NVIDIA silicon roadmap: Blackwell (2025) → Vera Rubin (2026) → Vera Rubin Ultra (2027) → Feynman (2028) — annual architectural cadence.
Predictor: Jensen Huang
Prediction text
NVIDIA silicon roadmap: Blackwell (2025) → Vera Rubin (2026) → Vera Rubin Ultra (2027) → Feynman (2028) — annual architectural cadence. | Vera Rubin → Feynman product milestones
Key catalyst: Vera Rubin → Feynman product milestones
Watch events: Vera Rubin sampling; Vera Rubin Ultra tape-out; Feynman disclosures.
Resolution evidence
Blackwell shipped 2025 as planned; Vera Rubin samples disclosed 2026; Vera Rubin Ultra and Feynman named on public roadmap.
Predictor: Jensen Huang
Calibration plot (stated vs observed)
Evidence about this node from Jensen Huang 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: gpu_roadmap_on_time_1y
GPU vendor product roadmap milestone (architecture / production / launch) hits within 1y of announced date
Tetlock-style outside view: at TRF=1 (just predicted), outside view dominates (w_in=0.3). At TRF=0 (deadline), inside view dominates (w_in=1.0). The blend regularizes overconfident inside views toward the historical base rate.
Probability over time
Milestone chain
- 2025-01-01 → 2025-12-31overdueBlackwell B200 / GB200 NVL72 generally available (2025 cohort)How: NVIDIA + hyperscalers publicly confirm GB200 NVL72 systems in commercial production / customer deploymentSource: https://www.tomshardware.com/pc-components/gpus/nvidia-announces-rubin-gpus-in-2026-rubin-ultra-in-2027-feynam-afterconf 99%
- 2025-09-21overdueQ1 window check-in (25%)
- 2026-03-01 → 2026-04-30overdueGTC 2026: Rosa CPU (successor to Vera) announced; Feynman 3D-stacked GPU details disclosedHow: NVIDIA at GTC 2026 publicly announces Rosa CPU successor and updated Feynman roadmap detailsSource: https://www.tomshardware.com/pc-components/gpus/nvidia-updates-data-center-roadmap-with-rosa-cpu-and-stacked-feynman-gpus-optical-nvlink-groq-lpus-with-nvfp4-and-nvlink-also-on-deckconf 85%
- 2026-06-11pendingQ2 window check-in (50%)
- 2026-06-01 → 2026-12-31pendingVera Rubin NVL144 launch (2026): 3.6 EFLOPS FP4, 1.2 EFLOPS FP8 inference/trainingHow: NVIDIA publicly ships Vera Rubin NVL144 with confirmed FP4 ≥3.6 EFLOPS / FP8 ≥1.2 EFLOPS specsSource: https://www.nextbigfuture.com/2025/03/nvidia-ultra-rubin-and-feynman-chips-and-data-center-roadmap.htmlconf 85%
- 2027-03-01pendingQ3 window check-in (75%)
- 2027-06-01 → 2027-12-31pendingRubin Ultra (2027): SXM8 with 4 reticle-limited GPU chiplets, 100 PFLOPS FP4, 1TB HBM4EHow: NVIDIA publicly ships Rubin Ultra with stated 4-chiplet socket, 100 PFLOPS FP4, 1TB HBM4ESource: https://www.tomshardware.com/tech-industry/semiconductors/nvidia-enterprise-roadmap-rubin-rubin-ultra-feynman-and-silicon-photonicsconf 70%
- 2028-06-01 → 2028-12-31pendingFeynman GPU launch (2028): advanced 3D stacking, custom HBM, Rosa CPU pairingHow: NVIDIA publicly ships Feynman GPU with 3D-stacked architecture; Rosa CPU GASource: https://www.tweaktown.com/news/110521/nvidia-updates-roadmap-with-new-details-on-its-next-gen-gpu-feynman-coming-in-2028/index.htmlconf 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": 0.653445934685756,
"kappa": 0.8077,
"base_rate": null,
"predictor": "Jensen Huang",
"total_llr": 1.6094379124341,
"bayesian_v2": true,
"prior_logit": 1.169934592028501,
"bayes_factor": "3.7:1 favoring",
"blend_reason": "no reference_class linked",
"inside_prior": 0.7631331927250531,
"kappa_source": "predictor_table",
"blend_applied": false,
"contributions": [
{
"llr": 1.6094379124341,
"kappa": 0.8077,
"label": "Roadmap remains intact and on cadence; Vera CPU + Rubin GPU launches affirm 2026 schedule.",
"adjusted_llr": 1.299943001873023
}
],
"evidence_kind": "intake_event_update",
"inside_source": "history_v2",
"inside_weight": 1,
"outside_weight": 0,
"posterior_prob": 0.9220029626744276,
"evidence_origin": "daily_intake",
"llm_suggestions": [
{
"polarity": "corroborates",
"status_change": "unchanged",
