13F Portfolio Analysis — Peter Thiel | He Sold NVIDIA and Bought Electricity
Peter Thiel sold NVIDIA: how to keep the theme but cut the exposure
In the second half of 2025, Peter Thiel’s investment firm Thiel Macro sold its entire NVIDIA stake.
For the next two quarters, its 13F showed no holdings at all.
Then came the second quarter of 2026.
72% of the newly disclosed portfolio was filled with power and energy names. The largest position was Amazon; the rest was made up of a power generator, regional utilities, a nuclear startup and an Argentine energy company.
On the surface, it looks simple.
Sell NVIDIA, buy power stocks.
But put this portfolio into DIVA Quantizer and a somewhat different picture appears.
The earlier portfolios were “Rodeo Riders,” moving 1.4–1.7 times as much as the market. The June 2026 portfolio became an “Immovable Rock,” with a market beta of 0.85 and annual volatility of 18.0%.
Yet 85% of its risk is concentrated in three stocks: Amazon, Vista Energy and Vistra.
In other words, Thiel didn’t simply reduce risk.
He moved it somewhere else.
In this piece we run seven quarters of Thiel Macro’s 13F filings through DIVA Quantizer to see what risk structure he used to put the market’s shift toward AI and power into a portfolio.
1. What the 13F shows first — he didn’t abandon AI
Thiel Macro’s portfolio changed in three big steps between the end of 2024 and the second quarter of 2026.
In Q2 2025, NVIDIA made up 40.1% of the portfolio. Add Tesla and it was a very strong bet on growth and technology stocks.
Then came Q3 2025.
NVIDIA was gone.
In Q4 2025 and Q1 2026 that followed, the 13F showed $0 in holdings.
Then, in Q2 2026, a $419 million portfolio reappeared.
| Holding | Weight | |
|---|---|---|
Amazon AMZN | 28.2% | |
Vista Energy VIST | 18.1% | |
Vistra VST | 14.1% | |
AEP AEP | 10.1% | |
DTE Energy DTE | 9.6% | |
FirstEnergy FE | 9.5% | |
CMS Energy CMS | 9.4% | |
X-energy XE | 0.9% |
What matters here is what newly appeared.
NVIDIA, ASML, Apple, Microsoft and Tesla did not come back.
Instead, Amazon and Vistra returned, surrounded by power generators and utilities.
Amazon runs AI data centers and consumes power.
Vistra is a power generator.
X-energy is developing next-generation nuclear reactors, in an investment round led by Amazon.
The utilities run the grids in regions where data centers are multiplying.
So rather than a simple “basket of power stocks,” this portfolio is closer to one laid out along a single industrial chain.
AI → data centers → power demand → generation → the grid
Which raises a question.
Has this portfolio really become safer than before?
This is where we bring in DIVA.
2. The first question for DIVA — how much did the portfolio’s character change?
We turned each quarter’s holdings into its own portfolio and compared them in DIVA over the same period.
Every portfolio uses the same data: 72 months, July 2020 to June 2026.
| 2024.12 | 2025.03 | 2025.06 | 2025.09 | 2026.06 | |
|---|---|---|---|---|---|
| Market beta | 1.55 | 1.48 | 1.68 | 1.36 | 0.85 |
| Value beta | −0.93 | −0.55 | −1.13 | −1.13 | +0.21* |
| Volatility | 41.3% | 30.7% | 43.0% | 36.1% | 18.0% |
| Diversification | 16.4% | 26.5% | 21.3% | 13.3% | 42.9% |
| DIVA type | Rodeo Rider | Rodeo Rider | Rodeo Rider | Rodeo Rider | Immovable Rock |
From the end of 2024 through Q3 2025 the portfolios were consistent.
They moved more than the market and leaned hard toward growth stocks.
The June 2025 portfolio in particular had a market beta of 1.68 and annual volatility of 43.0%.
That is why DIVA classed it as a “Rodeo Rider.”
By June 2026, though, everything is different.
Market beta: 0.85.
Volatility: 18.0%.
The diversification effect climbed to 42.9%.
The important thing is that this should not be read as “he switched to value stocks.”
The value beta is +0.21, but it is not statistically significant.
A more accurate way to put it is this.
A portfolio that bet heavily on growth stocks set the growth bet down altogether.
Did the bet on AI disappear too, then?
No.
The way he holds AI changed.
It moved from chips and consumer tech to power and infrastructure.
3. The second question for DIVA — so where is the risk?
With market beta down to 0.85, the portfolio looks fairly defensive overall.
But look at DIVA’s risk contribution and the story changes.
| Holding | Weight | Risk contrib. | Market beta | Annual vol. |
|---|---|---|---|---|
| Amazon | 28.4% | 34.4% | 1.11 | 32.2% |
| Vista Energy | 18.3% | 29.8% | 1.07 | 48.8% |
| Vistra | 14.2% | 20.6% | 1.20 | 41.3% |
| Three-stock subtotal | 61.0% | 84.8% | ||
| AEP | 10.2% | 5.2% | 0.47 | 20.4% |
| DTE Energy | 9.7% | 3.4% | 0.49 | 17.4% |
| FirstEnergy | 9.6% | 3.0% | 0.34 | 21.3% |
| CMS Energy | 9.5% | 3.6% | 0.43 | 18.0% |
| Four-utility subtotal | 39.0% | 15.2% |
Here the portfolio’s structure comes into focus.
