The AI Bubble: Why Manias Never Really Die (Part 4)
How to size a portfolio for survival of the AI Bubble; What survival looks like and what it costs
Part 2 of this series, published in December 2025, did not just diagnose the mania. It prescribed a specific portfolio architecture, the “sustainable allocation framework for the In-Between phase”, with named securities, specific weights and rules governing when to trim, when to add and when to hold. This is Part 4. It prices that framework against reality.
This Part 4 covers:
The full In-Between portfolio: every named position, YTD return, and contribution to total return
Sleeve attribution: where the +5.15% came from across all five sleeves
The standout performers within the AI Tactical sleeve
The one detractor that the framework anticipated and sized for
The Five Warning Lights: how the signals predicted in Part 2 have moved
The framework instructions that now apply, and the three trip wires that would change the allocation materially
The portfolio period is January 1 to April 27, 2026, using year-to-date return data sourced from named financial databases.
The result: +7.20% total return, against the S&P 500’s +4.85% over the same period. Outperformance of +2.35 percentage points against the primary benchmark, and +1.27 percentage points against the MSCI All-Country World Index, which returned approximately +5.93% YTD.
While there was an outperformance, the In-Between framework was not designed to maximise return. It was designed to stay in the game, to participate in the AI opportunity, protect against the drawdown, and preserve the dry powder needed to buy assets at reset prices. Over four months that included a fresh S&P 500 all-time high, a geopolitical shock from the Iran war, and the highest bearish retail sentiment reading in years, the framework did exactly what it was built to do.
The detail the contribution of each sleeve, the star performers, the one position that cost the portfolio meaningful ground, and the framework instructions that now apply is below.
The Framework Being Tested
Part 2 prescribed the In-Between portfolio to solve a specific problem: how does a disciplined investor stay exposed to the greatest capital formation event of their lifetime without being destroyed by the correction that historically follows every mania?
The answer was structural, not speculative. Three sleeves. Three mechanical rules. One overriding principle: size for survival.
The Carlota Perez framework anchored the macro case.
The In-Between Portfolio
The framework from Part 2 prescribed three sleeves and five sub-categories. The table below prices each named security at YTD returns. Where individual securities were named, they are priced individually. Where ETF alternatives were named, the primary ETF is used.
January 1, 2026 entry prices are estimated by working backward from verified April 27, 2026 closing prices and sourced YTD return data. Some returns are estimated from available reporting and should be treated as approximate. Not investment advice. (Sources: FinanceCharts, Macrotrends, Robinhood, Investing.com, Dividend.com, Yahoo Finance.)
Where the Returns Came From
The most important number in the table is not in the return column. It is −0.57 ppts, the damage from the entire AI Hopes sleeve. Palantir fell 23.2% over the period. The Hopes sleeve as a whole returned approximately −19.10%. On the framework’s 3% combined AI Hopes allocation, that cost the portfolio just half a percentage point.
The Part 2 rule — “AI Hopes deserve a seat at the table, never dependence” — was structural. An AI Hopes allocation of 15% (the ceiling the article permitted) would have cost 2.6 percentage points instead of 0.57. The discipline of undersizing the Hopes sleeve absorbed what would otherwise have been a meaningful drag.
The ballast returned +0.60 percentage points, contributed by gold’s exceptional YTD return of +8.47%, driven by central bank buying, Iran war safe-haven demand and inflation concerns. This is the framework’s most underappreciated element: the ballast was not designed to return zero, it was designed to return something uncorrelated to the AI trade.
What the Framework Named Correctly
Vertiv (VRT): +86.0% , The article’s most specific call
Part 2 named Vertiv explicitly as a “Railway Engineer”, a company providing “the power and cooling plumbing without which the digital brain cannot function.” Q1 2026 confirmed it: net sales of $2.65 billion, up 30% year-on-year; operating profit up 51%. The Nvidia Vera Rubin DSX collaboration and S&P 500 inclusion in March 2026 validated both the business thesis and the institutional recognition the framework anticipated. At +86.0% YTD on a 2.3% tactical weight, VRT contributed 2.0 percentage points to total portfolio return — the single largest individual contribution. The +2 sigma trim rule is now triggered. The rule says trim 10% of the position, not exit.
Equinix (EQIX): +53.6%, The energy infrastructure call
Part 2 named Equinix explicitly as a data centre REIT that “owns the digital dirt — in a world of infinite software, specialised physical space is the ultimate scarcity.” Multiple analyst upgrades in April 2026 confirmed the re-rating: Guggenheim to $1,235, HSBC to $1,250, Wells Fargo to $1,200. At +53.6% YTD on a 1% energy sleeve weight, EQIX contributed 0.54 percentage points — the third-largest individual contribution despite sitting in the smallest sleeve by weight. Per dollar allocated, it was the portfolio’s most efficient position.
