Can AI Beat the Benchmark? Public Prompts, Portfolios, and Tracking

We’re launching an experiment where AI is put to work publicly, and the results are also measured publicly.

The idea is simple: portfolio construction requests are brought to the AI here, the AI’s proposed portfolio is published, and then later, we track how well it actually performed.

The point of this thread is to test whether well-formulated assignments can actually generate usable investment ideas and functional portfolios. Many have already noticed that the AI’s response is often exactly as good or bad as the task given to it. A bad prompt produces generic mush. A good prompt can force the AI to address the right things:

  • what is the goal
  • what is the time horizon
  • what level of risk is acceptable
  • what investment targets can be used
  • with what weights the portfolio is constructed
  • what macro assumptions need to be considered
  • how the choices are justified

The post should at least state:

  • the AI application used, the model or version used
  • the date the prompt was run
  • the original prompt
  • the AI’s proposed portfolio with weights
  • benchmark index or ETF

In my opinion, the most interesting question is not whether AI can write a convincing-looking analysis. The most interesting question is: Does a well-prompted AI portfolio actually perform well? And if it does, where does it perform:

  • in industry selection,
  • in diversification,
  • in risk identification,
  • in valuation consideration,
  • or just in sounding smart?

The goal of this thread could be to build over time a proper practical comparison:

  • what kind of prompts work and what don’t
  • where AI actually helps
  • what kind of portfolios built by AI have performed well and where AI has gone badly wrong

The intention is not to outsource one’s own thinking to a machine. AI is a tool here, not an authority. That’s why monitoring is essential. Only actual development shows whether the proposal was truly good or just believable. Ihre Datenschutzeinstellungen is a good and free place to track portfolio development.

32 Likes

I started with this experiment. My prompt is merciless in that it doesn’t allow fixed-income investments. Additionally, I included a couple of my own views to prevent the AI from making completely free predictions. It’s about a global portfolio with the restriction that I must be able to buy and sell through Nordnet.

The prompt for the AI is hidden here:

Summary

Task: Build me a feasible investment portfolio for 2026

Provide a thorough and concrete analysis of which industries will be the best investment targets in 2026, with the most important selection criterion being strong revenue and operating profit growth over the next 2–5 years.

Based on this analysis, build me a ready-made investment portfolio that could form my entire portfolio.

1. Portfolio Goal

The portfolio should be:

  • growth-oriented
  • risk-managed
  • structured to avoid binary, speculative, and lottery-ticket-like risk
  • weighted towards high-quality, profitable, and sustainable winners rather than the most aggressive growth stories

Avoiding risk specifically means I do not want investment targets in the portfolio whose return expectation depends on one uncertain event, a single regulatory decision, a technology breakthrough, a funding round, or other strongly binary factor.

2. Portfolio Constraints

Follow these conditions precisely:

  • The portfolio may contain a maximum of 15 investment targets
  • There may be a maximum of 3 investment targets from one industry
  • Portfolio allocation:
    • ETFs 70 %
    • individual stocks 30 %

ETF Requirements

Accept only ETFs that meet all of the following conditions:

  • UCITS
  • listed on an EU-area stock exchange
  • available for purchase through Nordnet

For each selected ETF, the following must be stated:

  • the fund’s full name
  • ticker
  • ISIN
  • listing exchange
  • information on whether it is UCITS
  • information on whether it is EU-listed
  • an assessment or verification that it is available for purchase on Nordnet

Stock Requirements

  • Stocks can be listed on any major stock exchange
  • OTC-listed stocks must not be selected

For each selected stock, the following must be stated:

  • company name
  • ticker
  • exchange
  • country
  • information that it is not an OTC listing

3. Macro Assumptions that Must Be Considered in the Analysis

Build the analysis on the following assumptions:

3.1 The Iran war continues for at least 3 months

Assess its impacts on at least the following:

  • energy
  • oil and gas
  • electricity price
  • inflation
  • interest rates
  • defense industry
  • logistics
  • raw materials
  • manufacturing
  • consumer demand

3.2 Electricity availability problems for new data centers continue

Assess its impacts on at least the following:

  • data centers
  • electrical infrastructure
  • electricity grids
  • grid expansion investments
  • cooling solutions
  • energy efficiency
  • nuclear power
  • gas power
  • renewable energy
  • semiconductors
  • AI infrastructure

3.3 SaaS company valuations have fallen

The market fears that AI will weaken SaaS companies’ business. Separately assess:

  • where this fear is justified
  • where it is exaggerated
  • which SaaS companies or SaaS sub-segments can emerge as winners
  • where AI is likely to destroy pricing power
  • where AI can strengthen competitive advantage

4. Analysis Method

Do not just provide a general market commentary, but proceed systematically.

Phase A: Industry Screening

First, evaluate a wide range of industries based on the following criteria:

  • revenue growth outlook 2–5 years
  • operating profit growth outlook 2–5 years
  • risk level
  • valuation level relative to growth
  • benefiting from or suffering from the given macro assumptions

Then, rank the industries from best to worst.

