Peter Thiel Sold Nvidia and Went All-In on Cloud Giants — Here's What the Filing Really Says

Peter Thiel's hedge fund just sold every share of Nvidia and went all-in on cloud giants. The headline-grabbing stock is Amazon — up a jaw-dropping multiple since its 1997 IPO — which makes for a dramatic click. But the actual filing tells a more specific story than the clickbait suggests, and it's worth reading carefully because the bet Thiel is making says something about where AI money flows next.

Thiel Macro, Peter Thiel's hedge fund, filed its Q2 2026 13-F in mid-August. The fund holds about $419 million, and two companies dominate it: Microsoft at roughly 34% of the portfolio and Amazon at 28%. Together they make up 62% of everything the fund owns. The rest is energy and utility names. Nvidia, which the fund held last year, is gone — completely sold in Q3 2025. Tesla was cut 76%. Palantir, Thiel's own co-founded company, has been systematically reduced over the past two years.

The message is clear: Thiel is shifting from AI silicon to AI infrastructure. He doesn't believe the biggest AI opportunity is in the chip that powers the models. He believes it's in the platform that hosts them.

That's a legitimate investment thesis. But is it a good one right now — at these prices?

The case starts with Microsoft, where the thesis is easiest to test because the numbers are public and accelerating. Azure, Microsoft's cloud platform, grew 43% year-over-year in the most recent quarter, with total Azure revenue for fiscal year 2026 crossing $100 billion for the first time. That's double-digit growth on a $3.75 trillion market cap. More importantly, the acceleration is real — Azure grew 40% in Q3, then 43% in Q4, and management is guiding for approximately 45% in the next quarter. The growth acceleration came from easing capacity constraints: Microsoft added 88 data centers throughout the year and cut the time for new GPUs to go online by nearly half.

The moat question is harder to dismiss here than with most tech giants. Microsoft doesn't just host AI models; it has the largest enterprise software distribution channel in the world. Microsoft 365 Copilot now has over 30 million paid seats, and customers with more than 50,000 seats have grown more than sevenfold year-over-year. The time for a new Copilot user to reach an 80% monthly active rate has shortened from months to days. That's not adoption in the sense of a trial — it's the kind of rapid integration that suggests once an organization puts Copilot in, it doesn't come out. Microsoft Foundry, its AI platform, hosts over 11,000 models, and the number of customers using multiple model providers has increased fivefold. The platform is designed so that enterprise data, permissions, and governance stay locked into Microsoft's ecosystem regardless of which underlying model is used. That's moat architecture.

Then there's the monetization layer. Microsoft is shifting from pure seat-based Copilot billing to a "seats plus usage" model. GitHub Copilot consumption revenue grew over 60% quarter-over-quarter with improving margins. The commercial backlog — the remaining performance obligation that represents future revenue — jumped 84% to $678 billion. That number matters because it's not a forecast; it's contracted work that Microsoft is obligated to deliver and will be paid for.

Valuation is where the case gets tested. Microsoft trades at a trailing P/E of 28x and a forward P/E of 36.8x. Revenue grew 18% for the full fiscal year, operating margin expanded to 46.8%, and return on invested capital sits at 28%. The PEG ratio comes to about 0.89, which means the forward multiple is actually lower than the revenue growth rate. That's not dirt cheap, but it's not a stock where the price has outrun the numbers, either. The company generated $67 billion in free cash flow for the year while spending $116 billion on capital expenditures — an enormous buildout, but one that's producing accelerating revenue growth.

Now compare that to Amazon, which Thiel added as a 28% position. Amazon's cloud division, AWS, grew 37% in Q2 2026 — its fastest growth rate in 18 quarters — reaching a $169 billion annualized revenue run rate. AWS generates roughly 60% of Amazon's operating profit despite being about 17% of total revenue. That margin disparity is the reason the thesis works: as AWS grows faster than the retail commerce business, Amazon's overall profitability expands at a rate that doesn't match its top-line growth. Amazon trades at a trailing P/E of 20.4x — notably cheaper than Microsoft on an earnings basis — and its PEG ratio is 0.23, the lowest among major tech peers.

But Amazon carries a different risk profile. Its free cash flow turned negative last quarter at -$11.6 billion trailing twelve months, because the company spent $173 billion on capital expenditures while generating $161 billion in operating cash flow. That's a brutal buildout cycle. Amazon also has $544 billion in total debt versus Microsoft's $316 billion, and Amazon's operating margin of 11.5% is nowhere near Microsoft's 46.8%. Amazon is the higher-growth, lower-margin, higher-capex play. Microsoft is the lower-growth, higher-margin, higher-cash-flow play. Both are valid, but they're not the same bet.

Here's what the Thiel move doesn't tell you: whether these prices already bake in the thesis. A fund with $419 million in assets making concentrated bets on the two largest cloud platforms isn't a signal of broad institutional conviction. It's one small fund's specific view. The 13-F filing is also backward-looking — it covers Q2, so the actual purchases could have been made months ago at significantly different prices. Microsoft has risen 23.5% over the past 120 days and Amazon is up 20% over the same period. You're not buying into a contrarian moment; you're buying two stocks that have already moved.

The Nvidia exit is the more interesting part of the filing. Thiel sold the chip leader that's been the most visible AI beneficiary for two years. The signal isn't that Nvidia is broken — it grew revenue at 70%+ rates through most of 2025 — but that the money capture in AI may shift from the hardware layer to the platform layer. Once data centers are built and GPUs are deployed, the ongoing revenue stream flows to the cloud provider that owns the rack space, the enterprise relationships, and the software distribution. Nvidia sells the engine; Microsoft and Amazon own the highway.

That's a reasonable framework, but it requires patience. Cloud infrastructure investments pay off over quarters and years, not weeks. The $116 billion Microsoft spent on capex this year and the $173 billion Amazon burned through aren't generating proportional earnings today. They're building capacity for demand that may not fully materialize for another fiscal year or two. If that demand delivers, both companies re-rate on the margin expansion that comes with operating leverage on deployed infrastructure. If it doesn't, the capex cycle becomes overhang.

The bottom line: Thiel's filing points to a specific, defensible thesis — AI infrastructure over AI silicon, cloud platforms over chips. The fundamentals support the direction. Azure growing 43% and AWS accelerating to 37% show the platform layer is winning, and Microsoft's enterprise distribution moat is real. But the prices already reflect a lot of this. There's no valuation disconnect here, no overdone selloff, no forward P/E converging with the market while growth runs at multiples of it. These are expensive companies with expensive growth, and the opportunity is in the execution, not the entry price.

If you're watching rather than holding, the setup to track is whether Azure and AWS maintain their acceleration into the next fiscal year while margins recover from the capex buildout. That's when the thesis either compounds or stalls.