PORTFOLIO LAB / RETROSPECTIVE EXPERIMENT
Own the model builders.
Test the investment.
Eight public companies connected to model development. Seven investable from 2020. SpaceX joins only after its 2026 IPO. One transparent test against QQQ and SPY.
Selected on 2026-09-19, not in 2020. This is a hindsight-selected universe, not the fund’s actual portfolio, a live trading record or proof of alpha. The weights are judgment calls, not optimized to beat the benchmarks.
THE RESULT / 2020-01-02 → 2026-10-13
What happened to $10,000?
254.94% total return
20.55% annualized · -49.23% max drawdown257.35% total return
20.67% annualized · -35.12% max drawdown163.12% total return
15.34% annualized · -33.72% max drawdownDots on the baseline mark annual resets. The vertical line marks the SpaceX purchase at the IPO-day close; the model earns none of the offering-to-first-close jump.
The investment lesson
The model trailed QQQ by 2.41 percentage points of cumulative return. Its worst daily-close drawdown was -49.23%, versus -35.12% for QQQ. Exposure to important technology does not by itself produce superior risk-adjusted returns.
These are diversified parent-company stocks, not pure claims on AGI profits. Shared technology, capex and country risks mean eight names are not eight independent bets.
Prices through 2026-10-13 · 1,704 matched sessions · updates with Daily market research after this change is deployed. End-of-day data, not live quotes. Refresh failures leave the last dated version visible.
Download every chart point (CSV) ↗Full calculation and trade ledger (JSON) ↗
ALLOCATION / WHY THESE WEIGHTS?
Strong anchors. Limited new-issue risk.
The initial US platform block is 85%; the China block is 15%. We prefer multiple revenue channels and established distribution. These are research allocations, not fair-value estimates or a recommendation tailored to your finances.
| Company / model connection | 2020 target | Next annual target* | Actual model weight 2026-10-13 |
|---|---|---|---|
| Alphabet GOOGL · Gemini / Google DeepMind | 25.00% | 23.75% | 25.02% |
| Microsoft MSFT · MAI models, Azure and enterprise distribution | 25.00% | 23.75% | 25.72% |
| Amazon AMZN · Amazon Nova / AWS | 20.00% | 19.00% | 20.32% |
| Meta META · Llama and AI distribution | 15.00% | 14.25% | 15.69% |
| Alibaba BABA · Qwen / Alibaba Cloud | 6.00% | 5.70% | 3.83% |
| Tencent TCEHY · Hunyuan / product distribution | 6.00% | 5.70% | 3.62% |
| Baidu BIDU · ERNIE / Baidu AI Cloud | 3.00% | 2.85% | 1.49% |
| SpaceX SPCX · xAI acquired by SpaceX in February 2026 | 0.00% | 5.00% | 4.31% |
*After IPO admission, the standing annual targets are 95% of the original seven-company targets plus 5% SpaceX. On IPO day itself we preserve the seven companies’ relative market weights, not reset them to these targets. Weights drift between trades; 5% is an entry target, not a continuously enforced cap.
GOOGL / Alphabet
25%: a core model developer with search, cloud and distribution. Its model capability can also disrupt its own search economics.
Model-development connection ↗MSFT / Microsoft
25%: enterprise monetization and cloud distribution balance pure model-development risk. Capex intensity and partner dependence remain material.
Model-development connection ↗AMZN / Amazon
20%: models, cloud and custom silicon support multiple routes to revenue. Retail dilutes direct model exposure, so the weight is below the two anchors.
Model-development connection ↗META / Meta
15%: model development with consumer-scale distribution and advertising monetization; offset by heavy spending and less diversified revenue.
Model-development connection ↗BABA / Alibaba
6%: Chinese model and cloud exposure. Country, governance and ADR/VIE risks constrain the allocation, not a claim that its technology is weaker.
Model-development connection ↗TCEHY / Tencent
6%: models embedded in a large product ecosystem. This test uses the USD OTC ADR, not Hong Kong 0700; liquidity, ADR fees and corporate-action treatment can differ.
Model-development connection ↗BIDU / Baidu
3%: model and application exposure with greater business concentration. The smallest legacy position keeps the combined China block at 15%.
Model-development connection ↗SPCX / SpaceX
5% on admission: participation in xAI through a newly public, capital-intensive conglomerate. IPO volatility, valuation uncertainty and launch/satellite exposure argue against an anchor-sized weight.
