Berkshire Dumped Amazon. The Numbers Say It Was a Portfolio Tidy-Up, Not a Thesis Call.

Berkshire Hathaway sold its last 2.3 million shares of Amazon in the first quarter of 2026, closing a position Warren Buffett opened in 2019 after calling himself an "idiot" for passing on the stock for years. The market took the exit as a vote of no-confidence on runaway capital spending and negative free cash flow. Then Amazon reported Q2 earnings, and the bear case looked even thinner than it did on paper.

The sale wasn't a thesis call. It was a portfolio housecleaning. The 2.3 million shares represented a rounding error in Berkshire's $263 billion equity portfolio, and they were part of a broader liquidation of stocks managed by Todd Combs, who left for JPMorgan in December. Greg Abel - Buffett's successor - eliminated at least 15 positions in Q1, from Visa to Domino's to Charter. Amazon was collateral damage in a personnel shuffle, not a conviction bear call.

What matters isn't who owns the shares. It's whether the underlying math supports the stock heading into the third quarter. And right now, the numbers are running in the opposite direction of the narrative that supposedly justified the exit.

Here's the breakdown.

1. AWS is accelerating, not decelerating.

Amazon Web Services revenue grew 37% in Q2 to $42.2 billion - its fastest growth in 18 quarters and the fifth straight quarter of acceleration. Five quarters ago, AWS grew 17%. The annualized revenue run rate is now $169 billion, which would rank AWS 24th on the Fortune 500 as a standalone company. More importantly, AWS backlog reached $496 billion, up triple digits year over year. That backlog is contracted revenue, not aspirational guidance. Andy Jassy said on the earnings call that Amazon still won't have enough capacity to meet demand in 2026, and likely not in 2027 either. "The demand we already have for 2028 is striking," he told analysts.

This isn't a business peaking. It's a business in the middle of a re-acceleration curve.

2. Profitability is expanding alongside growth, not being sacrificed for it.

AWS operating income jumped 64% to $16.6 billion, pushing the segment's operating margin to 39.4%, up from 32.9% a year ago. Amazon's AI business has surpassed a $25 billion annualized revenue run rate and is growing at a triple-digit percentage. Its custom-chip business (Trainium and Graviton) is above $25 billion as well, up from $20 billion just three months earlier. Both are growing faster than the core cloud business, which means AWS isn't just getting bigger - it's getting more profitable at the margin.

Jassy described AI adoption as "barbelled": frontier labs consuming massive compute at one end, enterprises harvesting cost savings at the other, with the middle - current enterprise production workloads - still mostly on the sidelines. He conceded he doesn't know whether the middle will accelerate as fast as the labs. But the base is $169 billion and growing at 37%. The math doesn't require every segment to triple-digit for the whole to compound.

3. The negative free cash flow is a timing gap, not a model problem.

Amazon's trailing free cash flow turned negative - $7.6 billion over the past 12 months - after the company spent $169 billion on property and equipment against $161.4 billion in operating cash flow. Capital spending in Q2 alone was $53.1 billion, and Jassy raised the full-year 2026 capex forecast to $220 billion, up from $200 billion, blaming inflated memory chip prices.

That's the number the bears fixate on. But the mechanism Jassy laid out on the call explains why it's a timing issue. Data centers require capital roughly two years before servers are installed and start generating revenue, but once open, they monetize for 30-plus years without repeating that upfront cost. Servers and networking equipment break even in under three years against a five-to-six-year useful life, and most of Amazon's AI capacity is contracted for five years or longer. Jassy's framing: "We'll spend a lot of capex and encounter free cash flow headwinds until these data centers come online, can be monetized, and we get a few years into these servers being utilized."

In other words, Amazon is front-loading spending on assets that will pay for themselves multiple times. The negative FCF is the gap between construction and occupancy, not a sign the business is burning money. Most hyperscalers are in the same boat - Alphabet turned cash flow negative for the first time on record, and Meta's cash generation plummeted 91% - but Amazon's contracted backlog gives it the cleanest visibility into when the spending converts to revenue.

4. The valuation disconnect is enormous.

Amazon trades at 3.86 times trailing sales. Microsoft, the closest comparable hyperscaler, trades at 11.0 times. Alphabet sits at 10.4 times. Meta is at 6.6 times. Yet AWS is re-accelerating, with 37% growth, a 39.4% operating margin, and a $496 billion backlog - metrics that the market's current multiple fails to reflect.

At 3.86x sales, the market is pricing Amazon as if its most profitable, fastest-growing segment is the problem rather than the engine. The stock has rallied roughly 20% since the Q2 earnings report, closing the gap a bit, but the forward multiple still doesn't reflect an AWS business on a path that Jassy himself described as a "$1 trillion revenue business." Even if you discount that ceiling aggressively, a $169 billion cloud business growing at 37% with a 39.4% operating margin doesn't justify a 3.8x sales multiple in a market that pays 11x for slower-growing peers.

5. The catalyst just fired.

The Q2 earnings report flipped the narrative. Before Amazon reported, the entire AI trade was caught in anxiety over runaway capex with no clear return path. Amazon provided the example other hyperscalers lacked: accelerating revenue, expanding margins, a $496 billion backlog, and management that walked through exactly how the spending converts to profit. Wedbush called it the "cleanest beat" among the hyperscalers; Evercore's Mark Mahaney said the growth rate "is just the breakout that the stock needed." The stock rallied 20-25% from pre-earnings levels in the week after the report.

The stock has already started to re-rate. But the re-rating is incomplete - 3.86x sales with AWS at a $169 billion run rate and $496 billion in backlog is still a discount to where the fundamentals point.

The risk

The real question isn't whether AWS is growing. It's whether enterprise AI adoption - the "middle of the barbell" Jassy described - arrives fast enough to sustain the buildout. Jassy conceded he doesn't know the trajectory of that segment. If enterprise inference adoption stalls, the capex pipeline could outpace revenue for longer than the current timeline assumes, and free cash flow could stay negative well into 2027. That's the scenario where the stock stalls.

But the contracts are five years, the backlog is $496 billion, and the margin is 39.4%. The downside requires enterprise AI to be a dud, not just slow.

The verdict

Berkshire's Amazon sale was a portfolio mechanic, not a market signal. The 2.3 million shares were tiny, the timing coincided with a manager departure, and the fundamental case for the stock has strengthened, not weakened, since the exit. AWS is accelerating at a pace the market hasn't priced in, the backlog is contracted rather than aspirational, and the capex story is a timing gap with a defined monetization path.

At 3.86x trailing sales, Amazon trades at less than half the multiple of Microsoft despite running a cloud business that is re-accelerating with higher margins. The stock doesn't need to be cheap on earnings to be attractive - it needs the AWS re-acceleration to persist, and the Q2 print suggests it already is. The gap between where the stock trades and where the backlog points is the edge. Whether that edge widens or narrows from here depends on enterprise AI adoption, but the contracted demand already on the books gives the thesis a floor most growth stories don't have.