Phase B: Investment Target Selection

Select the best ETFs and stocks for the most attractive industries.

Phase C: Portfolio Construction

Build a complete portfolio such that:

  • ETFs total 70 %
  • stocks total 30 %
  • the overall portfolio is diversified but not messy
  • there is not too much overlap between investment targets
  • each selection has a clear role in the portfolio

5. Mandatory Output Structure

A. Industries Ranked

Present the best industries in order from best to weakest.
For each industry, state:

  • why it is attractive
  • what supports growth in 2–5 years
  • how macro assumptions affect it
  • what is the main risk

B. Proposed Portfolio as a Table

Present the complete portfolio as a table, showing at least:

  • investment target
  • type (ETF / stock)
  • industry / theme
  • weight in portfolio
  • ticker
  • ISIN (mandatory for ETFs)
  • exchange
  • brief justification

C. Justifications for Each Investment Target

For each selection, state:

  • why this particular target was chosen
  • what is its main growth driver in 2–5 years
  • what are its main risks
  • why it is better than the closest alternatives
  • what role it serves in the overall portfolio

D. ETF Verification

For each ETF, provide a separate verification row:

  • UCITS: yes / no
  • EU-listed: yes / no
  • Available on Nordnet: yes / no / probably yes, but verification recommended

E. Risk Analysis for the Entire Portfolio

Conduct a separate risk analysis:

  • main macro risks
  • main industry risks
  • biggest individual faulty assumptions
  • what could go wrong most severely
  • what factors the portfolio is most exposed to

F. Critical Self-Assessment

Finally, critically assess your own proposal:

  • where the portfolio is strongest
  • where it is most vulnerable
  • what alternative weighting could be considered if the market outlook changes

6. Additional Instructions

  • Favor high-quality and profitable businesses
  • Avoid hype companies
  • Do not select targets merely because a theme is fashionable
  • Always consider valuation level
  • If a popular theme looks poor on a risk-adjusted return outlook, state it directly
  • If an unpopular sector looks strong, highlight it
  • If no good ETF can be found for an industry, state it directly
  • If no sufficiently high-quality stocks with low risk can be found for an industry, do not force a selection

7. Important Final Requirement

Do not stay on a general level. The final result must be so concrete that based on it I can:

  1. understand which industries look best in 2026
  2. see the complete portfolio with its weights
  3. verify the buyability of ETFs in practice
  4. compare selected targets to alternatives

The goal is not to create the most exciting portfolio, but the strongest, growth-oriented, and risk-managed investment portfolio for 2026.

The AI’s portfolio proposal will be in the next post because I hit the forum character limit :wink:

I eagerly await your own experiments here :face_savoring_food:

15 Likes

ChatGPT 5.4 Pro, Deep Research responded on April 7, 2026:[details=“Summary”]

A Viable Growth-Oriented Investment Portfolio for 2026

This is a model portfolio based on research and public sources, not personal investment advice. It is constructed for the situation on April 7, 2026 (Europe/Helsinki) and complies with the macro assumptions you provided (Iran war, data center power bottlenecks, decline in SaaS valuations).

A Sectors in Order of Preference

“Mandatory” Conclusions for Sector Screening Based on Macro Assumptions

The continuation of the Iran war for at least 3 months (and disruptions in the Strait of Hormuz) steers the model towards sectors that have (a) pricing power, (b) non-discretionary demand, and (c) the ability to benefit from or at least withstand an energy price shock. In this scenario, energy prices have risen sharply, physical crude oil prices have reached extremely high levels, and the IMF and central bank actors have warned of the impact on inflation and growth (as well as the risk of “higher for longer” interest rates).

The continuation of electricity availability issues for data centers is a clear signal that the bottleneck is not just “AI data centers,” but above all grid interconnection, power grid capacity, component availability, and execution schedules. The International Energy Agency (IEA) highlights power grids as a bottleneck, estimating, among other things, that over 2,500 GW of generation/load/storage projects are stuck in the queue, and that grid investments need to grow by about 50% by 2030 (from the current level of ~$400 billion/year) – while planning, permitting, and building new grids can take 5–15 years, whereas building data centers often takes 1–3 years.

The decline in SaaS valuations and the fear that “AI is eating SaaS” is twofold in the analysis: part of the software landscape is genuinely at risk of eroding pricing power, but the market may also generalize the fear too broadly. In this framework, I use:

  • Gartner forecasts on IT and AI spending growth and the proliferation of AI agents into applications.
  • BVP Nasdaq Emerging Cloud Index metrics as a “thermometer” for the valuation level and growth of cloud software (e.g., average revenue multiple).
  • Goldman Sachs publications in framing the AI risk for application software (“AI can change the user interface and workflows, but the market may also price the risk too broadly”).

Systematic Screening and Ranking of Sectors

Below is the ranking (best → weakest) based precisely on the criterion of “strong revenue and operating profit growth over 2–5 years” with risk managed and binary risks avoided.

