Hyperscaler capital expenditure specifically on memory architecture leaps from approximately 8% of total AI capex (2023) to 30% by end-2026 — High Bandwidth Memory (HBM) and DRAM face severe structural supply constraints with DRAM prices more than doub...
Predictor: Morgan Stanley
Prediction text
Hyperscaler capital expenditure specifically on memory architecture leaps from approximately 8% of total AI capex (2023) to 30% by end-2026 — High Bandwidth Memory (HBM) and DRAM face severe structural supply constraints with DRAM prices more than doubling. | Q2 2026 memory-maker earnings
Key catalyst: Q2 2026 memory-maker earnings
Watch events: DRAMeXchange spot prices; HBM allocation; Micron/Samsung/SK Hynix quarterly guidance
Resolution evidence
HBM4 pricing, DRAM spot prices up 80%+ YoY Q4-2025 → Q1-2026; Micron/Samsung/SK Hynix record capex guidance for agent-era memory.
Predictor: Morgan Stanley
Calibration plot (stated vs observed)
Evidence about this node from Morgan Stanley is multiplied by κ in /api/intake. Lower κ = less weight; floors at 0.10 (effectively silenced) and caps at 1.00 (full weight).
Reference class
This node isn't linked to a reference class. The Bayesian update applies without outside-view blending.
Probability over time
Milestone chain
- 2026-03-05overdueQ1 window check-in (25%)
- 2026-04-03hitMemory consumes ~30% of hyperscaler AI capex in 2026 confirmedHow: SemiAnalysis or equivalent firm publicly confirms memory accounts for ~30% of total hyperscaler AI capex in calendar 2026 (4x increase vs 2023)Source: Tom's Hardware — Memory Will Consume 30% of Hyperscaler AI Spending This Yearconf 99%Notes: HIT — SemiAnalysis confirmed 30% of hyperscaler capex going to memory in 2026, exactly matching prediction. 4x the ~8% 2023 baseline.
- 2026-05-07overdueQ2 window check-in (50%)
- 2026-07-09pendingQ3 window check-in (75%)
- 2026-07-15 → 2026-08-31pendingQ2 2026 memory-maker earnings confirm 30%+ memory capex shareHow: SK Hynix, Samsung, Micron Q2 2026 earnings disclose hyperscaler memory orders ≥30% of total AI capex per customer disclosureSource: ChronicleJournal/TokenRing — The AI Memory Supercycle: Micron Shatters Records as HBM Capacity Sells Out Through 2026conf 90%Notes: Direct measurement of headline criterion.
- 2026-04-01 → 2026-12-31pendingNVIDIA secures preferential HBM supply terms below market ratesHow: Industry reporting confirms NVIDIA receives HBM supply at below-spot pricing through long-term agreementsSource: Tom's Hardware — Memory Will Consume 30% of Hyperscaler Spending (Nvidia preferential supply)conf 85%
- 2026-09-01 → 2026-12-31pendingDRAM prices more than double in 2026 vs 2025 baselineHow: DRAM ASP in Q4 2026 ≥2.0x Q4 2025 baseline per TrendForce/SemiAnalysis trackingSource: SemiAnalysis via Tom's Hardware / Cloud News — Memory Threatens to Devour Hyperscale Cloud Spendingconf 85%
- 2026-11-01 → 2027-03-31pendingHyperscaler total capex in 2026 reaches ~$600-725 billionHow: Big Five hyperscalers combined 2026 capex disclosed at ≥$600B (36%+ YoY growth)Source: Yahoo Finance — Magnificent 7 earnings reveal AI spending surge, hyperscaler capex set to reach $725B in 2026conf 90%
- 2027-01-01 → 2027-12-31pendingMemory pricing surge triggers 2027 industry margin compression and EPS missesHow: Hyperscaler 2027 industry median gross margin compresses ≥5pp; consensus EPS miss ≥10% per Morgan Stanley frameworkSource: Morgan Stanley framework via Random Tech — How AI Is Reshaping Global Memory Supply Chainconf 55%
What if this resolves?
Click a button to clamp this prediction and run a Gibbs sample. Returns the predictions whose marginals shift most. ~30s per run; ideal for stress-testing "if X resolves, what else moves?"
