An in-depth look at the hyperscalers; My penny-pinching parents
I concluded yesterday's e-mail by writing:
I continue to be bullish on four of the five hyperscalers: Alphabet (GOOGL), Amazon (AMZN), Meta Platforms (META), and Microsoft (MSFT).
As I've outlined in many previous e-mails, they have highly profitable, dominant businesses that will benefit from their huge investments in AI – even if (or when) the bubble bursts.
The Financial Times chart below captures the key to my bullishness...
It shows a huge decline in the hyperscalers' free cash flow ("FCF") for 2026 and 2027 due to hundreds of billions of dollars in capital expenditures. (In fact, based on the latest earnings and guidance – which aren't included in the chart – FCF will likely be negative for all but Microsoft for both years.)
But then, estimates show there will be a massive resurgence as the spending pays off – with FCF soaring to more than $600 billion at the end of the decade:
This future FCF is, of course, uncertain. That's why these stocks are currently trading at roughly market multiples despite having far above-average businesses. But that's the bet that the companies – and investors in their stocks – are making.
Two of the smartest investors I know, Dan Loeb of Third Point and my college buddy Bill Ackman of Pershing Square (PS), share my view on the hyperscalers.
Dan wrote in his third-quarter letter:
Amara's Law states that we tend to overestimate the impact of a technology in the short run and underestimate it in the long run. We believe the AI trade is transitioning from the former to the latter: investors are beginning to underestimate AI's long-term potential after initially overestimating its near-term impact.
Numerous anecdotal examples demonstrate meaningful productivity gains from AI, yet the evidence remains less apparent in aggregate economic data, lending some credence to the bubble thesis. We remain confident that demand for compute will continue to grow, although the ultimate distribution of that demand between open-source models and proprietary large language models remains uncertain. Regardless of which models prevail, the need for compute infrastructure should persist, and we believe data centers will benefit.
We believe hyperscalers (and certain related project finance structures) currently offer the most attractive way to maintain exposure to the theme. We do not expect issuance fears to materially change their cost of capital: the incremental financing needs are relatively modest compared with their balance sheets and cash flow generation. Hyperscaler access to relatively inexpensive capital provides a significant advantage as the market works through the current wave of infrastructure investment and financing.
Bill went even more in-depth in his third-quarter letter, writing at length about the hyperscalers in general and about Amazon, Meta, and Microsoft specifically. (He bought Alphabet in 2023, but sold it earlier this year.)
In a section entitled "The Hyperscalers' Growth and Return Opportunity Is Underappreciated," he wrote:
We believe that the hyperscalers' capital spending is offensive rather than defensive. Rather than merely protecting existing franchises, we believe the hyperscalers are investing to accelerate growth and are deploying capital behind a transformative technology that plays directly to their competitive advantages. These include unmatched distribution, long-standing customer relationships, proprietary data and context, security and reliability at global scale, and the balance sheet strength to invest at a level few competitors can match...
Each of these companies has a differentiated and substantial opportunity to monetize its compute, whether on a first-party basis through its own applications or through the sale of cloud services, and each is uniquely positioned to benefit from the technology it is building.
Regarding Amazon, he highlighted how AI is strengthening its Amazon Web Services ("AWS") division and agentic shopping:
Amazon's compute investment directly leverages AWS's dominant position as the largest cloud provider. Even at its $170 billion annualized revenue scale, AI has helped AWS accelerate its revenue growth to 39% in the third quarter, its fastest growth in nearly five years, despite nearly doubling in size over this period...
AI is also strengthening Amazon's retail business where agentic shopping is emerging as a new channel for both commerce and advertising. More than 350 million customers have used Alexa for Shopping, Amazon's agentic shopping assistant, over the past 12 months, and these customers spend more per order and convert more often. Across both of its major businesses, we believe AI is strengthening Amazon's competitive position while adding to its already substantial growth runway.
Turning to Meta, Bill argued that AI is giving the company a way to monetize beyond its core social media and advertising businesses:
For Meta, AI and the associated investment in compute are transforming its business in two important ways. First, Meta's core digital advertising business has effectively used AI for better ad targeting, generative ad copy, and more effective content recommendation. These benefits are already visible in the company's revenues, with growth accelerating from 22% in 2025 to an estimated 27% in 2026. Second, AI is unlocking entirely new products, namely consumer and business agents, which Meta can launch at a scale, that we believe few companies can match.
Entering the year, the magnitude of Meta's compute investment exceeded what many investors believed its core business alone would support. It was also unclear what new products could justify the incremental investment. In recent months, the company has addressed that question directly with a series of launches, including its Muse Spark and Wonder Full family of models, which are now at the frontier, the Muse personal agent, and Meta Enterprise Platform.
Muse appears to be a resounding success, now surpassing ten million downloads within weeks of its launch...