Model-development connection ↗JUNE 12, 2026 / SPACEX ENTERS
A 5% position. No extra money.
The SEC offering document identifies SPCX, its June 12 trade date and a $135 offering price. Our model does not assume access to that allocation. It buys at the recorded first closing price of $160.95, plus a modeled 0.50% transaction cost.
Each legacy holding is reduced proportionally, with a small additional reduction to fund costs. This avoids inventing a hindsight judgment about which stock was “the right one” to sell. SpaceX receives 5% of post-cost NAV: $31,454.19. The modeled purchase is 9.7714 fractional-share equivalents.
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 30.20% | 28.69% | -$477.83 | $0.48 |
| MSFT | 21.93% | 20.83% | -$346.99 | $0.35 |
| AMZN | 22.26% | 21.15% | -$352.30 | $0.35 |
| META | 13.83% | 13.14% | -$218.88 | $0.22 |
| BABA | 4.64% | 4.40% | -$73.36 | $0.07 |
| TCEHY | 4.70% | 4.47% | -$74.44 | $0.37 |
| BIDU | 2.44% | 2.32% | -$38.65 | $0.04 |
| SPCX | 0.00% | 5.00% | +$1,572.71 | $7.86 |
SEC: offering terms and first trade date ↗ · SEC: xAI became a SpaceX subsidiary ↗
SPCX previously identified a different security. Pre-IPO prices are explicitly rejected; no retired ETF history, private valuation or SpaceX price before listing enters the test.
THE DECISION RECORD / MODELED, NOT EXECUTED
When did we rebalance—and why?
6 annual resets, plus IPO admission. Reset at the first US trading close of each new calendar year to control drift with limited turnover. Daily monitoring does not mean daily trading. No earnings, AI release or price crash has been retroactively declared a trading signal.
2020-01-02Initial investment$12.38 costs
Invest the initial $10,000 at this session’s close, after funding modeled costs. NAV before: $10,000.00; after costs: $9,987.62. Traded notional / pre-trade NAV: 99.88% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 0.00% | 25.00% | +$2,496.90 | $2.50 |
| MSFT | 0.00% | 25.00% | +$2,496.90 | $2.50 |
| AMZN | 0.00% | 20.00% | +$1,997.52 | $2.00 |
| META | 0.00% | 15.00% | +$1,498.14 | $1.50 |
| BABA | 0.00% | 6.00% | +$599.26 | $0.60 |
| TCEHY | 0.00% | 6.00% | +$599.26 | $3.00 |
| BIDU | 0.00% | 3.00% | +$299.63 | $0.30 |
2021-01-04Annual target reset$1.59 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $13,812.10; after costs: $13,810.52. Traded notional / pre-trade NAV: 10.07% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 22.80% | 25.00% | +$303.62 | $0.30 |
| MSFT | 24.76% | 25.00% | +$32.70 | $0.03 |
| AMZN | 24.28% | 20.00% | -$591.60 | $0.59 |
| META | 13.91% | 15.00% | +$150.94 | $0.15 |
| BABA | 4.50% | 6.00% | +$207.34 | $0.21 |
| TCEHY | 6.35% | 6.00% | -$48.94 | $0.24 |
| BIDU | 3.40% | 3.00% | -$55.66 | $0.06 |
2022-01-03Annual target reset$5.89 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $18,102.72; after costs: $18,096.83. Traded notional / pre-trade NAV: 23.23% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 32.04% | 25.00% | -$1,276.07 | $1.28 |
| MSFT | 29.57% | 25.00% | -$829.53 | $0.83 |
| AMZN | 16.32% | 20.00% | +$665.31 | $0.67 |
| META | 14.40% | 15.00% | +$106.84 | $0.11 |
| BABA | 2.42% | 6.00% | +$648.02 | $0.65 |
| TCEHY | 3.67% | 6.00% | +$421.59 | $2.11 |
| BIDU | 1.57% | 3.00% | +$257.97 | $0.26 |
2023-01-03Annual target reset$2.83 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $10,990.55; after costs: $10,987.72. Traded notional / pre-trade NAV: 18.64% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 25.30% | 25.00% | -$33.90 | $0.03 |
| MSFT | 29.74% | 25.00% | -$521.38 | $0.52 |
| AMZN | 16.59% | 20.00% | +$374.74 | $0.37 |