Evidence chain
Raw metadata
{
"trf": 0.6293214636488222,
"kappa": 0.5833,
"base_rate": null,
"predictor": "Morgan Stanley",
"total_llr": -0.4054651081081644,
"grace_days": 7,
"bayesian_v2": true,
"prior_logit": 1.029158575771784,
"bayes_factor": "1.3:1 against",
"blend_reason": "no reference_class linked",
"inside_prior": 0.7367527356802764,
"kappa_source": "predictor_table",
"n_milestones": 1,
"blend_applied": false,
"contributions": [
{
"llr": -0.4054651081081644,
"kind": "quartile_checkpoint",
"kappa": 0.5833,
"label": "Q2 window check-in (50%)",
"weight": 0.05,
"strength": "weak",
"confidence": null,
"source_url": null,
"adjusted_llr": -0.2365077975594923,
"expected_date": "2026-05-07",
"measurement_criterion": null
}
],
"evidence_kind": "metadata_milestone_miss_sweep",
"inside_source": "history_v2",
"inside_weight": 0.5594749754458244,
"outside_weight": 0.44052502455417564,
"posterior_prob": 0.6884002204839857,
"posterior_logit": 0.7926507782122918,
"predictor_brier": 0.01,
"inside_posterior": 0.6884002204839857,
"blended_posterior": 0.6884002204839857,
"reference_class_id": null,
"total_adjusted_llr": -0.2365077975594923,
"predictor_n_resolved": 1
}Raw metadata
{
"trf": 0.6650500679109432,
"kappa": 0.5833,
"base_rate": null,
"predictor": "Morgan Stanley",
"total_llr": -0.4054651081081644,
"grace_days": 7,
"bayesian_v2": true,
"prior_logit": 1.265666373331276,
"bayes_factor": "1.3:1 against",
"blend_reason": "no reference_class linked",
"inside_prior": 0.78,
"kappa_source": "predictor_table",
"n_milestones": 1,
"blend_applied": false,
"contributions": [
{
"llr": -0.4054651081081644,
"kind": "quartile_checkpoint",
"kappa": 0.5833,
"label": "Q1 window check-in (25%)",
"weight": 0.05,
"strength": "weak",
"confidence": null,
"source_url": null,
"adjusted_llr": -0.2365077975594923,
"expected_date": "2026-03-05",
"measurement_criterion": null
}
],
"evidence_kind": "metadata_milestone_miss_sweep",
"inside_source": "prior_prob",
"inside_weight": 0.5344649524623397,
"outside_weight": 0.4655350475376603,
"posterior_prob": 0.7367527356802764,
"posterior_logit": 1.0291585757717838,
"predictor_brier": 0.01,
"inside_posterior": 0.7367527356802764,
"blended_posterior": 0.7367527356802764,
"reference_class_id": null,
"total_adjusted_llr": -0.2365077975594923,
"predictor_n_resolved": 1
}Network propagation neighbors
No propagation data yet. Run inference/.venv/bin/python scripts/ops/run_loopy_belief_propagation.py on the droplet, or wait for the Sunday 02:00 UTC weekly cron.