The four utilities, 39% of the weight, produce only 15% of the risk.
Conversely, Amazon, Vista and Vistra, at 61% of the weight, produce 84.8% of the risk.
So the portfolio can be seen in two parts.
The engine
Amazon, Vista and Vistra.
Market betas around 1.1–1.2, and high volatility.
This is the part that carries both the return and the risk.
The anchor
AEP, DTE, FirstEnergy and CMS.
Market betas around 0.3–0.5.
Their job is to lower the whole portfolio’s market sensitivity and create the diversification effect.
In other words, Thiel didn’t eliminate risk. He deliberately separated the risky holdings from the ones that lower risk.
4. The third question for DIVA — is the anchor really four anchors?
Can the four utilities really be counted as four different holdings?
DIVA’s correlations suggest something is off.
AEP, CMS and DTE have correlations of 0.81–0.89 with each other.
That means they move almost together.
In other words, he bought four stocks, but from a risk point of view they are hard to call four genuinely different bets.
Amazon’s correlations with the utilities, by contrast, are around −0.08 to 0.05.
Vista Energy’s correlations with almost every other holding are also low, at about −0.04 to 0.13.
That is why the portfolio’s diversification effect rises to 42.9%.
Bring in DIVA’s HRP as well, and the result gets even more interesting.
5. What DIVA’s HRP shows — what if it were truly defensive?
We put the same seven stocks into DIVA’s HRP (Hierarchical Risk Parity) model.
HRP doesn’t just split weights equally. It groups stocks that move alike, then allocates weights so the portfolio’s risk isn’t concentrated in any one group.
The result is quite different.
Allocated by risk, HRP would have put more than 80% in the four utilities.
The actual portfolio does the opposite.
If reducing risk had been the goal, the portfolio would most likely have looked closer to HRP.
It didn’t.
The utilities got only 39%, and the other 61% went to Amazon, Vista and Vistra.
So it is hard to see this simply as a “safe power-stock portfolio.”
It is closer to this structure.
An aggressive AI-and-energy bet kept in place, with utilities cushioning part of its risk
6. The odd one out — why is Vista Energy here?
Vista Energy holds a slightly unusual place in this portfolio.
It isn’t a utility, and it isn’t an American company.
It is an Argentine oil producer.
Yet its weight is a hefty 18.1%.
DIVA’s Fama–French five-factor analysis shows something distinctive, too.
FF5 explains only about 15% (R²) of Vista’s moves.
For Amazon it is about 71%.
In other words, Vista’s price moves are poorly explained by the traditional equity factors of market, size, value, profitability and investment. Its correlations tell the same story: −0.04 to 0.13 with each of the other six holdings, so it effectively moved on its own.
Of course, this alone can’t tell us why Thiel bought Vista.
But within the whole portfolio, Vista looks like an independent bet that the shared AI-power theme alone can’t explain, unlike the other holdings.
That is another feature of this portfolio.
It is built around a core theme, yet one separate high-conviction stock is given a large position.
7. Thiel’s market view, read through DIVA — what matters more than why he sold NVIDIA
Let’s pull the results together.
① He didn’t abandon AI. He changed where in AI he invests.
Chipmakers such as NVIDIA and ASML and tech stocks such as Tesla, Apple and Microsoft disappeared.
In came Amazon, Vistra, X-energy and regional utilities.
In other words, the focus moved from companies that build AI to companies that supply the power and infrastructure AI needs.
② He lowered the portfolio’s overall risk but kept an aggressive bet.
The earlier portfolios had market betas of around 1.4–1.7.
By June 2026 it had fallen to 0.85.
Yet the risk contributions show three stocks, Amazon, Vista and Vistra, accounting for 84.8% of the total risk.
In other words, he didn’t eliminate risk; he divided up its roles.
· Amazon, Vista, Vistra → the core bet that takes on high risk
· Regional utilities → the cushion that lowers market sensitivity
DIVA’s HRP also confirms that the actual portfolio is quite different from a structure that spreads risk as evenly as possible.
③ So the heart of this portfolio isn’t “power stocks” but “AI combined with power.”
Amazon consumes power in its AI data centers.
Vistra produces power.
X-energy is aiming at the nuclear supply chain of the future.
The regional utilities run the grids where data centers are multiplying.
They look like different industries, but they connect into a single flow.
AI data centers → rising power demand → generation → grid investment
What looks like a simple stock swap in the 13F alone reads, through DIVA, as a bet on a single industrial structure.
8. And the twist DIVA reveals — low beta doesn’t mean low risk
So far, the June 2026 portfolio looks fairly stable.