Investment Suitability 2026 (Best→Weakest) Sector / Theme Why Attractive 2–5 Yrs Impact of Macro Assumptions Key Risk
1 Power Infrastructure & Electrification (grids, distribution, switchgear, power modules, cooling, energy efficiency) Structural, investment-driven growth: electricity demand is rising and the grid is a bottleneck; component demand + investments support revenue and margins for leading suppliers. Data center power bottlenecks = “forced” grid and electrification capex. Iran scenario supports energy efficiency and security-of-supply investments. Capex cycle timing, component shortages (transformers, switchgear), project delays; potential “overcapacity” later.
2 Cybersecurity (network, cloud, and identity security, SOC automation) Demand is close to non-discretionary: AI increases the attack surface, and the geopolitical environment raises priority. Gartner emphasizes AI, geopolitical tensions, and regulation driving cyber trends in 2026. Iran scenario increases cyber risk and the need to protect critical infrastructure; the proliferation of AI agents also increases “machine identity” security. Price pressure in “subcategories” (commoditizing features), as well as intensifying competition in platforms.
3 Semiconductors & AI Infrastructure (chips, networking, ASIC/XPU, memory chain) The continuation of the AI investment cycle (AI spending and infrastructure) supports strong revenue growth. Gartner estimates AI spending will grow strongly in 2026 and IT spending will grow clearly. Data center power bottlenecks can delay deliveries (delays), but the capacity being built does not disappear → demand shifts. Iran scenario: energy prices and interest rates could weigh on valuations. Cyclicality and valuation risk; additionally, geopolitical constraints in supply chains (export controls).
4 Defense & Aerospace Long “re-arm/restock” theme: defense spending has risen sharply, and geopolitical uncertainty supports order books. NATO reports that European allies’ and Canada’s defense spending has risen sharply in real terms, and SIPRI reports record global defense spending. Iran scenario supports defense demand and investments in logistics/air and missile defense. Political budget risk (if tensions ease), as well as valuation stretching in “one direction” (crowded trade).
5 Healthcare Innovation (drug development, medtech, diagnostics) Population aging and the burden of chronic diseases support long-term demand; at the same time, drug development innovation continues (e.g., new FDA approvals). Iran scenario: inflation/interest rates could weigh on riskier biotech parts; on the other hand, demand is more defensive than in consumer sectors. Regulatory and reimbursement system changes; the innovation theme also includes high dispersion (winners vs. losers).
6 High-Quality Cloud & Enterprise Software (SaaS winners, platforms) IT spending will grow in 2026 and AI infrastructure will remain strong. At the same time, SaaS valuations have fallen: in BVP Cloud metrics, revenue multiple levels have moderated compared to peak years. AI agents are changing usage and could erode “seat-based” pricing in some tools (justified fear), but the market may also price the risk too broadly for those with strong data/workflow moats. Loss of pricing power in certain horizontals; “AI crowding-out” (infrastructure taking budget from applications) in weaker categories.
7 Consumer Discretionary Weaker: energy price and inflation pressure impair purchasing power and growth in the scenario. Iran scenario negative; interest rate risk negative. Demand shock and margin pressure.
8 Leveraged / Rate-Sensitive (REITs, low quality) Weaker: interest rate risk and energy price risk, plus potential financing availability. Iran scenario increases inflation → interest rate pressure. Financing and valuation risk.

SaaS and AI Fear Separately: Where Justified, Where Exaggerated

Justified Fear (AI can destroy pricing power), especially when:

  • The product is a “thin” user interface around work and an AI agent can do the same work directly in backend systems; seat-based pricing is vulnerable if work is automated.
  • Product differentiation is low (commoditizing categories where features are quickly copied by AI).

Exaggerated Fear (AI is more likely to strengthen competitive advantage) when:

  • Software is a “system of record” and deeply embedded in processes: data, audit trail, compliance, and integrations create high switching costs; AI acts as an additional layer, not a replacement. (This is also a theme visible in SEG’s 2026 SaaS reports regarding buyer preferences: buyers reward sustainable growth, engagement, and workflow/data-anchored products.)

Winner Segments 2026–2030 (Probable):

  • Cybersecurity (AI increases both attacks and the need for protection).
  • Cloud platforms where AI infrastructure and distribution already exist (budget flows go to both infrastructure and AI features).Substation Transformer Guide: Design, Power Classes, and Grid ApplicationsCommissioning a new hyperscale data centre | Aggreko PHASML. The Dutch Company That Dominates Tech | by Dion F. Lisle | VentureVINE | MediumThe $150 Million Machine Keeping Moore's Law Alive | WIRED

B Ready-Made Portfolio Proposal as a Table

Construction principle: 1) global core + risk brake, 2) clear “winner themes” (power infra, AI infra, cyber), 3) high-quality, profitable individual companies where growth drivers are structural rather than driven by a single event.

