Public Cloud Earnings: Is Cloud Revenue Clearing the Hurdle Rate

This is the key chart of the board. The green line is the three clouds’ combined revenue growth, the red line is the four hyperscalers’ combined depreciation growth, and the depreciation line is the hurdle rate. Margins hold only while revenue outruns depreciation; today the green line is on top. But the bar does not stay here: the depreciation projection in group 4 shows depreciation growth climbing to 53 to 55% in 2027, so today’s 55.7% is not yet enough. Read the trend, not a single quarter.

The three clouds’ revenue growth on one chart, latest quarter Google Cloud 102.6%, Azure 48.2%, AWS 39.3%. All three are around 40% with Google Cloud jumping to 80%, a steady recovery from the early-2023 trough.

One panel per company: the red line is the company’s own depreciation growth (dashed where projected), the green line is its cloud revenue growth (Meta has no cloud segment, so company-wide revenue is used), and the grey dashed line is the 2027 threshold. The numerator is company-wide depreciation, so the bar is systematically high; the right use is to judge how thick the cushion is and which way it is heading. Google is the only one beating the bar, still ahead even with all company depreciation charged to Google Cloud; Amazon is on the line and one more step up in AWS growth clears it, and its shortened server life makes its bar the most honest; Microsoft has a wide gap but the strictest measure, with Office depreciation in the numerator; Meta’s gap is structural, because its AI returns come through advertising, so faster depreciation presses straight on margins.

AWS at $46B a quarter, Azure $33B, Google Cloud $31B, all three still accelerating. Azure is Microsoft's restated September 2026 definition, excluding GitHub, developer services, Security Copilot and the healthcare cloud; quarters before 3Q24 are estimates projected back from the actuals and drawn dashed. Google Cloud includes TPU system sales from 2Q26, so it is not fully comparable with earlier quarters. The three charts below are each company's revenue bars with its growth line.

Left: revenue each cloud added over the same quarter a year earlier. The three combined went from about $6B in 2Q23 to $39B in 3Q26, the most direct revenue evidence that AI investment is paying back. Right: each company's share of the combined increment; installed share moves slowly, so the share of new revenue is what shows competitiveness today. Azure's share is slipping and Google Cloud's is rising, and one reading is that in-house chips are the deciding factor.

Left: each company's annualized AI revenue; solid dots are the company's own disclosed anchors, the rest is interpolation or back-solving between anchors, drawn dashed. Right: annualized AI revenue divided by four and by the quarter's cloud revenue, a gauge of how much of the rebound in cloud growth AI accounts for; AI is already 10 to 40% of quarterly cloud revenue, with Microsoft's broader AI definition running high.

Left: all three backlogs jumped in the latest quarter, AWS $578B, Google $568B, Microsoft's company-wide RPO $683B, mostly long contracts signed by model companies: about 45% of Microsoft's cloud orders are OpenAI-related, OpenAI plus Anthropic are about a third of AWS's, and Google's includes Anthropic. Right: capex divided by backlog; the ratio has been fairly stable historically, so capex is effectively governed by the order book, and a falling ratio means orders are growing faster than spending, leaving room for capex to be revised up. The three measures differ, so read each trend on its own.

AWS 38.7%, Microsoft Intelligent Cloud 35.2%, Google Cloud 36.2%, all three between 35 and 40%, with Google Cloud having climbed there from a loss. Microsoft's Intelligent Cloud segment disappears from fiscal 2027, leaving only the dashed Microsoft Cloud gross margin to follow, which is a gross margin rather than an operating margin, so read its direction only.

Left: each company's depreciation and amortization, accelerating every quarter. Right: the four combined, dollars plus year-on-year growth; the pale bars and dashed line are a mechanical projection from capex already spent plus analyst estimates at fitted depreciation lives, not company guidance, and the growth rate keeps rising into 2027, which is why the hurdle rate climbs.

Left: capex as a multiple of depreciation, Google 6.1×, Meta 4.9×, Microsoft 3.8×, Amazon 3.6×; multiples this high mean depreciation must accelerate, and only a return to about 1× marks the end of the investment cycle. Right: company-wide depreciation divided by cloud segment revenue; the numerator includes non-cloud depreciation, so read the trend rather than the level. It is falling, which means cloud monetization is outrunning depreciation, and a turn upward is when to worry.

Left: assets bought but not yet in service, which neither depreciate nor earn, Google $145B, Meta $89B; a record pile means capex is running ahead of the ability to power up and commission. Amazon discloses only annually, Oracle only in its May fiscal year, Microsoft never. Right: the same stock expressed as months of capex parked, like days of inventory, Google about 11 months, Meta about 10; flat means commissioning keeps up with spending, rising means it is falling behind. The third chart is capex efficiency: new annualized cloud revenue per dollar of earlier capex, with 0.33 as the three-year revenue payback line, a pass mark rather than a good one, since it is revenue paid back, not profit.

Who has changed server useful lives, by how much, and what it did to earnings. Extending the life flatters current earnings and pushes the depreciation tail into the future; Amazon is the only one that moved the conservative way, so its hurdle rate is the most honest. What to watch: anyone else shortening AI server lives is admitting GPUs have a shorter economic life, which pushes the whole industry’s hurdle rate up; conversely, another extension by Meta would be a red flag.