Market beta 0.85, annual volatility 18.0%.
The four utilities, 39% of the weight, have market betas of only 0.34–0.49.
But what actually happened after the end of June was not what you’d expect.
From June 30 to September 30, 2026, the four utilities fell 14.2% on average.
As a result, about 95% of the portfolio’s −5.75% loss came from the utilities.
| Return | Contribution | |
|---|---|---|
| Whole portfolio | −5.75% | −5.75%p |
| Four utilities AEP −12.6 · DTE −19.3 · FirstEnergy −8.0 · CMS −16.9 | −14.2% | −5.49%p |
| Amazon · Vista · Vistra Amazon +4.5 · Vista +2.4 · Vistra −12.6 | −0.1% | −0.07%p |
| X-energy | −22.8% | −0.20%p |
| For reference: S&P 500 | +2.4% | |
| For reference: US 10-year yield | 4.42% → 5.29% |
Why did the low-beta utilities produce most of the portfolio’s loss?
Because the stock market and interest rates are different risks.
Utilities have stable cash flows and pay dividends. So investors treat them as stocks that are at the same time compared with bonds.
When interest rates rise, things change.
The higher Treasury yields go, the less attractive utilities’ relatively low dividends become, and their share prices can come under pressure.
Over this period, the US 10-year Treasury yield in fact rose from 4.42% to 5.29%.
In other words, the utilities were relatively insensitive to the stock market but sensitive to interest rates.
DIVA’s Fama–French five-factor analysis points to a similar limit.
FF5 explains about 26% of the four utilities’ moves.
The other 74% can’t be explained by the five equity factors of market, size, value, profitability and investment.
That means there can be risks that DIVA’s standard FF5 model doesn’t include, such as interest rates.
A low market beta means a holding moves less with the stock market. It doesn’t mean every risk is low.
In this portfolio the utilities were the anchor that lowered market risk, but when rates moved sharply, the anchor shook too.
So when you analyze a portfolio, don’t stop at “how sensitive is it to the market?” Go on to ask “what risks is it exposed to besides the market?”
9. When you read a 13F, go one step past the weights table
Looking at a 13F, people usually ask this.
“What did this investor buy?”
With DIVA alongside, you can change the questions.
Wrapping up
The investor who sold NVIDIA did not abandon AI.
He changed how he invests in it.
From chips to electricity.
And to Amazon, which uses the power; Vistra, which produces it; and the regional utilities that run the grid.
In the 13F alone, this is a simple stock swap.
Open up the risk structure with DIVA, though, and a different picture appears.
The old portfolios were “Rodeo Riders” that moved more than the market.
The new one became an “Immovable Rock” with a market beta of 0.85.
Yet even inside that rock, 85% of the risk was concentrated in three stocks.
The utilities weren’t the center of the returns but the anchor that lowered risk.
And even that anchor could shake once rates rose.
So the most interesting thing about this portfolio is not that he bought power stocks.
It is which risks he chose to take, and which he chose to reduce.
If what you buy is your market view, where you put the risk is your design.
Try analyzing a portfolio yourself in DIVA Quantizer.
Factor exposures, risk contributions, correlations and the portfolio’s structure, invisible in a weights table, start to show.
① A 13F shows only US-listed stocks (and some options) at quarter-end. Cash, bonds, foreign-listed shares, private stakes, short positions and derivative hedges are not included, and purchase timing and prices are unknown.
② The June 2026 risk contributions, correlations, HRP and FF5 results exclude X-energy (listed April 2026, 0.9% of the 13F total) because of its short price history, with the other seven weights rescaled. It is included in the post–June 30 performance.
③ To compare quarters, every portfolio uses the same 72 months (July 2020 to June 2026). Entering the same stocks into DIVA today uses the latest six years, so the figures may differ slightly.
④ Of the June 2026 portfolio’s FF5 betas, only the market beta is statistically significant (size beta is borderline, p≈0.10). The past 72 months include large run-ups in Vista Energy and Vistra, so the expected return and Sharpe ratio should not be read as forecasts.
⑤ The cause of the utilities’ decline (rising rates) is an interpretation based on rate moves over the same period, not something the DIVA models measure directly.
This piece is for education and information only and does not recommend buying or selling any security or copying any 13F.
Data · 13F: SEC EDGAR 13F-HR, Thiel Macro LLC (CIK 1562087), Q4 2024 to Q2 2026 (Q4 2025 and Q1 2026 filed with no holdings) · prices: yfinance adjusted close; analysis uses 72 month-ends from July 2020 to June 2026, post–June 30 performance uses closes from June 30 to September 30, 2026 · factors: Ken French Data Library, North America FF5 · risk-free rate: US 13-week T-bill, 4.00% annual · rates: US 10-year Treasury yield (^TNX) close · X-energy listing and Amazon’s investment: TechCrunch (Apr 24, 2026). All risk, factor, HRP and risk-contribution metrics are DIVA Quantizer model outputs.
This article was written with the help of AI.