Note: ETFs have been selected as EU-region exchange listings (in practice Xetra here), and can be found on Nordnet’s listings with the chosen ticker.

Investment target Type Sector / theme Weight Ticker ISIN (ETF) Exchange Short rationale
Vanguard FTSE All-World UCITS ETF (USD) Accumulating ETF Global equity core 30% VWCE IE00BK5BQT80 Xetra Broad, cost-effective “all-world” core that diversifies the risk of individual themes.
Amundi MSCI World Minimum Volatility Advanced UCITS ETF Acc ETF Risk brake / lower volatility 10% WMMV IE0001DKJVC2 Xetra Cuts the portfolio’s “worst drawdowns” in a scenario where an energy price shock keeps volatility and interest rates sensitive.
VanEck Semiconductor UCITS ETF ETF AI infra / semiconductors 9% VVSM IE00BMC38736 Xetra Broad exposure to the semiconductor chain; benefits from AI capex, but weight kept controlled due to cyclicality.
First Trust Nasdaq Clean Edge Smart Grid Infrastructure UCITS ETF - Class A ETF Power infra / smart grids 9% GRID IE000J80JTL1 Xetra Direct “grid and electrification capex” theme: bottlenecks are driving investments.
iShares Digital Security UCITS ETF USD (Acc) ETF Cybersecurity 5% L0CK IE00BG0J4C88 Xetra Cyber is one of the most defensive themes of 2026: AI + geopolitical trends force spend.
iShares Global Aerospace & Defence UCITS ETF USD (Acc) ETF Defense 4% 5J50 IE000U9ODG19 Xetra Upward trend in defense spending, but weight kept small so the theme doesn’t dominate with “war cycle” risk.
iShares Healthcare Innovation UCITS ETF USD (Acc) ETF Healthcare innovation 3% 2B78 IE00BYZK4776 Xetra Long-term demand (aging) + innovation flow, but high dispersion → moderate weight.
Schneider Electric SE Stock Electrification, energy management, data centers 6% SU – Euronext Paris Proven growth and profitability; data center drive and margin development.
Eaton Corporation plc Stock Power infra, power modules, data center demand 6% ETN – NYSE Exceptionally strong order/backlog visibility and segment profitability; benefits from electrification and data center cycle investments.
Broadcom Inc. Stock AI chip infra (ASIC/XPU) & networking 6% AVGO – Nasdaq Strong growth in AI chip and networking demand and long contract visibility (e.g., Google TPU partnership).
Microsoft Corporation Stock Cloud & enterprise software (AI + “core SaaS”) 6% MSFT – Nasdaq Wide distribution and cloud earnings: Azure has grown strongly and the company reports growth in both revenue and operating profit.
Palo Alto Networks, Inc. Stock Cybersecurity platform 6% PANW – Nasdaq Cyber platform also positioned for securing agentic AI; complements the cyber ETF’s diversification nature with an individual “quality winner” selection.

C Rationale for Each Investment Target

Below for each selection: why chosen, 2–5 year growth driver, risks, why better than alternatives, role in the portfolio.

Vanguard FTSE All-World UCITS ETF (VWCE)
The selection is the core because it provides global stock market exposure through a single instrument, preventing theme bets from making the portfolio “messy.” The cost structure is low and the ETF is large, which is important in practical implementation. Vanguard
The growth driver in 2–5 years is the compounding of global earnings growth and dividends (company-level diversification across ~thousands of holdings).
Risk: global stock market risk (drawdowns), currency risk, and the fact that the index also includes lower-growth sectors. Role: the “spine” upon which themes are built.Amundi MSCI World Minimum Volatility Advanced UCITS ETF (WMMV)
Chosen because, in a scenario where an energy price shock and inflationary pressure may keep volatility high and increase the risk that rate-sensitive, long-duration growth stocks face valuation pressure. Amundi
Growth driver: not “maximal growth,” but risk-adjusted return; lower fluctuation makes the overall portfolio easier to manage.
Risk: lags behind in a strong “risk-on” rally (when high-volatility stocks lead). Better than alternatives in this portfolio because it acts specifically as a risk brake without moving into fixed-income instruments (which were not requested). Role: lowers the portfolio’s overall risk and reduces the fluctuation caused by the binary macro scenario (oil/inflation).

VanEck Semiconductor UCITS ETF (VVSM)
Chosen because semiconductors are the economic center of the AI infrastructure “bottleneck chain”: the growth of AI spending and IT spending creates a demand base for computing, memory, and networking chips. VanEck
2–5 year driver: the continuation of the AI infrastructure investment cycle.
Risk: cyclicality and valuation (chip demand could normalize; energy prices/interest rates could weigh on it).
Why an ETF instead of a “single chip company”: avoids binary company risk (product or customer dependency). Role: growth engine, but with only a 9% weight to manage risk.

