research:8f34
Raised datacenter mix assumption
assumption changed in NVDA v3
Private beta
Version control for investment research.
subThesis is an agentic investment research workspace. It tracks your assumptions, evidence, decisions, and outcomes as each thesis evolves — while AI agents, tailored to how you invest, work in the background, scenarios branch and run in parallel, and company financials stay current in real time. Every revision is sourced, timestamped, and reviewable, so your process compounds instead of disappearing into notes and memory.
Coming soonSee how it works
AssumptionsEvidenceDecisionsOutcomes
overview.subthesis
Overview
5 positions · 117 research commits · last synced 2m ago
+1.38%
portfolio · today
PositionLastTodayConviction
NVDANVIDIAthesis v3
$168.42+2.14%78
ASMLASML Holdingthesis v5
$942.10+0.87%71
TSMTaiwan Semithesis v2
$204.55-1.02%64
MSFTMicrosoftthesis v7
$488.31+0.41%82
AMDAdvanced Microthesis v4
$176.09-0.63%55
NVDA - new evidence committed to thesis v3 after the latest 10-Q.
Research version control
The whole investment flow, versioned over time.
subThesis records the chain from assumption to evidence to decision to outcome. You can inspect what changed, who changed it, what source supported it, and whether reality later confirmed or broke the call.
Assumptions
What must be true for the thesis to work.
Evidence
The filing, transcript, note, or datapoint behind each claim.
Decisions
Why you bought, added, trimmed, held, or changed conviction.
Outcomes
What actually happened, tied back to what you believed then.
audit-log.subthesis
append-onlyAssumptionEvidenceDecisionOutcome
evidence:41ac
Linked Q1 10-Q segment disclosure
source added to revenue claim
decision:b911
Held position after earnings
conviction +6 with memo attached
outcome:2d07
Compared forecast to realized margin
miss logged against base case
compare decision:b911 against outcome:2d07
See the exact evidence available at the decision point, then review what the market and company reported afterward.
Versioned thinking
Every belief, versioned over time.
Investment theses evolve. subThesis tracks every version — what you believed, when you believed it, and why it changed. Look back at v3 of your NVDA thesis and see exactly what evidence shifted your conviction.
thesis.subthesis
NVDA· $168.42thesis v3
Accelerated-compute demand stays supply-constrained through 2027.
Conviction
78 / 100
Hyperscaler capex guidance continues to outrun foundry capacity. Lead times on datacenter GPUs extended again this quarter.
changed in v3
Datacenter revenue mix rose to 92% — gaming is no longer material to the thesis.
Primary risk: a capex air-pocket if AI training ROI disappoints. See scenario branch capex-peak-2026.
Evidence
Truth grounded in sources.
Every claim in your thesis is anchored to filings, transcripts, and your own research notes. Memos compound into a permanent record of your thinking — append-only, never overwritten, always citable. And everything is stored as plain, open Markdown, so your research stays portable and never locked into a proprietary format.
memos.subthesis
Memos
NVDA · append-only · 41 entries
May 14, 2026filing
Q1 datacenter mix at 92%
Segment disclosure confirms the shift. Re-rated conviction +6.
NVDA 10-Q · p.12
May 02, 2026transcript
Supply commentary from earnings call
CFO: lead times "extended versus last quarter" — capacity still the binding constraint.
Q1 call transcript · 18:40
Apr 21, 2026note
Own read on hyperscaler capex
Cross-checked three cloud capex guides. All raised. Demand side intact.
Research note · my analysis
Timeline
Assumptions to outcomes, on one timeline.
See thesis versions, sourced evidence, buy or hold decisions, and realized outcomes on a single time axis. Pan into the past to understand what you knew then. Pan forward to see which assumptions reality confirmed, contradicted, or left unresolved.
timeline.subthesis
Research timeline · NVDA
past| nowforecast
Assumptions
v1v2v3
Evidence
10-Qcall10-Q
Decisions
BUYHOLD
Outcomes
confirmedcontradictedforecast
NovJanMarMayJulSep
Pan left to see the assumptions, evidence, and decisions that built conviction. Pan right to compare those calls against realized outcomes.
