTAP, SIMO, FORM, and PSTG/P were reviewed as tails. RMBS and NTAP are chartable but not direct memory or storage-media suppliers; FORM belongs mainly to the semicap/probe route; SIMO and PSTG/P need local ticker coverage repair before they can support right-rail charts under this node.
Price and volume context supports setup labels, triggers, and invalidation levels only. It does not establish why a stock moved without same-window earnings, filing, pricing, customer, policy, or supply evidence.