Your investment research, in one place
Follow the AI roadmap.
From expert views to business evidence. See who could benefit, what must happen next, and what would change the thesis.
Editorial review 2026-10-03 · Next review due 2026-11-03
A route can succeed while its stocks lose money. These are conditional watch candidates: current valuation, your risk budget and existing exposure still need review. No automatic trades.
Five routes, three validation stages
Stages describe dependencies, not promised dates or assigned probabilities. Routes can develop together.
Infrastructure
Paid deployment
Real-world transfer
Demand evidenced · valuation open
Compute, interconnects & memory
NVDA · AVGO · MU
More useful inference and parallel agents could raise compute and memory demand, even as each task gets cheaper. Total paid workload must grow faster than efficiency improves.
What would strengthen it
Over the next two reported quarters, compare AI revenue, margins and customer spending with the prior guidance.
What would weaken it
Spending remains high while utilization, pricing or cash generation weaken; customers reduce orders.
Track paid inference growth and margins together. Favor demonstrable economics over capacity announcements.
Inspect the original views (4)
Amodei expects very powerful AI within a few years and discusses how capability may spread through the economy.
A lab leader’s expectation is not independent confirmation. Technical capability and widespread adoption are different questions.
Reviewed view · Watch source videoBrown explores parallel agent work and the challenges of coordinating many contributors.
Research examples do not establish gains for every workflow or agent count.
Reviewed view · Watch source videoAtallah and Midha challenge the idea of a single model winning every use case.
A marketplace perspective does not establish the best model for your task.
Reviewed view · Watch source videoKarpathy describes agents as a decade-long engineering effort, pointing to gaps including continual learning.
This is a view recorded in October 2025, not a current capability test or a resolved AGI deadline.
Reviewed view · Watch source videoNVDA NVIDIA
Priority researchAccelerated compute and networking
Reported Data Center revenue growth; this establishes current demand, not a cheap share price.
Next check: Data Center growth, gross margin, customer concentration and cash conversion.
Downside: Custom chips, export limits and customers building ahead of demand.
Q2 FY2027 · Issuer source
AVGO Broadcom
Priority researchCustom accelerators and interconnects
Reported growth in custom AI accelerators and networking revenue. Customer concentration still matters.
Next check: AI revenue, next-quarter guidance, program timing and customer breadth.
Downside: Few large customers; internal chip programs can shift spending.
Q3 FY2026 · Issuer source
MU Micron
Wait for confirmationHigh-bandwidth memory and memory capacity
The September 30 release reports strong AI-led demand and record results. Peak earnings are not normalized earnings.
Next check: HBM mix, DRAM supply, inventory days, capital spending and through-cycle cash flow.
Downside: Memory supply can overshoot demand; falling prices compress peak margins.
Q4 FY2026 · Issuer source
Demand evidenced · valuation open
Power delivery & cooling
VRT · ETN
If compute use expands, physical power delivery and heat removal can limit deployment. This is our economic inference, not a stock recommendation from the speakers.
What would strengthen it
At the next two reports, verify backlog conversion, cancellation rates where disclosed, and cash conversion.
What would weaken it
Backlogs rise but deliveries and cash flow stall, or data-center customers delay projects.
Watch delivery schedules and whether revenue converts into cash.
Inspect the original views (2)
Amodei expects very powerful AI within a few years and discusses how capability may spread through the economy.
A lab leader’s expectation is not independent confirmation. Technical capability and widespread adoption are different questions.
Reviewed view · Watch source videoBrown considers how more capable models might organize increasingly large groups of agents.
This is a forward-looking scenario, not evidence of autonomous firms operating reliably.
Reviewed view · Watch source videoVRT Vertiv
Priority researchPower and thermal management
Reported rising data-center demand and raised full-year guidance. Orders still need to convert to cash.
Next check: Orders, backlog conversion, organic growth, working capital and free cash flow.
Downside: Construction delays, order cancellations and capacity expansion costs.
Q2 2026 · Issuer source
ETN Eaton
Priority researchElectrical distribution and power equipment
Reported growth in electrical orders and backlog; data centers are one driver, not the whole company.
Next check: Electrical orders, book-to-bill, delivery capacity and segment margins.
Downside: Industrial cyclicality, grid delays and a valuation that already assumes growth.