"evidence_strength": "strong",
"delta_prob_suggestion": 0.06
}
],
"posterior_logit": 2.469877593901523,
"predictor_brier": 0.01276,
"evidence_doc_ids": [
"1935"
],
"inside_posterior": 0.9220029626744276,
"blended_posterior": 0.9220029626744276,
"reference_class_id": null,
"total_adjusted_llr": 1.299943001873023,
"predictor_n_resolved": 8
}Raw metadata
{
"trf": 0.6664919347394406,
"kappa": 0.8077,
"base_rate": null,
"predictor": "Jensen Huang",
"total_llr": -1.2163953243244932,
"grace_days": 7,
"bayesian_v2": true,
"prior_logit": 1.4605894275995546,
"bayes_factor": "2.5:1 against",
"blend_reason": "no reference_class linked",
"inside_prior": 0.8116228096469142,
"kappa_source": "predictor_table",
"n_milestones": 3,
"blend_applied": false,
"contributions": [
{
"llr": -0.4054651081081644,
"kind": "llm_pre_event",
"kappa": 0.799623,
"label": "Blackwell B200 / GB200 NVL72 generally available (2025 cohort)",
"weight": 0.4,
"strength": "weak",
"confidence": 0.99,
"source_url": null,
"adjusted_llr": -0.3242192261407747,
"expected_date": "2025-07-02",
"measurement_criterion": "NVIDIA + hyperscalers publicly confirm GB200 NVL72 systems in commercial production / customer deployment"
},
{
"llr": -0.4054651081081644,
"kind": "quartile_checkpoint",
"kappa": 0.8077,
"label": "Q1 window check-in (25%)",
"weight": 0.05,
"strength": "weak",
"confidence": null,
"source_url": null,
"adjusted_llr": -0.32749416781896434,
"expected_date": "2025-09-21",
"measurement_criterion": null
},
{
"llr": -0.4054651081081644,
"kind": "llm_pre_event",
"kappa": 0.686545,
"label": "GTC 2026: Rosa CPU (successor to Vera) announced; Feynman 3D-stacked GPU details disclosed",
"weight": 0.4,
"strength": "weak",
"confidence": 0.85,
"source_url": null,
"adjusted_llr": -0.2783700426461197,
"expected_date": "2026-03-31",
"measurement_criterion": "NVIDIA at GTC 2026 publicly announces Rosa CPU successor and updated Feynman roadmap details"
}
],
"evidence_kind": "metadata_milestone_miss_sweep",
"inside_source": "history_v2",
"inside_weight": 0.5334556456823916,
"outside_weight": 0.46654435431760843,
"posterior_prob": 0.629601118599425,
"posterior_logit": 0.5305059909936958,
"predictor_brier": 0.01276,
"inside_posterior": 0.629601118599425,
"blended_posterior": 0.629601118599425,
"reference_class_id": null,
"total_adjusted_llr": -0.9300834366058588,
"predictor_n_resolved": 8
}Network propagation neighbors
Top incoming (parents)
Edges that influence THIS node's belief
| Kind | Node | Their prob | P(c|s=T) | P(c|s=F) | Δ implied |
|---|---|---|---|---|---|
| killer | TK02 AI Compute Supply Shock (TSMC/Taiwan Disruption) | 12.0% | 0.050 | 0.880 | -0.090 |
Top outgoing (children)
Predictions THIS node influences
No outgoing edges.
Ticker exposure
Beneficiaries (22)
Prerequisites (1)
| Type | Pred | Title | Domain | Lag |
|---|---|---|---|---|
| killer | TK02 | AI Compute Supply Shock (TSMC/Taiwan Disruption) | — | — |
Dependents (0)
| Type | Pred | Title | Domain | Lag |
|---|---|---|---|---|
| No dependents | ||||
Expected milestones (1)
| Expected by | Description | Status |
|---|---|---|
| 2028-12-31 | [Capability 2028-12] [CMQ_026] Vera Rubin sampling; Vera Rubin Ultra tape-out; Feynman disclosures. [SEM_005] Stargate site groundbreaking announcements; OpenAI infrastructure spend disclosures [CMQ_009] FDA-approved autonomous di | pending |
Validations (1)
| Observed at | Status | By | Notes |
|---|---|---|---|
| 2026-04-29 | partial | thesis_timeline_v1.0_import | Blackwell shipped 2025 as planned; Vera Rubin samples disclosed 2026; Vera Rubin Ultra and Feynman named on public roadmap. |
Linked documents (10)
| Sim | Source | Title | Market prob | Polarity | Reviewed | Published |
|---|---|---|---|---|---|---|
| 0.659 | gdelt | 10729180 | — | mentions | pending | 2026-04-30 |
| 0.656 | github_release | facebookresearch/faiss v1.14.2 | — | mentions | pending | 2026-05-22 |
| 0.646 | github_release | pytorch/pytorch v2.11.0 | — | mentions | pending | 2026-03-23 |