Holdings
See what your holdings are actually doing.
subThesis reads your position and trade activity — entries, adds, trims, and sizing — alongside the thesis behind each move. Every decision is tied back to the reasoning that drove it, so you can spot drift between what you believe and how you're actually positioned.
overview.subthesis
Overview
5 positions · 117 research commits · last synced 2m ago
+1.38%
portfolio · today
PositionLastTodayConviction
NVDANVIDIAthesis v3
$168.42+2.14%78
ASMLASML Holdingthesis v5
$942.10+0.87%71
TSMTaiwan Semithesis v2
$204.55-1.02%64
MSFTMicrosoftthesis v7
$488.31+0.41%82
AMDAdvanced Microthesis v4
$176.09-0.63%55
NVDA - new evidence committed to thesis v3 after the latest 10-Q.
AI research agents
AI agents that work while you don't.
Pick a research task. The agent reads the 10-Q, scans monitored sources, executes the analysis, and lands a structured finding in your inbox. Some tasks run once. Some run forever — like watching a competitor's IR page every week, pinging you only when something changes.
Each agent is an AI tailored to your workspace and deeply personalized to your flow — it knows your theses, your sources, and how you reason, so every finding lands in your context. You decide what to track, what to investigate, and what to believe. The agent handles the legwork.
tasks.subthesis
Research tasks
3 agents activeQueued2
TSM
Re-underwrite on Q1 10-Q
triggered by filing
AMD
Segment-margin breakdown
one-time
Running2
NVDA
Read 10-Q, draft thesis update
62% · ~8m left
ASMLloop
Watch competitor IR page
weekly
Landed2
NVDA
Datacenter-mix finding
conviction +6
MSFT
Capex cross-check
no change
Earnings auto-reunderwrite
10-Q lands. Thesis update drafted within 30 minutes.
Source monitoring
Substack, X, IR pages, RSS — surface what's relevant to your positions.
Hypothesis testing
Track speculative claims separately. Accept them into the thesis when evidence accumulates.
Continuous insights
Confirmed, contradicted, missed — see how reality is testing your thesis in real time.
Gaps
Find the holes before the market does.
Agents read your thesis against the evidence and flag what's missing — an untested assumption, a claim with no source, a risk you never priced. Gaps surface as work to do, not surprises after the fact, so the weak points in your reasoning are visible while you can still act on them.
memos.subthesis
Memos
NVDA · append-only · 41 entries
May 14, 2026filing
Q1 datacenter mix at 92%
Segment disclosure confirms the shift. Re-rated conviction +6.
NVDA 10-Q · p.12
May 02, 2026transcript
Supply commentary from earnings call
CFO: lead times "extended versus last quarter" — capacity still the binding constraint.
Q1 call transcript · 18:40
Apr 21, 2026note
Own read on hyperscaler capex
Cross-checked three cloud capex guides. All raised. Demand side intact.
Research note · my analysis
Data layer
Real data. Real filings. No black boxes.
subThesis indexes SEC filings, earnings transcripts, and IR materials for every tracked position and keeps them current in real time. Explore company financials as they update — search across line items, drill into segment data, model scenarios with reverse-DCF and sensitivity tables. Your research is grounded in primary source documents — structured for how you actually think.
resources.subthesis
Source documents · NVDA
10-Qindexed
Q1 FY26 Quarterly Report
May 2026
CALLindexed
Q1 FY26 Earnings Transcript
May 2026
10-Kindexed
FY25 Annual Report
Feb 2026
IRindexed
Investor Day — slide deck
Jan 2026
datacenter revenue
Q1 FY26 · pulled from 10-Q
Datacenter revenue$39.1B+18%
Gross margin74.2%+90bps
Gaming revenue$2.9B-4%
Operating margin62.0%+120bps
Scenarios
Branch your thesis. Stress-test the future.
Branch your main thesis into scenarios and run them in parallel. What if AI capex peaks in 2025? What if China demand drops 10%? Each branch evolves on its own — its own price target, conviction level, and per-event interpretation — letting you carry and stress-test multiple futures at once without losing your conviction in any single one.
scenarios.subthesis
Scenarios · NVDA
4 branches · 1 basemainBase case$210
Supply-constrained through 2027.
conviction
78primary
capex-peak-2026AI capex peaks 2026$132
Training ROI disappoints; orders soften.
conviction
41carried in parallel
china-demand-cutChina demand −10%$188
Export curbs tighten; mix absorbs most of it.
conviction
55carried in parallel
sovereign-buildoutSovereign AI buildout$256
National compute programs add a demand leg.
conviction
62carried in parallel
Your edge is your process. Make it durable.
subThesis builds the memory for you — what you thought, why you thought it, and what changed your mind, all kept and sourced. Six months in, you have a structured archive of your thinking. Two years in, it's irreplaceable.
Coming soon