Q2 2026 · Issuer source
Paid adoption to verify
Enterprise agents that earn their keep
MSFT · NOW
Distribution, workflow data and reliable recovery can turn model capability into paid work. Existing software benefits only if new usage exceeds price and seat erosion.
What would strengthen it
Check disclosed paid adoption, renewals and incremental AI revenue against serving and review costs.
What would weaken it
Demos grow while paid retention stays weak, or agent revenue merely replaces higher-margin seat revenue.
Track specific paid tasks and renewals instead of counting launched agents.
Inspect the original views (5)
Almeida argues that AI can strengthen existing software by becoming a small, programmable part of it.
A product thesis does not establish adoption, reliability or an advantage for every software business.
Reviewed view · Watch source videoWeinstein emphasizes agents retrying and debugging after problems, beyond merely starting a task.
A vendor’s description is not an independent comparison across tasks.
Reviewed view · Watch source videoShihipar discusses software that users can continually reshape with agents.
A customization vision is not evidence that ongoing maintenance disappears.
Reviewed view · Watch source videoKarpathy describes agents as a decade-long engineering effort, pointing to gaps including continual learning.
This is a view recorded in October 2025, not a current capability test or a resolved AGI deadline.
Reviewed view · Watch source videoThe discussion highlights distribution and access as important parts of the AI market.
Distribution is a business thesis, not evidence of demand or a durable advantage.
Reviewed view · Watch source videoMSFT Microsoft
Priority researchCloud distribution and enterprise workflows
Reported Azure growth and paid Copilot adoption. Revenue must be weighed against infrastructure spending.
Next check: Azure growth, paid usage, AI monetization, capex and free cash flow after investment.
Downside: High capital intensity; monetization can lag infrastructure spending.
Q4 FY2026 · Issuer source
NOW ServiceNow
Wait for confirmationEnterprise workflow and agent orchestration
Reported AI annual contract value and subscription growth. Contract value is not recognized revenue.
Next check: AI contract conversion, cRPO, renewals and net gains after seat cannibalization.
Downside: Agents can replace seats as well as sell new products; valuation sensitivity.
Q2 2026 · Issuer source
Paid adoption to verify
Permission, identity & trusted deployment
PANW · MSFT
Agents acting across systems create permission, monitoring and incident-response needs. Demand for safeguards does not guarantee a separate paid market for every vendor.
What would strengthen it
Look for attributable paid deployments, organic security growth and renewal evidence; separate acquisitions.
What would weaken it
Controls become free bundles, adoption stalls or losses from incidents exceed the expected business value.
Verify customers pay for controls rather than receiving them in existing bundles.
Inspect the original views (4)
Huang argues that agentic systems should operate with restrictive rights.
Restricting access reduces exposure but cannot establish that an entire system is safe.
Reviewed view · Read original interviewHuang frames AI safety as an engineering problem that can be addressed through system design.
This is an executive’s position, not proof that all safety problems are solved.
Reviewed view · Read original interviewKvist argues that trust and accountability can constrain adoption even when AI capabilities improve.
An assurance vendor’s perspective does not prove its service resolves every adoption barrier.
Reviewed view · Watch source videoKvist emphasizes testing failure cases beyond a successful demonstration.
Passing a test set does not guarantee safe behavior outside its coverage.
Reviewed view · Watch source videoPANW Palo Alto Networks
Wait for confirmationIdentity, cloud and AI security
Reports security platform revenue and ARR. These do not isolate revenue from AI-agent security.
Next check: Organic next-generation security ARR, GAAP profit, cash flow and acquisition effects.
Downside: Bundling by cloud vendors, dilution and acquisitions obscuring organic growth.
Q4 FY2026 · Issuer source
MSFT Microsoft
Priority researchCloud distribution and enterprise workflows
Reported Azure growth and paid Copilot adoption. Revenue must be weighed against infrastructure spending.
Next check: Azure growth, paid usage, AI monetization, capex and free cash flow after investment.
Downside: High capital intensity; monetization can lag infrastructure spending.
Q4 FY2026 · Issuer source
Longer-term research
World models & physical AI
GOOGL · NVDA
Interactive simulation could support creation, training and robotics. A technical leap still needs reliable real-world transfer and paying users.
What would strengthen it
Require repeat external use, comparable real-world evaluations and separately evidenced economics before upgrading the thesis.
What would weaken it
Simulation improves while real-world reliability, costs or commercial retention fail to improve.