First Trust Nasdaq Clean Edge Smart Grid Infrastructure UCITS ETF (GRID)
Chosen because the electricity bottlenecks of data centers make power grids and electrification a “mandatory investment destination.” According to the IEA, grids are a bottleneck and the need for grid investments is growing significantly, while implementation times are long.
2–5 year driver: grid expansion and modernization, investments related to unlocking capacity (switchgear, transformers, automation), and connecting data centers to the grid.
Risk: component availability and project delays; additionally, some holdings are cyclical infrastructure operators.
Why better than a “general industrials ETF”: hits the bottleneck (the power grid) more directly, which the IEA highlights as critical. Role: the portfolio’s most important “datacenters meet electricity” satellite.

iShares Digital Security UCITS ETF (L0CK)
Chosen because the 2026 cyber trend is driven by both AI and geopolitical tensions, and Gartner highlights these in its 2026 cybersecurity trends. BlackRock
2–5 year driver: companies increasing security in AI environments (data, identities, cloud, SOC automation).
Risk: competition and pricing pressure in certain categories.
Why this ETF: according to justETF, this is a cost-competitive and large digital security index tracker. Role: the diversified cornerstone of the cyber theme, complemented by individual stock selection.

iShares Global Aerospace & Defence UCITS ETF (5J50)
Chosen because the defense spending trend has been upward, and NATO’s latest report highlights the growth of defense investments in Europe and Canada; SIPRI reports record levels of global defense spending.
2–5 year driver: order backlog growth and rearmament.
Risk: the theme may be “crowded,” and if the geopolitical situation unexpectedly normalizes, valuations could unwind. Hence the 4% weight.
Why global instead of just Europe: reduces the risk of a single region’s budget policy (diversification). Role: a “geopol-hedge” but with a controlled weight.

iShares Healthcare Innovation UCITS ETF (2B78)
Chosen because healthcare innovation combines demographic demand (aging) with new drug development approvals. The WHO estimates the 60+ population will grow strongly toward 2030, and the FDA reports a significant number of new drug approvals annually (e.g., the 2025 and 2026 lists).
2–5 year driver: innovation flow (drugs, diagnostics, medtech) and demand stability.
Risk: dispersion in the innovation theme is high (some companies are more “binary” biotech cases), so the weight is kept at 3%. Role: diversification and a more defensive growth component.

Schneider Electric (SU, Euronext Paris)
Chosen because the company reported strong growth and profitability in 2025: organic revenue grew, Data Center stood out as a growth driver in Q4, and adj. EBITA and margin improved. This is directly aligned with the data center electricity bottleneck and electrification theme.
2–5 year driver: electrification (distribution, automation), data center investments, and energy efficiency (electricity is expensive → optimization).
Risks: industrial cycle and capex timing; currency effects (as a global operator).
Why better than closest alternatives (e.g., other electrical engineering conglomerates): in this portfolio, it is chosen for “proven” growth and data center-driven momentum (the company’s own reported figures). Role: the portfolio’s “quality winner” in the electrical infrastructure theme.

Eaton (ETN, NYSE)
Chosen because Eaton reported record sales and strong profitability in Q4/2025; furthermore, order trends and backlog grew significantly, especially in the Electrical segments, and the company is providing guidance for growth and strong margins in 2026. This fits the “grid is a bottleneck” scenario, where demand is visible in order backlogs.
2–5 year driver: investments in the electrification and data center cycle, which drive up demand for electrical components. (Conclusion based on the IEA’s growing need for grid investments and the order/backlog momentum reported by Eaton.)
Risks: capex overheating and subsequent normalization; valuation if the market prices in a “perfect cycle.”
Why better than alternatives: in this portfolio, it is chosen because the company simultaneously reports (1) growth, (2) profitability, and (3) backlog visibility. Role: the second “cornerstone” in the electrical infrastructure theme alongside Schneider.Broadcom (AVGO, Nasdaq)
Selected because Broadcom is reporting extremely strong AI segment growth and visibility: according to the company, Q1/2026 AI revenue grew strongly (company earnings release), and Reuters reported on a long-term agreement with Google to develop and supply custom AI chips (TPUs), which improves predictability.
2–5 yr Driver: growth in custom ASIC/XPU demand, AI networking, and customer (hyperscaler) capex.
Risks: AI capex “breather”, customer concentration, and semi-cycle volatility.
Why better than closest alternatives: the selection emphasizes a profitable, cash flow-generating, already scaled AI infrastructure player rather than a small “single product” story. Role: a single “AI infrastructure” stock in the portfolio that complements the diversification of the semiconductor ETF.

Microsoft (MSFT, Nasdaq)
Selected because Microsoft reported strong revenue and operating profit growth (FY2025) and Azure business scale (over $75 billion annualized), as well as rapid Azure growth also in the FY2026/Q2 reporting.
2–5 yr Driver: growth in IT spending in 2026 and AI spending, where cloud and AI features are central. (Conclusion based on Gartner IT/AI spend forecasts and Microsoft’s own reported cloud figures.)
Risks: AI investment capex pressure may temporarily weigh on cash flow and margins; additionally, in the SaaS market, some categories may face pricing pressure due to AI agents.
Why better than “narrower” SaaS companies: this portfolio aims to avoid the binary disruption risk of a single application category and instead take broad exposure (cloud + platform + ecosystem). Role: the portfolio’s “SaaS winner without a lottery ticket.”