| META | 9.10% | 15.00% | +$647.95 | $0.65 |
| BABA | 7.55% | 6.00% | -$170.38 | $0.17 |
| TCEHY | 7.78% | 6.00% | -$195.72 | $0.98 |
| BIDU | 3.95% | 3.00% | -$104.15 | $0.10 |
2024-01-02Annual target reset$6.01 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $18,412.25; after costs: $18,406.24. Traded notional / pre-trade NAV: 21.39% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 23.13% | 25.00% | +$342.77 | $0.34 |
| MSFT | 23.30% | 25.00% | +$311.60 | $0.31 |
| AMZN | 20.85% | 20.00% | -$157.93 | $0.16 |
| META | 24.85% | 15.00% | -$1,814.51 | $1.81 |
| BABA | 2.95% | 6.00% | +$561.33 | $0.56 |
| TCEHY | 3.19% | 6.00% | +$517.68 | $2.59 |
| BIDU | 1.73% | 3.00% | +$233.04 | $0.23 |
2025-01-02Annual target reset$3.53 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $25,029.58; after costs: $25,026.04. Traded notional / pre-trade NAV: 12.79% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 25.30% | 25.00% | -$75.06 | $0.08 |
| MSFT | 20.90% | 25.00% | +$1,024.15 | $1.02 |
| AMZN | 21.60% | 20.00% | -$401.88 | $0.40 |
| META | 19.16% | 15.00% | -$1,042.02 | $1.04 |
| BABA | 5.12% | 6.00% | +$219.39 | $0.22 |
| TCEHY | 6.33% | 6.00% | -$82.97 | $0.41 |
| BIDU | 1.58% | 3.00% | +$354.86 | $0.35 |
2026-01-02Annual target reset$8.39 costs
Restore the predeclared allocation at this session’s close. Sell names above target and buy names below target; do not anticipate the next day’s returns. NAV before: $33,262.99; after costs: $33,254.60. Traded notional / pre-trade NAV: 21.65% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 31.41% | 25.00% | -$2,135.92 | $2.14 |
| MSFT | 21.41% | 25.00% | +$1,192.47 | $1.19 |
| AMZN | 15.48% | 20.00% | +$1,502.98 | $1.50 |
| META | 12.29% | 15.00% | +$900.88 | $0.90 |
| BABA | 8.42% | 6.00% | -$804.09 | $0.80 |
| TCEHY | 6.89% | 6.00% | -$297.88 | $1.49 |
| BIDU | 4.10% | 3.00% | -$366.84 | $0.37 |
2026-06-12SpaceX admission · 5%$9.74 costs
The dated IPO event admits the eighth company; reduce the seven existing positions proportionally. NAV before: $31,463.93; after costs: $31,454.19. Traded notional / pre-trade NAV: 10.03% (buys + sells, not one-way turnover).
| Security | Before | After | Buy / sell USD | Cost |
|---|---|---|---|---|
| GOOGL | 30.20% | 28.69% | -$477.83 | $0.48 |
| MSFT | 21.93% | 20.83% | -$346.99 | $0.35 |
| AMZN | 22.26% | 21.15% | -$352.30 | $0.35 |
| META | 13.83% | 13.14% | -$218.88 | $0.22 |
| BABA | 4.64% | 4.40% | -$73.36 | $0.07 |
| TCEHY | 4.70% | 4.47% | -$74.44 | $0.37 |
| BIDU | 2.44% | 2.32% | -$38.65 | $0.04 |
| SPCX | 0.00% | 5.00% | +$1,572.71 | $7.86 |
A thesis invalidation could justify a faster future review. This backtest contains no such rule because we have not reconstructed a point-in-time evidence ledger for 2020–2026. A future rule change needs a new policy version; it must not silently rewrite this experiment.
ROBUSTNESS / KEEP THE UNCOMFORTABLE COMPARISONS
What did the rules contribute?
| Same dates and capital | Ending value | Annualized return | Max drawdown |
|---|---|---|---|
| Base policy | $35,493.63 | 20.55% | -49.23% |
| No annual resets; same 5% IPO entry | $33,908.25 | 19.74% | -48.18% |
| Annual resets; never add SpaceX | $35,762.67 | 20.68% | -49.23% |
| Base policy; zero transaction costs | $35,599.09 | 20.60% | -49.21% |
| Base policy; doubled transaction costs | $35,388.50 | 20.49% | -49.25% |
Annual resets changed ending wealth by $1,585.38 versus no annual reset. Adding SpaceX changed it by -$269.04 versus keeping seven names. These are path-dependent outcomes, not evidence that either rule will help next time.