Advance only when simulation benefits a measured real task.
Inspect the original views (4)
Germanidis discusses real-time video and world models as a direction for interactive experiences.
A research direction does not establish production reliability or demand.
Reviewed view · Watch source videoGermanidis emphasizes references, steerability and iteration within creative workflows.
Creative controls do not guarantee originality, consistency or a commercially usable result.
Reviewed view · Watch source videoThe discussion examines the difficulty of transferring simulated success to complex real-world tasks.
A simulation gap can vary by task; it does not show that all benchmarks are useless.
Reviewed view · Watch source videoThe panel debates whether research and engineering bottlenecks could constrain recursive AI improvement.
The participants disagree; this is not a shared timetable or a settled result.
Reviewed view · Watch source videoGOOGL Alphabet
ExploreWorld models and model distribution
DeepMind demonstrated an interactive world model. This is dated technical evidence, not verified world-model revenue.
Next check: External paid adoption, cost per useful session and real-world transfer tests.
Downside: A convincing demo may not produce a profitable business; Alphabet is not a pure-play.
Genie 3 · 2025-08-05 · Issuer source
NVDA NVIDIA
Priority researchAccelerated compute and networking
Reported Data Center revenue growth; this establishes current demand, not a cheap share price.
Next check: Data Center growth, gross margin, customer concentration and cash conversion.
Downside: Custom chips, export limits and customers building ahead of demand.
Q2 FY2027 · Issuer source
Keep only the stocks you want to follow
Add a candidate above. Your selection stays in this browser; it does not alter the fixed 12-stock portfolio or configure Telegram alerts.
Your existing 12-stock experiment
Open returns and toolsLoading the latest complete published valuation…
Fixed October 2, 2026 NY-close entry; equal-weight model, not actual holdings. SPY is an S&P 500 ETF proxy. TQQQ targets daily 3× exposure. The roadmap does not rebalance this record.
Evidence that could change the route
SunWatch
The existing private owner channel checks roadmap revisions every 30 minutes. Related company disclosures remain in the disclosure monitor. New video titles are review leads, not investment verdicts.
/roadmap · /roadmap_pause · /roadmap_resume
Owner integration only. Public visitors do not receive these alerts by opening the bot or saving a local watchlist.
Open SunWatch botCompany announcements
Open original releases, read what changed, then compare the route’s confirmation and failure conditions. Announcement metadata alone does not establish better earnings.
Open SunWatch evidence monitorNVDA, MU, MSFT, NOW, PANW and GOOGL use the existing company monitor. AVGO, VRT and ETN have direct issuer links here; automatic company-release coverage is not yet connected.
New videos to review (100)
Matched by title only. No transcript judgment or automatic stock recommendation. Publisher language is retained. Source check: 2026-10-25T07:49:32Z
Your Agent Fix Just Broke Something Else - Sofia Jakovcevic, Arize AI 2026-10-24
Why Inference Matters 12x More Than Training | YC Paper Club 2026-10-24
AI:AM Highlights: Clean Neurons, Black October, the GPU Credit Crunch, and Buying the Supply Chain 2026-10-24
Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca 2026-10-23
NVIDIA Halos SAFE: Autonomous Vehicle Safety Evaluation From Road Testing to Simulation 2026-10-23
Accelerating Robot Development With AI Agents and NVIDIA Isaac 2026-10-23
Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg 2026-10-23
How AT&T is building AI-powered legal workflows with OpenAI 2026-10-23
Why Compute Still Trades in Group Chats — Affaan Mustafa & Alejandro Soumah Fernandez, Jump Trading 2026-10-23
Buying GPUs Should Work Like Buying Stocks — Carmen Li, Compute Exchange 2026-10-23
How insect intelligence can inspire more efficient autonomous robots 2026-10-23
Self-replicating AI, your doctor’s chatbot & agents at dinner 2026-10-23
One route, the right tool for each question
Method and revision history
Interview summaries retain attribution and dates. We infer industry dependencies, then link them to issuer evidence and stocks. Speaker agreement is not independent confirmation. Research priority is editorial, not a numerical probability or a forecast return. No price target is fabricated from a video.
New video metadata refreshes through the existing daily site job; reviewed theses change only after source review. Failed checks keep their original dates. The owner receives revisions and bounded new-review leads, not automatic buy/sell orders.
