Same cloud, three different ways to earn the money back

They look like the same business. The path from AI to cash is not.

Microsoft, Alphabet and Amazon are all giant cloud operators. All three build data centers, buy GPUs, design their own AI silicon and sell models and agent platforms to enterprise customers.

From the outside it looks like one business run three times.

The capital recovery path is different.

Microsoft pushes AI into workflows that already sit inside the enterprise: Microsoft 365, GitHub, Azure, Security, Dynamics. Alphabet defends the enormous consumer surface of Search and YouTube while binding Gemini, TPU and Google Cloud into a single vertically integrated stack. Amazon builds on AWS as the base layer of corporate IT, integrating compute, storage, databases, security, models and its own chips.

Which means you cannot judge all three by the same number.

For Microsoft, what matters is paid seats, Azure consumption and cloud gross margin. For Alphabet, it is the defense of Search, cloud growth and the economics of the TPU. For Amazon, it is AWS growth, operating income and the payback period on infrastructure.

The AI capex may look identical. The channel it flows back through is not.

Microsoft: the strongest weapon is not the model, it is the workflow that already exists

Read Microsoft's AI strategy as a model race and you miss the point.

Microsoft's strongest asset is that it is already inside the building.

Windows, Microsoft 365, Teams, GitHub, Azure, Dynamics, Security.

Companies already run these products. So Microsoft does not have to sell a new AI product from a standing start. It puts AI into the documents, email, code, meetings, CRM and security operations that are already in use.

Azure and other cloud revenue grew 43 percent in the fourth quarter of FY2026. Paid Microsoft 365 Copilot seats passed 30 million. Foundry reached 100,000 customers, and the number of Foundry customers running more than one trillion tokens on an annualized basis was four times higher than a year earlier. Agent 365 registered roughly 40 million agents within two months of launch.

These numbers say more than "lots of people use AI."

They say Microsoft is extending monetization from revenue per human seat to revenue per unit of agent usage.

The real test is cloud gross margin, not Copilot seats

Thirty million paid Copilot seats is a strong signal. But the number that matters more to investors comes right after it.

Microsoft Cloud gross margin was 65 percent in the fourth quarter of FY2026. That is down from 68 percent in the same quarter a year earlier, and from 69 percent for FY2025 as a whole.

More AI usage lifts revenue. It also lifts the cost of GPUs, networking, power and data center depreciation.

So Microsoft's AI success cannot be judged on how fast Copilot seats accumulate.

Does cloud gross margin stabilize after Copilot and agents have lifted ARPU and Azure consumption?

That is the far more important question.

Microsoft is the company best positioned to demonstrate AI monetization fastest. It is also the company that will show fastest how heavy the capital cost of AI sits on an income statement.

What Microsoft is really aiming at is the operating system for enterprise AI

Once agents multiply inside a company into dozens, hundreds, thousands, a new management problem appears.

Who built this agent. What data does it reach. Which model does it use. How much is it spending. Who is accountable.

Microsoft wants that management layer too.

That is why Agent 365 matters.

If a company's existing identity, security, compliance and governance systems can be extended to cover agents, Microsoft stops being a supplier of AI apps and starts looking like the operating system for enterprise AI.

In that case the moat runs much deeper than model quality.

You can swap out a model. Swapping out a company's identity system, security policy, data permissions and workflows is a far harder job.

Alphabet: the most exposed company, and the best armed

Alphabet's position is different from Microsoft's.

For Google, AI is an opportunity and a threat to the existing Search economy at the same time.

If generative AI changes the search interface, the click structure that traditional search advertising rests on can change with it. Alphabet's AI investment is therefore offense and defense at once.

The Q2 2026 data suggests the defense is holding up rather well so far.

Google Search and other revenue was 63.27 billion dollars, up 17 percent year on year. YouTube ads came in at 11.1 billion dollars, up 13 percent. Google Cloud grew 82 percent to 24.77 billion dollars.

So far, AI is producing Search growth and cloud expansion at the same time rather than cannibalizing Search on contact.

Alphabet's biggest investment question is not whether Gemini is a good model.

It is whether Google can keep connecting user intent to advertiser budget even after the interface for search changes.

That matters far more.

A heavier number than 82 percent cloud growth: a 513.9 billion dollar backlog

The eye-catching figure in Alphabet's Q2 2026 is the 82 percent growth at Google Cloud.

For a long-term investor, the backlog may matter more.

At the end of June 2026, Alphabet's remaining performance obligations stood at 519.5 billion dollars. Of that, 513.9 billion dollars relates to Google Cloud.

The company expects to recognize slightly more than 50 percent of that backlog as revenue within the next 24 months.

That number shows Alphabet's enormous AI capex does not rest on future expectation alone.

Contracted enterprise demand already exists.

One caution. Starting in the first quarter of 2026, Alphabet changed its methodology to include contracts with original terms of one year or less in the backlog. Straight comparisons with prior periods need care.

Even so, a 513.9 billion dollar cloud backlog is a strong signal that Google Cloud is no longer a distant third behind AWS and Azure, but a core platform absorbing large-scale enterprise demand in the AI era.

The TPU is one of Alphabet's most important economic weapons

What makes Alphabet's AI structure unusual is that it owns its silicon.

The TPU is not simply an alternative to buying NVIDIA.

Google has enormous internal workloads in Search, Gemini, YouTube and Cloud. So it can design a chip, validate it at scale internally first, and then offer it to cloud customers.

That structure is powerful.

Chip design, model, data center, cloud service and end user can all be optimized inside one company.

In Q2 2026, Alphabet began recognizing revenue for the first time from sales of TPU systems deployed in customer data centers.

