How much AI do you own? Adding up ETF, VC funds and angel stakes

An investor who holds a world equity ETF, a few venture funds and a handful of angel stakes owns artificial intelligence three times over, and usually counts it once. Added up across all three, the share of a portfolio that depends on the AI build-out can easily be several times what the owner would have guessed. Whether valuations are a bubble is a question nobody can answer in advance. How much of one theme you own is a question you can answer this weekend, and it is the more useful one.

Where the concentration stands in October 2025

Start with the public side. At the end of August 2025 the ten largest companies made up 39 per cent of the S&P 500, and Nvidia alone about 8 per cent, according to Boston Trust Walden. Ten years earlier the top ten were 18 per cent and the largest company, Apple, about 3 per cent. US shares in turn make up more than 70 per cent of the MSCI World, the index behind many of the ETFs German private investors hold. A product bought in order to own everything has become, to a considerable degree, a holding in a few American technology groups whose investment plans and share prices now turn on AI.

The private side moved faster. On 2 September Anthropic announced a USD 13 billion financing at a post-money valuation of USD 183 billion. On 2 October current and former OpenAI employees sold USD 6.6 billion of shares at a valuation of USD 500 billion, up from USD 300 billion in the company's previous round. In PitchBook's venture data, AI companies took 63 per cent of the dollars invested over the most recent twelve months, against 40 per cent a year earlier.

And the two sides are linked by contract. On 22 September Nvidia and OpenAI announced that Nvidia intends to invest up to USD 100 billion in OpenAI as OpenAI deploys at least ten gigawatts of Nvidia systems. The largest position in the public index is financing the largest private company, which buys its products.

We draw no conclusion about prices from this. Concentration has been rising for ten years, and the index rose with it. The point is narrower: whatever happens to this theme happens to all three parts of the portfolio together.

A worked example: counting it three times

The numbers are invented for illustration. Say a portfolio of EUR 3 million looks like this:

• EUR 1,000,000 in an MSCI World ETF
• EUR 500,000 current value in three venture capital funds, with another EUR 400,000 of unfunded commitment
• EUR 300,000 in eight angel investments, at the price of their last rounds
• EUR 900,000 equity in two rental flats
• EUR 300,000 in cash and bonds

Asked how much AI is in there, the owner thinks of the three angel companies that have it in their pitch: EUR 150,000, or 5 per cent.

Now look through each layer. For the ETF, open the provider's holdings list and add up the companies whose valuation you believe rests on AI spending: the chip makers, the large cloud platforms, the suppliers of data centre equipment. Where to draw the line is a judgement. Say you arrive at a quarter of the ETF: EUR 250,000.

For the venture funds, take the last quarterly reports. Each lists the portfolio companies with their valuation, and your share of the fund's NAV applies to each of them. Say the companies you would call AI make up 40 per cent of the three funds' combined value: EUR 200,000.

The angel stakes are EUR 150,000, as before.

Together that is EUR 600,000, or 20 per cent of the portfolio. Four times the first answer, and without a single new purchase.

One more line belongs on the sheet. The EUR 400,000 of unfunded commitment will be called over the next years and invested by managers who are, on PitchBook's numbers, putting most of their money into AI companies. The exposure grows without any decision on your part.

Why the three buckets do not diversify each other

On paper they are different asset classes: listed shares, fund interests, direct stakes. In a correction of the theme they are one position with three reporting dates.

The ETF shows it the same day. The venture funds show it two or three quarters later, because private valuations take their multiples from listed comparables and from the price of the next round, and both arrive with a delay. We described that lag in the tariff shock and the denominator effect. The angel stakes show it last and most abruptly, when a company has to raise money again and the new round is lower, or does not happen.

That order is dangerous for planning. For half a year after a fall in listed AI shares, the private part of the portfolio would look stable. It would not be stable. It would be unreported.

The reverse also holds, and it is why we make no bubble call. If the build-out pays off, all three parts gain, and an investor who sold the listed part early has cut the only part that could be cut and kept the parts that cannot be sold at all.

What to do with the number

Write it down. One sheet: each holding, its value, the share you count as AI, the source and date of that estimate. The definition you used matters less than using the same one next quarter. A number that can be compared with its own past is worth more than a precise one.

Decide on a range before the news does. Some investors will look at 20 per cent in one theme and find it right; they may know the field and have chosen it. Others will find it twice what they are comfortable with. Either view is defensible. Not knowing the figure is not.

Use the parts that can be steered. Existing fund interests and angel stakes cannot be adjusted. Three things can. The next fund commitment: ask the manager how much of the previous fund went into AI companies and what is intended for the new one. Follow-on investments in your own stakes, where each pro rata right is a new decision. And the liquid part, the only one that can change within days. Which way to adjust is your decision and depends on the range you set. We are describing the levers, not recommending a trade.

Repeat it when the reports arrive. The fund reports for the third quarter usually start arriving in November. They carry the company list and the valuations, which is the raw material for this sum. Valued reads the portfolio companies and their valuation changes from each quarterly report and keeps them with the fund, so the look-through does not start from a stack of PDFs each time. A spreadsheet with one row per company does the job too, if it is updated.

The exercise takes an afternoon the first time. Its result is a single percentage that no depot statement and no fund report will ever show you.