2026-10-03 · 2026-10-03.1: Established five routes and nine US-listed watch candidates. Speaker views, economic hypotheses and company evidence are separate; valuation remains unassessed.
Machine-readable evidence and route mapTurn an AI story into a research record
Trace the business benefit, inspect the evidence, then test how the price and your exposure could respond. Export a record you can revisit after the next filing.
Free research aid. US-listed company examples; not a stock ranking, live feed or buy/sell service. Framework reviewed September 28, 2026.
1. Map where the AI benefit becomes cash
Editorial research questions, not an opportunity ranking. For each company compare the same period, unit and accounting basis. Separate realized revenue from orders, guidance and market narratives.
| Business layer | Check | Counter-case |
|---|---|---|
| Compute & networking | Revenue, gross margin, concentration | Capacity growth exceeds durable demand |
| Memory & manufacturing | Inventory, utilization, cash capex | A supply cycle erodes pricing |
| Power & cooling | Order conversion, project returns, debt | Delays or financing absorb the benefit |
| Cloud platforms | Cloud growth, cash flow, leases | Spending grows faster than monetization |
| Applications | Paid retention, margins, cost to serve | Usage grows but unit economics worsen |
2. Start with a disclosed fact
NVIDIA (NVDA)
Data Center revenue: $89.0 billion; company-reported quarter, USD.
FY2027 Q2 · 2026-07-26
Published 2026-08-26 · Checked 2026-09-28
Segment revenue does not isolate end-user AI returns or prove a fair stock price.
Read company disclosureMicrosoft (MSFT)
Microsoft Cloud revenue: $59.3 billion; company-reported quarter, USD.
FY2026 Q4 · 2026-06-30
Published 2026-07-29 · Checked 2026-09-28
Microsoft Cloud is broader than Azure and AI; this is not standalone AI revenue.
Read company disclosureDated snapshots, not automatically refreshed. A source supports its reported metric, not the example thesis. Sample review dates remain blank on purpose.
3. Write your thesis and its failure condition
Worked example
Entering a link does not verify it. AI can help extract and compare evidence; the source still needs checking.
4. Stress-test growth, valuation and exposure
Fictional example: EPS grows 20% a year for 3 years, but P/E falls from 40 to 25. Earnings rise 72.8%; the implied price change is only 8.0%, before dividends and costs. These inputs are not forecasts for either company above.
Uses positive EPS and positive P/E only. Price factor = (1 + annual EPS growth)^years × ending P/E ÷ starting P/E. EPS growth must already reflect share dilution. Excludes dividends, fees, taxes and FX. Loss-making businesses need another method.
Stress is a separate one-off scenario: weight × sleeve shock + (1 − weight) × other-assets shock. It is not a forecast, a target allocation, or a maximum-loss bound.
- Cumulative EPS growth
- —
- Cumulative price change
- —
- Annualized price change
- —
- Annual EPS growth needed to break even
- —
- Portfolio stress return
- —
- Gain needed to recover from that stress
- —
Keep a record; ask AI to challenge it
Download a source-led brief with an AI review prompt, or a JSON file you can import next time. This tool does not call an AI model. Saving is optional and only uses this browser. Research text is never included in our tool events.
Research discipline and sources
A 13F can arrive up to 45 days after quarter-end and does not show short stock positions. A filing-day closing price is not necessarily available to a follower: check the publication time, then use the next executable price. Historical model returns are not realized investor profits.
SEC 13F FAQ · SEC EDGAR API · Damodaran: cash flows and valuation assumptions · FINRA: diversification
Next Q3 2026 13F deadline: November 16, 2026 (SEC calendar).
Need to check a number? Open FilingLens. Need to assess a paid alert service? Inspect SunWatch’s public record and current terms first. This research record is not automatically monitored by either service.
Embed this free workbench
<iframe src="https://agiscorecard.com/invest?embed=research&utm_source=embed" title="AI investment research" width="100%" height="1100" loading="lazy"></iframe>
Open full workbench on AGI Scorecard · Dated source data (JSON)
Historical investor holdings and background
How are the biggest investors positioned on AI?
Public portfolio tracker
Registered October 3, 2026: a twelve-stock basket versus SPY, QQQ and TQQQ, with position stress, contribution scenarios and saved reviews.