| 0.646 | gdelt | — | mentions | pending | 2026-04-30 | |
| 0.645 | github_release | pytorch/pytorch v2.7.0 | — | mentions | pending | 2025-04-23 |
| 0.645 | github_release | facebookresearch/faiss v1.8.0 | — | mentions | pending | 2024-03-01 |
| 0.643 | github_release | facebookresearch/faiss v1.12.0 | — | mentions | pending | 2025-08-12 |
| 0.641 | github_release | google-deepmind/alphafold v2.2.0 | — | mentions | pending | 2022-03-10 |
| 0.639 | github_release | pytorch/pytorch v2.8.0 | — | mentions | pending | 2025-08-06 |
| 0.639 | github_release | facebookresearch/neuroai v0.2.2 | — | mentions | pending | 2026-05-26 |
Raw metadata
{
"nia": false,
"qty": "annual cadence",
"mode": "FORECAST",
"role": "Cited-Executive",
"caveats": "Execution risk at every node; TSMC advanced-node capacity is binding constraint.",
"context": "Compressed cadence vs historical 2-3 year cycles; forces hyperscalers into perpetual refresh to maintain lowest cost per token.",
"to_year": 2028,
"conv_cues": "explicit named roadmap; CEO",
"direction": "HAPPEN",
"from_year": 2025,
"timeframe": "2025-2028",
"conv_level": "HIGH",
"milestones": [
{
"kind": "llm_pre_event",
"label": "Blackwell B200 / GB200 NVL72 generally available (2025 cohort)",
"source": "https://www.tomshardware.com/pc-components/gpus/nvidia-announces-rubin-gpus-in-2026-rubin-ultra-in-2027-feynam-after",
"status": "overdue",
"weight": 0.4,
"ordinal": -7,
"source_id": null,
"confidence": 0.99,
"expected_date": "2025-07-02",
"miss_emitted_at": "2026-05-02T22:07:21.384228+00:00",
"miss_emitted_by": "metadata_milestone_sweep",
"research_origin": "training",
"expected_date_range": {
"to": "2025-12-31",
"from": "2025-01-01"
},
"measurement_criterion": "NVIDIA + hyperscalers publicly confirm GB200 NVL72 systems in commercial production / customer deployment"
},
{
"kind": "quartile_checkpoint",
"label": "Q1 window check-in (25%)",
"status": "overdue",
"weight": 0.05,
"ordinal": -6,
"source_id": null,
"expected_date": "2025-09-21",
"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": "GTC 2026: Rosa CPU (successor to Vera) announced; Feynman 3D-stacked GPU details disclosed",
"source": "https://www.tomshardware.com/pc-components/gpus/nvidia-updates-data-center-roadmap-with-rosa-cpu-and-stacked-feynman-gpus-optical-nvlink-groq-lpus-with-nvfp4-and-nvlink-also-on-deck",
"status": "overdue",
"weight": 0.4,
"ordinal": -5,
"source_id": null,
"confidence": 0.85,
"expected_date": "2026-03-31",
"miss_emitted_at": "2026-05-02T22:07:21.384228+00:00",
"miss_emitted_by": "metadata_milestone_sweep",
"research_origin": "deep_research",
"expected_date_range": {
"to": "2026-04-30",
"from": "2026-03-01"
},
"measurement_criterion": "NVIDIA at GTC 2026 publicly announces Rosa CPU successor and updated Feynman roadmap details"
},
{
"kind": "quartile_checkpoint",
"label": "Q2 window check-in (50%)",
"status": "pending",
"weight": 0.05,
"ordinal": -4,
"source_id": null,
"expected_date": "2026-06-11",
"observed_date": null
},
{
"kind": "llm_pre_event",
"label": "Vera Rubin NVL144 launch (2026): 3.6 EFLOPS FP4, 1.2 EFLOPS FP8 inference/training",
"source": "https://www.nextbigfuture.com/2025/03/nvidia-ultra-rubin-and-feynman-chips-and-data-center-roadmap.html",
"status": "pending",
"weight": 0.4,
"ordinal": -3,
"source_id": null,
"confidence": 0.85,
"expected_date": "2026-09-15",
"research_origin": "deep_research",
"expected_date_range": {
"to": "2026-12-31",
"from": "2026-06-01"
},
"measurement_criterion": "NVIDIA publicly ships Vera Rubin NVL144 with confirmed FP4 ≥3.6 EFLOPS / FP8 ≥1.2 EFLOPS specs"
},
{
"kind": "quartile_checkpoint",
"label": "Q3 window check-in (75%)",
"status": "pending",
"weight": 0.05,
"ordinal": -2,
"source_id": null,
"expected_date": "2027-03-01",
"observed_date": null
},
{
"kind": "llm_pre_event",
"label": "Rubin Ultra (2027): SXM8 with 4 reticle-limited GPU chiplets, 100 PFLOPS FP4, 1TB HBM4E",
"source": "https://www.tomshardware.com/tech-industry/semiconductors/nvid
... (truncated)