Prior to Meta's recent product launches, investors had limited visibility into how Meta's consumer and business agent efforts would take shape, but now the narrative on the stock has quickly shifted as investors increasingly recognize the multiple paths to monetizing Meta's compute assets beyond its core business. Meta's earnings multiple has increased from 18 times to 22 times. We continue to view Meta as undervalued given the momentum in its core business and the large new addressable markets that Muse and Meta Enterprise Platform provide to the company.
Lastly, regarding Microsoft, he wrote that the narrative around the stock has shifted now that its AI investments look to be paying off:
Microsoft occupies an enviable position among the hyperscalers in having two large channels by which to monetize its compute. Like Meta, it has a first-party channel in Microsoft 365, its productivity suite embedded in daily workflows of over 450 million users. Like Amazon, Microsoft's Azure, the second-largest cloud provider after AWS, offers a third-party channel, which surpassed $100 billion in revenue last fiscal year while growing 41%...
As with Meta, investors have started to reassess how beneficial AI could be to Microsoft's growth as its AI strategy is now starting to show in its recent business results. Microsoft's [price-to-earnings] multiple has increased from 19 to 25 times over the past three months as the narrative on the stock has shifted from the company being an AI laggard to an AI leader. We believe Microsoft's current valuation understates the degree to which AI will accelerate Microsoft's growth and deepen its entrenchment with customers.
Bill concluded that the hyperscalers' growth will likely exceed analysts' estimates:
In our view, analyst estimates penalize the hyperscalers for the cost of the buildout today while ascribing limited potential to the significant future growth in revenue and earnings that are likely to result from these investments. Most of the revenue acceleration visible in 2026 reflects capital deployed in 2024 and 2025, while we believe the larger levels of investment now will yield an additional increase in growth in the future. As a result, we believe it is highly probable that the hyperscalers' revenue growth will exceed analyst estimates over the next several years. To the extent that does not occur, we believe the hyperscalers will reduce their capital expenditures from current levels. In either scenario, we believe the hyperscalers' capital intensity will be lower and their free cash flow will be meaningfully higher over the next several years than what is implied by current consensus estimates.
He also shares my concerns about how companies like CoreWeave (CRWV), Oracle (ORCL), and SpaceX (SPCX) will fund their massive investments – an issue the hyperscalers don't face to nearly the same degree:
The availability and cost of financing required to fund a buildout of this magnitude has been another area of increasing investor concern. The four hyperscalers discussed above are collectively expected to spend a little over $1 trillion on capital expenditures in 2027 based on consensus estimates. Including Oracle and SpaceX and neoclouds such as CoreWeave and Nebius, total data center capital expenditures are expected to reach approximately $1.5 trillion next year based on consensus estimates. Investors have grown concerned about how this will be funded, as reflected in the widening of credit spreads on neocloud debt and certain hyperscaler asset-level financings as well as in the recent sharp declines in neocloud share prices.
In our view, the scrutiny of the neoclouds and the funding needs of companies such as SpaceX and Oracle should not be conflated with the hyperscalers' ability to fund their own capital expenditures. The hyperscalers are in a vastly stronger financial position. They have virtually no net debt and can finance the substantial majority of their capital expenditures from the prodigious cash flows of their core businesses. Their multi-trillion-dollar market capitalizations also mean that even large equity issuances, in the hundreds of billions of dollars, would amount to a low- to mid-single-digit percentage of their shares outstanding...
The hyperscalers already operate large installed bases, have ready access to capital, and are developing custom silicon that reduces their dependence on external chip supply. They also have been planning land, power, and cooling infrastructure on multi-year lead times. The more difficult the buildout proves for others, the more valuable the hyperscalers' capacity becomes, and the more durable their competitive position.
I agree entirely with Bill.
CoreWeave remains on my "Stinky Six" list of stocks to avoid – which was the "Stinky Seven" until November 5, when I decided to remove AppLovin (APP).
As I wrote back on September 25, I was even tempted to add Oracle to this list after a slew of bad news about its data-center build-outs.
And, of course, I still think SpaceX is one of the most overhyped, overvalued large-cap stocks of all time. As recently as November 6, it was severely underperforming both the S&P 500 Index and my "Discarded Dozen" list.
Best regards,
Whitney
P.S. I welcome your feedback – send me an e-mail by clicking here.
P.P.S. As you read this, I'm on my way back to New Hampshire to swap places with my sister and spend the week with my parents. My dad is getting better every day, but he's still weak after six days in the hospital, so my mom needs the help.
They've been staying in a big house on Lake Sunapee by themselves since Labor Day and, not surprisingly, it's expensive to heat. The heating-oil truck came last week, and the bill was $928, which was quite a shock. So they turned the thermostat down to barely above freezing, wear their heavy winter coats, and huddle under blankets.
It's driving my sister crazy! There's a daily tug-of-war between them. It's truly hilarious what my parents will do to save a buck...