Palo Alto Networks (PANW, Nasdaq)
Selected because the 2026 cyber trend highlights pressure created by AI, the geopolitical environment, and regulation – and Palo Alto is reporting and launching solutions specifically for securing agentic AI (company IR news feed).
2–5 yr Driver: “mandatory” growth of cyber budgets for AI environments and critical systems.
Risks: competition and valuation, as well as the commoditization of parts of the cyber market.
Why this stock + cyber ETF: the ETF diversifies the sub-theme, and the stock brings “convexity” (if platforms win in the new agentic AI security layer). Role: strengthens cyber weight in a controlled manner.

D ETF Review

Below is a separate review row for each ETF (UCITS / EU listing / Nordnet):

ETF UCITS EU-Listed Available on Nordnet
Vanguard FTSE All-World UCITS ETF (VWCE, IE00BK5BQT80) Yes Yes (Xetra) Yes (Nordnet listing VWCE Xetra)
Amundi MSCI World Minimum Volatility Advanced UCITS ETF (WMMV, IE0001DKJVC2) Yes Yes (Xetra) Yes (Nordnet listing WMMV Xetra)
VanEck Semiconductor UCITS ETF (VVSM, IE00BMC38736) Yes Yes (Xetra) Yes (Nordnet listing VVSM Xetra)
First Trust Smart Grid UCITS ETF (GRID, IE000J80JTL1) Yes Yes (Xetra) Yes (Nordnet listing GRID Xetra)
iShares Digital Security UCITS ETF (L0CK, IE00BG0J4C88) Yes Yes (Xetra) Yes (Nordnet listing L0CK Xetra)
iShares Global Aerospace & Defence UCITS ETF (5J50, IE000U9ODG19) Yes Yes (Xetra) Yes (Nordnet listing 5J50 Xetra)
iShares Healthcare Innovation UCITS ETF (2B78, IE00BYZK4776) Yes Yes (Xetra) Yes (Nordnet listing 2B78 Xetra)

E Risk Analysis for the Entire Portfolio

Key Macro Risks

Energy price shock + inflation + interest rates: In the Iran scenario, energy prices have risen and institutions have highlighted inflation pressure, which can keep the interest rate environment tight and raise the cost of capital. This weighs particularly on high valuation multiples (long-duration growth companies).

Economic slowdown: The IMF has signaled the risk of slowing growth due to war and energy disruptions. This increases cyclical risk (industries, some semiconductors).

Data center electricity bottleneck: Although this is the portfolio’s “thesis,” it is also a risk if the bottleneck changes abruptly (e.g., streamlining of permit processes + investment boom turning into overcapacity). However, the IEA emphasizes the long lead time and investment need of grids, which makes a quick fix unlikely – but not impossible.

Key Sector Risks

Power infrastructure: The biggest risk is that customers postpone projects (capex cycle) or that component shortages/delays disrupt deliveries. According to the IEA, the bottleneck is real and building grids takes a long time, which can cause “undulation” in the order and delivery cycle.

Semiconductors: AI capex may take a “breather” (budget rotation), or geopolitical restrictions may affect the supply chain. Return potential is high, but volatility can be among the portfolio’s highest.

Cyber: Competition and commoditization; parts of the security layers may turn into “bundled features” within large platforms. However, Gartner sees AI and geopolitical tensions increasing cyber priority.

Largest Single Misassumptions1. Assumption that the growth in power grid investments materializes quickly as revenue and operating profit: if permitting and labor shortages slow down “execution” more than demand, some companies may see their order books grow, but earnings will materialize with a delay.

  1. Assumption that AI infrastructure investments remain additive rather than substitutive: if AI infrastructure “crowds out” the budget for application licenses, some software sectors will suffer (this is part of the background fear behind the SaaS slump).
  2. Assumption that the geopolitical risk premium remains high: if the Iran scenario ends quickly and the risk premium unwinds, defense and energy shock hedges could lose momentum, while “risk-on” could bring higher volatility back to the forefront of growth.

What Could Go Wrong Most

The worst combined scenario for the portfolio would be:

  • the energy price shock continues → inflation stays high → interest rates do not fall,
  • at the same time, global growth slows down,
  • and AI infrastructure investments slow down (capex pause).

In this case, both semiconductor and power infrastructure orders could normalize simultaneously, and valuation multiples would compress. The portfolio’s shock absorber is min-vol (WMMV) and a broad core (VWCE), but a drawdown would still be possible.