That means Google is extending the TPU from an internal cost-reduction tool into an external revenue line.

Over the long run, a key variable in Alphabet's AI economics is not only how good Gemini is, but how much lower a cost-to-outcome the TPU can deliver against NVIDIA.

Alphabet's risk: cloud goes well, and cash flow feels the squeeze first

After Q2 2026, Alphabet raised its 2026 capex outlook to a range of 195 billion to 205 billion dollars.

The reason given is very strong cloud demand.

But large infrastructure investment comes back as a burden on the income statement and cash flow with a lag.

Alphabet's capex in the first half of 2026 was 80.6 billion dollars. That is more than double the 39.6 billion dollars in the same period a year earlier.

Depreciation on a data center starts the moment it is finished.

So Alphabet's most important AI test is not the cloud growth rate.

It is whether cloud operating margin holds at a high level while the cash generation of the Search business keeps carrying the AI infrastructure build.

The more realistic risk is not the scenario in which AI destroys Google. It is the one in which AI grows Google, and the capital cost of that growth turns out to be far larger than expected.

Amazon: the company selling AI most like enterprise infrastructure

Of the three, Amazon can look the least glamorous in consumer AI.

From an enterprise AI standpoint, that may be a strength.

Enterprise AI is not one model and done.

It needs data.

It needs storage.

It needs databases.

It needs security.

It needs networking.

It needs compute.

AWS already has every one of those layers.

So Amazon's AI strategy reads less like "we will build the best chatbot" and more like "we will be the base infrastructure companies run AI workloads on."

Bedrock, AgentCore, Trainium, Graviton, S3, Redshift, Aurora and the security stack all sit inside the same structure.

The more AI becomes the new workload of corporate IT, the more AWS can raise spend per existing customer.

What 37 percent AWS growth actually means

Amazon's Q2 2026 AWS revenue was about 42.2 billion dollars.

That is roughly 37 percent higher than a year earlier.

AWS operating income was 16.6 billion dollars.

On a simple calculation, that is an operating margin of about 39 percent.

The interesting part is that AWS operating income is holding up strongly even as AI infrastructure investment surges.

Amazon's logic for recovering AI capital is very concrete.

Build the data center.

Fill it with servers, including Trainium and Graviton.

Enterprise customers consume compute.

That consumption repeats.

That consumption becomes AWS revenue and operating income.

If Microsoft monetizes AI through work seats plus usage, and Alphabet through search defense plus cloud, Amazon is the closest to recovering money purely through infrastructure usage.

What Amazon's 53.1 billion dollars in capex means

Amazon's cash capex in Q2 2026 was 53.1 billion dollars.

That is 69 percent higher than the 31.4 billion dollars a year earlier.

Company disclosures say most of it is technology infrastructure investment, with a substantial portion going to support AWS growth.

Fifty-three point one billion dollars in a single quarter is a very large number.

But you should not read the absolute capex figure alone and call it overinvestment.

In the same quarter, AWS operating income was 16.6 billion dollars.

Amazon's total operating income was 27.5 billion dollars.

Amazon's AI investment is burning enormous cash and it is being done on top of a business already producing large-scale operating income.

The real risk is not that capex is big.

It is the case where AWS growth and operating income growth slow while capex stays high.

Amazon's own silicon: not just margin, but control of supply

Trainium and Graviton are not simply tools for cutting the NVIDIA bill.

In-house silicon does three things.

First, it lowers cost-to-performance on specific workloads.

Second, it reduces dependence on external GPU supply shortages.

Third, it creates a price and performance combination available only inside AWS.

Through 2026, Amazon has been emphasizing that the Trainium and Graviton businesses have each reached a large run rate.

This matters because the long-run competition in cloud is shifting from who holds the most GPUs to who can drive total cost of ownership lower from silicon all the way through to service.

Microsoft's Maia, Alphabet's TPU and Amazon's Trainium all point the same way.

The hyperscalers are NVIDIA's customers and are becoming compute suppliers in their own right.

The three moats are in three different places

Microsoft's moat is enterprise workflow, identity, security and software distribution.

Alphabet's moat is the vertical integration linking the user surface of Search and YouTube with Gemini, the TPU and Cloud.

Amazon's moat is the scale and breadth of AWS as enterprise infrastructure, plus operating efficiency including its own chips.

Which is why comparing the three on one model benchmark ranking is meaningless.

Microsoft can win in enterprise AI without building the best model itself.

Alphabet can come out of the AI era stronger if it holds most of search advertising while growing Cloud and the TPU.

Amazon can produce the steadiest monetization without being flashy in consumer AI, so long as it owns the base infrastructure for enterprise workloads.

That is the difference between them.

Who proves AI capital recovery first

It is hard to pick one of the three as the AI winner.

Microsoft shows the fastest monetization. Copilot seats, agents and Azure consumption already put AI directly on a price tag.

Alphabet has the broadest vertical integration, connecting its own silicon, models, Search, YouTube, Cloud and Android. If it works, it is strong on both cost structure and distribution.

Amazon is the most infrastructure-centric. The bigger enterprise AI gets, the more data and workloads move to AWS, and usage-based revenue accumulates on top.

In the end the investor's question is not who does AI best.

It is who turns one dollar of AI capex into the highest long-term cash flow.

That is the real race.

Sources

• Microsoft FY2026 Q4 Earnings
• Microsoft FY2026 Q4 Earnings Call
• Alphabet Q2 2026 SEC Earnings Release
• Alphabet Q2 2026 Form 10-Q
• Amazon Q2 2026 Form 10-Q

Insight Times Editorial Desk