Open tracker and investment toolsStart with one disclosed fact. Test your AI thesis against growth, valuation and portfolio exposure.
Open research workbenchNew: annual SEC facts for 20 AI infrastructure companies, source-linked AI research questions and cloud notes.
Open infrastructure researchThe predictor's own bet: Aschenbrenner's fund
The author of "AGI by 2027" didn't just predict — he ran Situational Awareness LP to a reported 439% return for 2026 through June (>1000% since inception). His Q1 2026 13F showed a very specific thesis: long AI power & infrastructure (Bloom Energy $879M, CoreWeave, SanDisk) with ~$8.5B notional in puts against chip names — Nvidia (~$1.57B), SMH (~$2B), Broadcom, AMD, Micron. Then July 2026: per CNBC and Bloomberg, the fund lost ~67% in a month as longs and shorts moved against it together, and transferred most of its public portfolio to Citadel — assets down from a ~$45B peak to ~$10B, keeping private stakes like Anthropic. He may still be right about the industry; the leverage didn't survive long enough to find out. The collapse, graded against his predictions → · The full holdings story → · How his predictions grade →
How 8 investing legends are positioned on AI
Snapshot from Q2 2026 filings (as of 2026-06-30, filed 2026-08-14). Two honest exceptions: Ackman's latest filing on record is Q1 2026 (filed 2026-05-15), and Burry has filed no 13F since winding down Scion in November 2025 — his stance is per his 2026-05 personal disclosure. Stances are our reading of their public AI-related moves, not their words:
| Investor | Firm | AI stance | Signature move |
|---|---|---|---|
| Warren Buffett | Berkshire Hathaway | Cautious | Alphabet added again in Q2 (A+C classes now 12.6% of the book) — the clearest AI bet; Apple still #1 at 22%; Amazon, exited in Q1, not rebought |
| Cathie Wood | ARK Invest | Bull | Aggressively long AI infra & next-gen compute: CoreWeave, Cerebras; nuclear (X-Energy) as AI's power base |
| Bill Ackman | Pershing Square | Bull | Amazon top AI position (17.4% of an 11-name book), Microsoft 15.3%, Meta 11.1% — per Q1 2026, his latest filing |
| David Tepper | Appaloosa | Bull | Amazon #1 (15.4%), Micron 14.6%, TSMC 10.2% — AI monetization and the memory cycle over pure compute |
| Stanley Druckenmiller | Duquesne | Cautious | TSMC top AI position (5.4%); rotated into storage and memory (Seagate 2.3%, SanDisk 1.5%) — the bottleneck trade, not the crowded names |
| Duan Yongping | H&H International | Cautious | Apple the anchor (41.1% of the book); Tesla 7.4%, Nvidia 6.6% — the 2026 full-chain AI tilt held through Q2 |
| Philippe Laffont | Coatue | Cautious | Full-stack AI infrastructure: TSMC #1 (8.8%), Lam Research 8.4%, Micron 7.5% |
| Michael Burry | Scion (wound down) | Bear | The clearest AI bear (per 2026-05 personal disclosure) |
Full profiles: Warren Buffett — Alphabet again in Q2 2026 · Cathie Wood — ARK's Q2 book (more migrating each week).
New: Is Nvidia overvalued? What its AI bet actually rides on — the bet-structure audit: 90% buildout, under 10% AGI, three 13F books that disagree completely.
🧭 Per-holding detail lives on the Compass — every investor profile with per-position notes, theses and sources, bilingual: open compass.agiscorecard.com →
You have seen what they hold. What does your basket actually bet on?
Pick your AI names and get two numbers: a score against the eight graded predictions on the same 0–100 scale as the Thesis Tracker, and the share of your basket that genuinely rides on AGI arriving rather than on the buildout continuing. A standard infrastructure basket scores 99 with about 2% riding on AGI itself.
Check my exposure →Read the historical simulation and its assumptions.
Version 13f-next-session-v2 uses the first completed trading-session close after the filing date, excludes Put/Call and principal rows, and requires complete adjusted prices on matching dates and contiguous quarters. These are gross historical simulations before costs, taxes and slippage, not live returns. Current-universe selection and survivorship bias remain; original filings do not reconstruct all amendments or confidential disclosures. As of 2026-09-28, over 8 rebalances: Druckenmiller +179.5% vs QQQ +55.5%; Buffett +34.6% vs QQQ +55.5% (2024-08-15 to 2026-08-17). Wood +161.8% vs her own-window QQQ +59.0%; Tepper +107.1% vs QQQ +55.5%. Different managers can have different windows. These results do not establish future alpha.