What Factors the Portfolio is Most Exposed To

  • Continuity of the power infrastructure and data center capex cycle (GRID + Schneider + Eaton).
  • Continuity of the AI infrastructure investment cycle (VVSM + Broadcom + Microsoft).
  • Cybersecurity “mandatory” spend and the attack surface created by AI (L0CK + Palo Alto).

F Critical Self-Assessment

Where the Portfolio is Strongest

The portfolio is at its strongest if two structural themes hold:

  1. Electricity is the bottleneck (grids, connection, capacity) and the need for investment grows, as the IEA describes; in this case, power infrastructure positions will get both a revenue and operating profit engine.
  2. AI infrastructure continues to grow, keeping semiconductor and cloud platform earnings strong.

In addition, the cyber weighting is reasonably “defensive growth”: Gartner links 2026 cyber trends directly to AI and geopolitical tensions.

Where the Portfolio is Most Vulnerable

Vulnerability is highest if AI capex and power infrastructure capex slow down simultaneously, or if the market prices in a higher required rate of return (interest rates) for the long term. This would directly hit the portfolio’s biggest themes (power infrastructure + AI infrastructure).

What Alternative Weighting Could Be Considered If the Market Outlook Changes

If the Iran scenario calms down quickly and inflationary pressure eases (interest rates fall), the portfolio could benefit from:

  • Reducing WMMV’s weight (e.g., 10% → 5%) and shifting part of the weight either to VWCE (core) or high-quality software winners (individual stock/ETF), because duration risk would ease. (Based on the idea that the interest rate environment heavily affects growth company valuations.)

If, on the other hand, the data center electricity bottleneck worsens and the need for grid investments accelerates, one could consider:

  • increasing the GRID + Schneider/Eaton allocation slightly, but with caution: the same theme already makes up 21% of the portfolio and there is overlap (GRID’s holdings already include power infrastructure companies).

If SaaS fears prove to be exaggerated and “AI features” turn into a driver for pricing (rather than a disruptor), one could consider:

  • increasing the weight in the high-quality cloud/software segment (e.g., by raising the MSFT weight or adding one broad, profitability-focused software ETF), while sticking to the principle of “no binary stories.”
    [/details]

I chose the following all-world index as a benchmark ETF: Vanguard FTSE All-World UCITS ETF - (USD) Acc ticker XETR:VWCE

And the portfolio suggested by AI is this, in case you don’t feel like reading the whole post. I made the purchases yesterday for Yahoo’s portfolio tracker.

Weight Ticker
30 % FTSE AllWld UCITS ETF USD A (XETR:VWCE)
10 % AmMSCIWldMinVltyAdvnUCITS ETFA (XETR:WMMV)
9 % VESemiconductorUCITSETFUSDA (XETR:VVSM)
9 % FTNasdaqClnEdgSGInfrUCITSETFAU (XETR:GRID)
5 % iShs Digtl Sec UCITS ETF USD A (XETR:L0CK)
4 % iShrsGblAerspc&DefUCITSETFUSDA (XETR:5J50)
3 % iSh Hlthcr Innov UCITS ETF (XETR:2B78)
6 % Schneider Electric SE (XPAR:SU)
6 % EATON CORPORATION PUBLIC LIMITED COMPANY (XNYS:ETN)
6 % BROADCOM INC. (XNAS:AVGO)
6 % MICROSOFT CORPORATION (XNAS:MSFT)
6 % PALO ALTO NETWORKS, INC. (XNAS:PANW)

Tracking starts here. I set up a 100,000-euro portfolio to make it easier to get full shares and units. Still, I had a little cash left over. If the AI suggests changes to the portfolio later, I will update this thread. At most 4 updates per year.

41 Likes

Maybe a silly comment, but if AI doesn’t actively manage investments, then the quality of the suggestions is measured precisely by long-term value growth.

My point is that many times my “best” investments have lost to the global index in the long run, even though for months/a couple of years I thought I was a genius.

4 Likes

I’ve been kind of doing that for half a year now (I spar and ask questions, and we bounce different perspectives around, then I finally make the decision myself) and so far the results have been quite okay. So, once a month, I do two things:

  1. I give the AI a list of the portfolio with its distribution and ask what it thinks
  2. And what it would change/replace

The only misses (so far) have been AMD, which hit a -20% stop-loss (Oct 29 → Nov 20), a bad miss, and AbbVie (since Dec 16), which is currently down almost -5%. But AbbVie’s history is so long and strong that it will probably be well in the black in a year.

But there have been many more successes:

  1. FSK lighten when the price was 15.38 USD (today 10.23 USD)
  2. Meta (Nov 13 → Nov 26) rose from 606 USD → 635 USD
  3. FSK lighten when the price was 16.14 USD
  4. FSK lighten (again) when the price was 15.14 USD
  5. Changing PMT (12.80 USD, then 12.01 USD) to Investor AB, current return +16.39%
  6. Selling VICI at 30.17 USD (today 27.76) and also changing to Investor AB
  7. Chevron (Jan 26 → Apr 1), rose from 167.34 → 195.10 USD (peaked around 215)

And two other changes, but so little time has passed with them that I can’t say much. They are, of course, in the black.