Calendar-year returns and volatility
| Year | Model | QQQ | SPY |
|---|---|---|---|
| 2020 | 40.39% | 45.82% | 17.12% |
| 2021 | 28.47% | 27.42% | 28.73% |
| 2022 | -40.13% | -32.58% | -18.18% |
| 2023 | 73.27% | 54.86% | 26.18% |
| 2024 | 33.29% | 25.58% | 24.89% |
| 2025 | 32.56% | 20.77% | 17.72% |
| 2026 through 2026-10-13 | 7.38% | 21.45% | 14.98% |
| Daily volatility × √252 | 27.74% | 24.78% | 20.00% |
METHOD / WHAT THIS CAN AND CANNOT TELL US
Inspect the assumptions.
Return and execution model
$10,000 invested at the January 2, 2020 close. Fractional shares, no deposits, leverage or withdrawals. Day-to-day adjusted-close ratios approximate split- and distribution-adjusted total returns with automatic reinvestment. Rebalance only after that day’s return; a purchase earns returns from the following session.
Base one-way friction: 0.10% on each buy or sale; 0.50% for TCEHY and SPCX. QQQ and SPY are buy-and-hold with 0.10% entry cost. No terminal liquidation. Actual spreads may be wider. Fees paid: $50.37; their final wealth effect also includes forgone compounding.
Data and limitations
Yahoo Finance research data, in USD. Tencent is TCEHY (OTC ADR); Alibaba and Baidu use US ADRs. No Hong Kong close or missing currency conversion is spliced into USD prices. All active assets share 1704 complete US sessions; missing internal sessions fail the calculation rather than being filled with invented prices.
Vendor-adjusted closes can be revised and are not an independently audited corporate-action ledger. Taxes, withholding, ADR fees, custody, liquidity limits and market impact beyond the fixed friction assumptions are not separately modeled. ETF expenses are embedded in their price history.
Hindsight is still here
The 2026 company selection, current model connections and SpaceX admission are known after the fact. Earlier stock returns were not necessarily earned from AGI. Avoiding a future-price look-ahead in the arithmetic does not remove universe-selection bias or make this an investable 2020 forecast. No parameter search or best-weight optimization was performed.
What comes next?
Observe this policy without replacing it after disappointing results. Compare model adoption, monetization, free cash flow, capex and dilution with valuation; any changed allocation requires a dated rationale. The official research portfolio and its forward record remain separate.
Return to the official research portfolio ↗Source files, retrieval dates and reproducibility
Calculation policy selected 2026-09-19. Snapshot hashes identify the exact cached inputs used; Git preserves their version. The graph ends at the oldest latest close across all ten series.
| Security | Retrieved (UTC) | Latest close | Data |
|---|---|---|---|
| GOOGL | 2026-10-14T15:55:41.080Z | 2026-10-13 | Cached rows · Vendor |
| MSFT | 2026-10-14T15:55:38.750Z | 2026-10-13 | Cached rows · Vendor |
| AMZN | 2026-10-14T15:55:37.575Z | 2026-10-13 | Cached rows · Vendor |
| META | 2026-10-14T15:55:43.384Z | 2026-10-13 | Cached rows · Vendor |
| BABA | 2026-10-14T15:56:44.425Z | 2026-10-13 | Cached rows · Vendor |
| TCEHY | 2026-10-14T15:56:58.780Z | 2026-10-13 | Cached rows · Vendor |
| BIDU | 2026-10-14T15:56:57.905Z | 2026-10-13 | Cached rows · Vendor |
| SPCX | 2026-10-14T15:56:37.368Z | 2026-10-13 | Cached rows · Vendor |
| QQQ | 2026-10-14T15:56:47.687Z | 2026-10-13 | Cached rows · Vendor |
| SPY | 2026-10-14T15:56:52.887Z | 2026-10-13 | Cached rows · Vendor |
Reproduce offline: node automation/build-model-builders.mjs --check. Projection hash: eba8e3808e08a91865092983de1a756f32ca8907d4f4eb609c84d86ce7407eb4. This page is educational research, not a securities offer or personalized financial advice.