Gold (GLD): +8.47%, The ballast that worked
Part 2 prescribed gold alongside T-bills as the ballast component that “thrives on volatility” and “hates what AI loves.” In 2026, with the Iran war in its eighth week, central banks buying in volume, and inflation above the Fed’s 2% target, gold returned +8.47% YTD — making it the third-best performing asset class in the portfolio by absolute return. Year to date as of April 22, 2026, gold prices were up approximately 9.2%, while the S&P 500 lagged at roughly 3.9% on a price return basis. The ballast was not passive. It was earning.
The Detractor: Sized to Absorb, Not Destroy
The AI Hopes Sleeve: −19.10% sleeve return, −0.57 ppts portfolio impact
The Hopes sleeve — Palantir, C3.ai, Symbotic, and IonQ — fell collectively. Palantir at −23.2% was the heaviest drag. C3.ai, Symbotic, and IonQ all continued their post-2025 corrections. The software and speculative AI segment underperformed the broad market materially in the period, consistent with the broader rotation away from pure-play AI software names that Part 2 explicitly anticipated.
But the damage to the total portfolio was −0.57 percentage points, a real cost but a contained one. This is the most important structural lesson in the portfolio audit: the framework was not wrong about the Hopes sleeve. It was right about how to size it. The Part 2 Golden Rule stated explicitly: “Finance your Hopes from the realised profits of your Core, not from new risk.” The 3% combined allocation to Hopes was the discipline that made a −23% Palantir drawdown a 0.35 percentage point portfolio event rather than a 2+ percentage point one.
The framework instruction for the Hopes sleeve now: hold Palantir at current weight pending May 4 Q1 2026 earnings. For C3.ai, Symbotic and IonQ, maintain positions at current minimal weights. No adds.
The Global Core, Doing Its Job Quietly
The 65% Global Core returned approximately +3.6% as a sleeve, positive but below the headline S&P 500 (+4.85%), Nasdaq 100 (+7.19%) and the non-US ACWI component (+5.93%) YTD. This is not a failure of the Global Core thesis. It is the thesis working correctly. The Global Core’s job is not to maximise return. It is to capture the Productivity Dividend, the adoption of AI by non-tech firms across the broad economy, without paying the Hype Premium concentrated in the AI Tactical sleeve. Over a full market cycle, that same diversification will be the cushion when the US AI concentration eventually corrects.
The Nasdaq-100’s +7.19% YTD return was the Global Core’s standout, consistent with Part 2’s inclusion of QQQ as a benchmark component and core exposure vehicle, capturing the most liquid end of the AI-adjacent market.
The Five Warning Lights, Predicted vs. Actual
Compared to December 2025, three signals are red in April 2026. Market Breadth is the signal that crossed the threshold. The Survival Directive the Part 2 article prescribed — “trim your winners, build your T-bill ladder, and stop assuming the next quarter’s return will look like the last” — is now active.
The framework’s predictive value is visible in hindsight: the two signals that were red in December correctly identified the compression that arrived in software and speculative AI stocks during the period. The In-Between portfolio was sized to absorb exactly this outcome.
The Framework Applied: What is required now
Trim within the AI Tactical sleeve. Vertiv has triggered the +2 sigma trim rule at +86% YTD, as prescribed: “when a stock trades two standard deviations above its own 3-year valuation band, trim 10% of the position. Not because the company is bad, but because the price has outrun the math.” Trim 10% of the VRT position. Tuck proceeds into the T-bill ladder. Do not exit. The thesis is intact.
Add to the AI Tactical Core on weakness. The entry trigger is firing on Microsoft: “when a Core leader is down 25–30% from its peak, but earnings revisions remain positive, you add.” MSFT is down 24% from its $555.45 all-time high. Azure grew 40% constant currency in Q3 FY2026 (reported April 29, two days after portfolio period — flag as post-period confirmation). EPS beat consensus. The business has not deteriorated. The framework prescribes adding. Do so within the AI Tactical Core weight, restoring MSFT toward its target allocation.
Hold the Global Core — do not rebalance away from international. Part 2 framework explicitly cautioned: “owning five different chipmakers isn’t diversification; it’s concentration in different wrappers.” Hold the international allocation. The Iran geopolitical premium that drove the current underperformance is episodic, not structural.
Start deploying the ballast — but selectively. Three warning lights are still red. The S&P 500 hit a fresh all-time high of 7,177 on April 27, up 4.85% for the year. Investor optimism, as measured by the American Association of Individual Investors (AAII) weekly survey, has bounced back above its long-run average for the first time in ten weeks. The market dipped, partially recovered, and moved on. The underlying risks have not.