7 Likes

Quarterly update: Portfolio value increase of 22.93% over 3 months vs. benchmark ETF VWCE 14.83%.

A good return is, of course, just random luck, but it motivates me to continue the experiment. I almost hit a minor motivation slump with this tracking, because I have already moved on to investing with real money using agent loops run in Claude Cowork as a tool.

New portfolio created on July 7, 2026, using ChatGPT 5.5 Deep Research. The prompt is the same as it was 3 months ago. Two ETFs remained in the portfolio, three are new. Two stocks remained in the portfolio, three are new.

Below is the portfolio shown above in the Yahoo Finance portfolio tracker. The starting capital is the final value of the previous portfolio.

19 Likes

Thanks! It would be interesting to hear more about this Clause Cowork, could you elaborate? So, have you automated it?

3 Likes

Hi everyone, long time no see. I’m not sure if this is the right section for this—moderators, feel free to move it to a better place if it’s currently in the wrong section.

Down to business. I’ve spent a long time fine-tuning functional strategies using AI. The starting point has been capital protection, minimizing the percentage drawdown, and long-term viability starting from 2013. We’ll see how the strategy performs in both bear and bull markets. At the same time, we’ll see how and what the strategy picks for the portfolio at any given time. As benchmarks for the strategy’s performance, I’ve used the S&P 500 and the NASDAQ 100. If the strategy doesn’t clearly beat these indices, there’s no point in implementing it at all.

The data used in backtesting includes, among other things:

Entire NYSE/NASDAQ incl. delisted + SPY, pre- and post-market, ~2,519 trading days (a total of about 5 billion price bars across six timeframes), as well as older daily data from the 1990s, and intraday from 2004 onwards.

SEC data (EDGAR) — approx. 15 million rows

News data — approx. 6.5 million headlines
Derived research datasets — approx. 57 million rows

The amount of data run through is absolutely massive (several hundred gigabytes) across various stages while polishing the strategy toward completion. Of course, one’s own input is also essential; while AI is a great tool and sifts through data efficiently, it still only does what you ask it to, and the results are often just “ballpark.” You need a fairly deep understanding of the stock market and trading when developing something like this. With all this data and backtesting, we generate insights into why a price moved up or down, what elements potentially influenced it, whether there are recurring patterns in the market, etc. After all this, we can start testing different parameters that form the foundation of the strategy, compare it against the aforementioned indices (NYSE/NASDAQ), and tweak (optimize) parameters until the strategy either works or doesn’t. If the strategy can’t be made to work (clearly beating the indices), it gets scrapped, and we start rebuilding from scratch.

If the strategy works, we keep it.

Now we’ve reached the point where a working strategy has been found: it has been backtested, walk-forward analysis has been run (the strategy validation phase), and paper tracking plus the scanner are ready (we validate the strategy in real time following the rules with real trades, but without real money).

In fact, there are already two working strategies, with more on the way, though they are still in different testing phases.

So now, I decided to document this backtest phase in video format as a record for myself and for anyone who happens to be interested in the topic. In the videos, the backtest starts from 2017. First up is the dollar (USD) version, which includes monthly contributions with two different savings amounts and shows how that affects portfolio growth and the final outcome. Even though that isn’t strictly related to backtesting, some might find it interesting, and running this version alongside the Euro version was quite convenient.

Once that dollar version has been rolled out, it will be immediately followed by the Euro version, where the backtest has intentionally been run with a small initial capital—a starting portfolio size that is fully accessible to any Finn. The test uses an equity savings account (osakesäästötili) and its real costs.

The test takes everything possible into account: currency conversion fees, real-time EUR/USD exchange rates, dividends, and spread (the price difference between bid and ask offers, which constitutes a hidden cost)—essentially as realistic a scenario for a Finnish investor as possible.

Apologies for the rambling post, but if anyone is interested, the first 2 dollar episodes of this 17-part series are available to watch on YouTube (link at the end of the post), and the remaining dollar episodes will be updated on a fast schedule. Right after those dollar episodes are out, the Euro episodes will be added. I’ll post updates here when new episodes go live, and it’s worth following the channel if the topic interests you.

One final note: currently, the scanner indicates that the first real buys will execute in the portfolio in early December. There will also be a separate update on the channel regarding portfolio performance and tracking, which I plan to update roughly once a week, covering the key news and events for the stocks held in the portfolio.

What makes this interesting is that viewers get to see how the portfolio’s holdings shift along the way as dictated by the strategy! An episode will be made for each shift so followers stay up to date, and then, say, in a year’s time, we can wrap up and see how the portfolio and Meri did with real money. Then we can all laugh together at what a crappy (or functional) strategy Meri built! :joy:

Link to the YouTube playlist where the series is updated: https://youtube.com/playlist?list=PLELcNXsDx8-8&si=XOAUqMX2h46Rsmv1

8 Likes