Ticker exposure
Beneficiaries (13)
Prerequisites (0)
| Type | Pred | Title | Domain | Lag |
|---|---|---|---|---|
| No prerequisites | ||||
Dependents (0)
| Type | Pred | Title | Domain | Lag |
|---|---|---|---|---|
| No dependents | ||||
Validations (1)
| Observed at | Status | By | Notes |
|---|---|---|---|
| 2026-04-29 | partial | thesis_timeline_v1.0_import | HBM4 pricing, DRAM spot prices up 80%+ YoY Q4-2025 → Q1-2026; Micron/Samsung/SK Hynix record capex guidance for agent-era memory. |
Linked documents (1)
| Sim | Source | Title | Market prob | Polarity | Reviewed | Published |
|---|---|---|---|---|---|---|
| 0.595 | gdelt | vig vs hdv growing income 155000719.html | — | mentions | pending | 2026-04-30 |
Raw metadata
{
"nia": false,
"qty": "8% → 30% CapEx; DRAM 2x",
"mode": "FORECAST",
"role": "Cited-Firm",
"context": "Memory-wall thesis extension of CMQ_028. Positions Micron / Samsung / SK Hynix as 'arbiters of agentic deployment'. Couples with SEM_021 (100x model size), SEM_022 (quantization architectures).",
"to_year": 2026,
"conv_cues": "specific % shift; specific price-doubling claim",
"direction": "UP",
"from_year": 2026,
"timeframe": "by end-2026",
"conv_level": "HIGH",
"milestones": [
{
"kind": "quartile_checkpoint",
"label": "Q1 window check-in (25%)",
"status": "overdue",
"weight": 0.05,
"ordinal": -6,
"source_id": null,
"expected_date": "2026-03-05",
"observed_date": null,
"miss_emitted_at": "2026-05-02T22:07:21.384228+00:00",
"miss_emitted_by": "metadata_milestone_sweep"
},
{
"kind": "llm_pre_event",
"label": "Memory consumes ~30% of hyperscaler AI capex in 2026 confirmed",
"notes": "HIT — SemiAnalysis confirmed 30% of hyperscaler capex going to memory in 2026, exactly matching prediction. 4x the ~8% 2023 baseline.",
"source": "Tom's Hardware — Memory Will Consume 30% of Hyperscaler AI Spending This Year",
"status": "hit",
"weight": 0.4,
"ordinal": -5,
"source_id": null,
"confidence": 0.99,
"source_url": "https://www.tomshardware.com/tech-industry/memory-will-consume-30-percent-of-hyperscaler-spending-this-year",
"expected_date": "2026-04-03",
"observed_date": "2026-04-03",
"research_origin": "deep_research",
"measurement_criterion": "SemiAnalysis or equivalent firm publicly confirms memory accounts for ~30% of total hyperscaler AI capex in calendar 2026 (4x increase vs 2023)"
},
{
"kind": "quartile_checkpoint",
"label": "Q2 window check-in (50%)",
"status": "overdue",
"weight": 0.05,
"ordinal": -4,
"source_id": null,
"expected_date": "2026-05-07",
"observed_date": null,
"miss_emitted_at": "2026-05-15T22:14:51.696830+00:00",
"miss_emitted_by": "metadata_milestone_sweep"
},
{
"kind": "quartile_checkpoint",
"label": "Q3 window check-in (75%)",
"status": "pending",
"weight": 0.05,
"ordinal": -3,
"source_id": null,
"expected_date": "2026-07-09",
"observed_date": null
},
{
"kind": "llm_pre_event",
"label": "Q2 2026 memory-maker earnings confirm 30%+ memory capex share",
"notes": "Direct measurement of headline criterion.",
"source": "ChronicleJournal/TokenRing — The AI Memory Supercycle: Micron Shatters Records as HBM Capacity Sells Out Through 2026",
"status": "pending",
"weight": 0.4,
"ordinal": -2,
"source_id": null,
"confidence": 0.9,
"source_url": "https://markets.chroniclejournal.com/chroniclejournal/article/tokenring-2025-12-30-the-ai-memory-supercycle-micron-shatters-records-as-hbm-capacity-sells-out-through-2026",
"expected_date": "2026-08-07",
"research_origin": "deep_research",
"expected_date_range": {
"to": "2026-08-31",
"from": "2026-07-15"
},
"measurement_criterion": "SK Hynix, Samsung, Micron Q2 2026 earnings disclose hyperscaler memory orders ≥30% of total AI capex per customer disclosure"
},
{
"kind": "llm_pre_event",
"label": "NVIDIA secures preferential HBM supply terms below market rates",
"source": "Tom's Hardware — Memory Will Consume 30% of Hyperscaler Spending (Nvidia preferential supply)",
"status": "pending",
"weight": 0.4,
"ordinal": -1,
"source_id": null,
"confidence": 0.85,
"source_url": "https://www.tomshardware.com/tech-industry/memory-will-consume-30-percent-of-hyperscaler-spending-this-year",
"expected_date": "2026-08-16",
"research_origin": "deep_research",
"expected_date_range": {
"to": "2026-12-31",
... (truncated)