This means the moment to deploy everything has not arrived — but the moment to deploy something has. Put 5% of the 15% ballast to work now, specifically into the names that have not bounced with the broader market: Microsoft and Eaton both remain well below their peaks and offer the better entry points. Keep the remaining 10% in reserve. A two-week rally does not make three red lights green.
Let the Hopes sleeve run its course. Do not add to any Hopes position before the May 4 Palantir earnings print. The framework’s Golden Rule remains the governing principle: Hopes are financed from Core profits, not new capital.
What the Framework was Designed to Produce
+7.20% against the S&P 500’s +4.85% is not a headline that will generate envy at a dinner table. It is a number that will matter enormously when the correction arrives — and the three red lights say it is coming.
The verified sleeve breakdown tells the real story. The Global Core returned +5.54% — quietly, without drama, doing exactly what 65% of the portfolio is supposed to do: participate in the broad market without paying the hype premium of the AI tactical layer. The AI Energy sleeve returned +26.2% on just 3% of capital. The AI Tactical Core returned +19.9% on 14%. The ballast returned +4.0%, with gold at +8.47% doing more work than T-bills — consistent with the Iran war safe-haven bid and central bank buying that defined the first quarter. The only sleeve that cost the portfolio ground was AI Hopes at −0.57 percentage points — and that is the number that matters most, not as a failure but as a proof of design.
The In-Between framework was built around one uncomfortable truth that stated plainly: “if your AI exposure dropped 50%, and your total portfolio fell by more than 10–12%, you are not invested — you are overexposed.” A concentrated AI portfolio that delivered +16% in this same period would deliver −30% to −40% in the reset that the three red lights are now signalling. The In-Between portfolio, with its 65% Global Core and 15% ballast, is sized to lose approximately 8–10% in that reset — not 30–40%.
And survival is what the next chapter of this cycle rewards. The infrastructure being built — the data centres, the grid transformers, the liquid cooling systems, the power converters — will endure long after the speculative layer that financed it has been repriced. The In-Between framework was not designed to capture all of that upside in a four-month sprint. It was designed to ensure the investor is still in the game when the productive phase of the revolution — Stage 4 of the Perez cycle, the Synergy phase — arrives.
The Three Tripwires that would require material reallocation:
1. A hyperscaler reports a quarter where AI revenue materially disappoints against capex deployed — Earnings Revisions crosses to RED (fourth light). Reduce the AI Tactical sleeve from 20% to 12%, increase the Ballast from 15% to 20%, and add 3% to the Global Core.
2. The S&P 500 falls to 6,300–6,400 (approximately −12% from April 27 levels). Deploy the second ballast tranche into the AI Tactical Core names at distressed prices — specifically MSFT, TSM, and VRT.
3. PLTR May 4 earnings show Artificial Intelligence Platform (AIP) commercial revenue growth below 25% AND government segment deteriorating simultaneously — both conditions required, not one alone. Exit the Hopes sleeve entirely. Redeploy into the Global Core.
Closing
The number that defines the period is not +7.20%. It is −0.57: the total portfolio impact of a Hopes sleeve that fell 19.1% on average, because the framework sized AI Hopes at 3% of total capital and refused to give speculation the weight that conviction deserves. Palantir fell 23.2%. C3.ai, Symbotic, and IonQ all continued their corrections. On a 3% combined sleeve weight, the damage was contained to less than six-tenths of a percentage point. That is not hindsight. It is the framework working exactly as designed.
The same framework that contained the Hopes damage also produced the period’s most instructive asymmetry: the AI Energy sleeve returned +26.2% on 3% of capital — contributing 0.79 percentage points — while the Hopes sleeve returned −19.1% on the same 3% weight, costing 0.57 percentage points. Equal capital. Opposite outcomes. The difference is the quality of the underlying thesis: physical infrastructure with contractual backlogs on one side, pre-revenue speculation on the other.
The framework’s next test arrives May 4, when Palantir reports Q1 2026 earnings — the first concrete data point that will determine whether the Hopes sleeve holds or exits, and whether the Survival Directive’s second-order consequences begin to unfold across the broader AI software complex.
The framework worked. The discipline is the next test.
THE AI BUBBLE SERIES
Part 1 — The Anatomy of a Mania: Railways to AI ·
Part 2 — Navigating the Mania: The Investor Framework ·
Part 3 — Cycle Position Update: April 2026 ·
Part 4 — The In-Between Portfolio Verdict · This article
Risk is a Privilege is written by Kenneth Legesi, CFA focuses on frontier capital markets and technology, African investment and the macro forces shaping the next decade of investing, legesi.substack.com
This article is for informational purposes only and does not constitute investment advice. January 1, 2026 entry prices are estimated by working backward from verified April 27, 2026 closing prices and sourced YTD return data. Past performance does not guarantee future results. Please consult your financial advisor before making investment decisions.