See the full scorecard → Method, caveats and the full table: does copying 13F filings work? →Past results do not predict future returns. An after-the-fact calculation on public filings, not a recommendation.
Historical model, not realized profit. Filing-day close backtests can include look-ahead bias when the filing arrives after the close. The linked historical returns have not been revalidated here against publication timestamps, next executable prices, costs or a survivorship-free universe. Do not treat them as attainable returns.
Tools in this section
Compare the same financial fact for the exact same period at two filing cutoffs. Inspect both source values and accession links. Free browser tool and local MCP beta; no investment signals.
Compare a financial fact →Score your own basket against the eight graded predictions — and see how little of it actually rides on AGI arriving.
Score your basket →Live watchlist with pre-registered buy/sell trigger levels, 52-week positioning, and an auditable forecast ledger — updated every 30 minutes.
Open SunWatch →Did copying the legends' AI holdings actually pay? Historical simulation at the next available session close after filing, compared with QQQ over matched windows.
See the scorecard →AI-sector themes, legends' 13F positioning, and a consensus score per AI stock.
Open the Compass →Build a US stock/ETF basket, compare target weights, equal weight and HRP over the same window, inspect drawdowns and correlations, test assumptions, and keep an evidence journal. Local saving and report export; English and Chinese interfaces. Historical research, not trading signals.
Open the workbench →The information edge: is AGI-2027 actually on track? One auditable score.
Thesis Tracker (62.5/100) →A public pre-registered experiment: kill conditions filed before entry, monthly audits. Plus: how much of Justin Sun's 2013 story is verified?
The live ledger → · The audit →Every dated AGI call on a live countdown — the claims behind the trades.
See the receipts →Six paper bots — SPY/QQQ holds, an equal-weight AI basket, a Thesis-Tracker mix, the Faber 200-day rule and an LLM arm — from one pre-registered start date, recomputed daily in public. Read date 2027-03-08.
Open the ledger →The Q2 13Fs landed August 14, 2026 — six of the eight books above are now June 30 snapshots.
The exceptions stay labelled: Ackman's latest filing is Q1, and Burry has filed nothing since Scion wound down. One filing remains unusually loaded: Situational Awareness's own Q2 13F is the last full picture of what it held on June 30 — at full leverage, weeks before it lost ~67% and sold most of the book to Citadel — and we grade it line by line the day we have read it. Holdings here change only after we have read a filing line by line; qualitative commentary never becomes a position on this page. The day this page changes, subscribers hear about it.
Tell me what changed, filing day →Why this section exists
AGI timelines aren't academic — trillions in capex, chip valuations and fund flows hinge on them. The scorecard grades the claims; this hub tracks the capital. Together they close the loop: research the thesis, see who's betting what, and use the tools — then make your own decisions.
Frequently asked questions
Per Q2 2026 SEC 13F filings (holdings as of June 30, filed August 14): Warren Buffett's Berkshire added Alphabet again — now 12.6% of the book across both share classes, its clearest AI bet — while staying out of Amazon; Cathie Wood is aggressively long AI infrastructure and next-gen compute; David Tepper leans bullish with Amazon at 15.4% his largest position; Bill Ackman remains bullish per his latest Q1 filing; Druckenmiller, Duan Yongping and Philippe Laffont are cautious; Michael Burry is the clearest bear.
Leopold Aschenbrenner — who predicted AGI by 2027 — runs Situational Awareness LP. Its Q1 2026 13F showed the fund long AI power and infrastructure (Bloom Energy, CoreWeave) with roughly $8.5B notional in put options against chip names including Nvidia. In late July 2026, per CNBC and Bloomberg reporting, the fund lost about 67% in a month and transferred most of its public equity portfolio to Citadel; assets fell from a ~$45B peak to about $10B, retaining private stakes such as Anthropic.
No. Everything in this section is educational information based on public SEC 13F filings and public statements. Holdings snapshots are quarterly and may not reflect current positions. Nothing here is a recommendation to buy or sell any security.
Follow the money, quarterly.
When 13Fs drop or an AGI verdict flips, we tell you what changed — free, no hype.
Subscribe free →