Hyperscalers at the Inflection

The largest stock-market winners in the AI era have been semiconductors and power equipment companies. These “picks and shovels” providers have seen massive revenue and profit growth during this capex supercycle.

Conversely, the companies funding the buildout – principally Alphabet, Amazon, Microsoft, and Meta – have lagged. The market worries their investment binge will not earn an adequate return. But look closely, and recent results suggest otherwise. The payoff is beginning to show, and analysts’ estimates haven’t caught up. We see hyperscaler financial results inflecting upward and growing evidence of attractive returns on investment.

Cumulative Total Return Since March 2023

Revenue Growth is Accelerating

In the last quarter, revenue growth across the Big Three cloud businesses – Amazon Web Services, Google Cloud, and Microsoft Azure – accelerated about ten percentage points sequentially to 48% year-on-year driven by surging demand for AI cloud as well as resilient demand for legacy cloud services.

This is faster growth than during the last major boom in the cloud industry during Covid. The pandemic catalyzed an accelerated digital transition as enterprises shifted workloads to the cloud for scalability, security, and costs. During 2021, the Big Three cloud platforms grew revenue around 40% year-on-year.

In other words, these businesses are currently growing significantly faster than they did at the peak of the Covid boom, despite having a revenue scale nearly 4x larger than in 2021. In the second quarter of 2026, the Big Three added $13 billion of revenue sequentially. This is roughly equivalent to an entire quarter of AWS revenue at the beginning of 2021.

Big Three Hyperscalers Cloud Revenue YoY

Analyst Estimates Haven’t Caught Up

This acceleration was unanticipated just a few quarters ago and remains underappreciated. Wall Street analysts model a slowdown from here. We think there are three reasons the market underappreciates the durability of this phase of AI-led cloud demand growth.

First, Google Cloud’s revenue growth could take another step up as external sales and deployment of its internally designed tensor processing units (TPUs) scale. TPU system sales represent a new revenue stream for Google and are difficult to model given limited disclosure. Anthropic is already a large user of Google TPUs and has committed to substantially expand its direct TPU purchases in 2027.

Second, hyperscalers disclose how much capex they will spend, but not how much compute capacity they have or are adding. The market is underestimating the amount of capacity that comes online in 2027-2028. We believe investors need to model not just capex, but the units of compute that spending buys, as well as the timeline for energizing it.

Third, current spot pricing for AI compute is rising while hyperscaler fleets remain contracted at older, lower prices, so the gap between what their capacity earns and what it could earn is widening. Rising spot prices are driven by the current supply/demand mismatch for compute, as well as improved monetization power by their customers, the frontier labs, as AI capabilities grow. While it is unlikely that hyperscaler blended prices reach current spot rates, we do expect hyperscaler revenue per gigawatt to rise over the coming years. This will be driven by new compute contracting at higher rates and the revenue mix shifting to higher-value managed services, such as databases, security, and software, that customers consume alongside compute infrastructure.

The Payoff is Visible

The spending is converting to profit sooner than feared. Google Cloud operating margins hit 36% last quarter, an all-time high, up 15 points from a year ago. AWS margins were 38%, up 5 points year-on-year. Also, operating margins include depreciation associated with infrastructure already placed into service, so improving profitability is occurring even as the depreciation burden from the buildout rises.

Moreover, return on invested capital (ROIC), when measured to reflect the timing lag between when capex is spent and when capacity comes online, already shows Alphabet ROIC expanding consistently in each of the last three years. A similar analysis for Amazon shows ROICs broadly stable despite the massive growth in invested capital. This evidence echoes comments from company management teams.

“[The] return on invested capital is very compelling. We’ve done this before in the first era of cloud computing, just over a longer time horizon, where demand built more gradually than it has in AI. But we see the margins and returns in AI tracking what we saw with core at the same point of evolution, actually a little ahead.” – Amazon CEO Andy Jassy, 2Q26 earnings call

Risks to Monitor

The market is worried about uncertain returns on capex, free cash flow turning negative, and asset-light businesses becoming capital intensive. We believe the emerging evidence points to attractive returns on current capex, as they were during the first era of cloud computing which made the hyperscalers large and valuable businesses in the first place.

Further risks we are monitoring are customer concentration and the supply/demand timing mismatch. Two customers – OpenAI and Anthropic – account for a large portion of the hyperscalers’ cloud revenue and backlog, which gives them greater bargaining power than a diverse and heterogeneous enterprise customer base for legacy cloud. Second, capacity is being added quickly in ever-growing amounts. If future demand growth undershoots supply, cloud returns could deteriorate.

The Largest Fleets, the Clearest Roadmaps

We see a bright future for the largest hyperscalers. They have the largest installed fleet of strategically valuable AI compute. They have the clearest, funded roadmaps to significant capacity growth in the years ahead. Their fleets increasingly run on custom-designed silicon, allowing them to earn superior economics. And they have the ability to further enhance those economics through attaching proprietary managed services.