Cloud market share got a fresh set of numbers this week. In the September quarter, Microsoft’s Azure reached $33.2 billion, up 48%, according to AFP’s report on Microsoft’s earnings. Google Cloud posted $30.7 billion, up 103%. We could not confirm AWS’s September quarter revenue or growth rate, so we do not print them. AWS remains the largest provider by revenue, but the two challengers are growing much faster.
That reversal is what makes this quarter worth a close read. Three of the biggest technology companies reported within 48 hours of each other, and each one sped up. AWS posted its fastest growth since the fourth quarter of 2021, according to CNBC. Azure cleared consensus expectations of 45%. Google Cloud more than doubled. If you pick a cloud for the next three to five years, these filings tell you more about capacity, pricing power and product direction than any feature checklist. For a service-by-service view, see our broader AWS, Azure and Google Cloud platform comparison.
Here is how this article works. It lays out the Q3 2026 scoreboard, converts it into share math, and then checks what each cloud is doing with the money: backlog, capex, AI models, token prices and scheduled price changes. It ends with use-case picks, a migration walkthrough, pros and cons, and a verdict. One warning up front. The three segments are not like for like. Google Cloud’s reported revenue includes Workspace subscriptions as well as infrastructure, and Microsoft only began reporting Azure in dollars this quarter, per CNBC. Treat every share figure below as a directional estimate, not an audited ranking.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Cloud Market Share in Q3 2026: What the Numbers Say
Two kinds of data describe cloud market share. The first is third-party tracking, mostly from Synergy Research Group, which estimates spending on cloud infrastructure services across the whole market. The second is what the three companies disclose in their own earnings. The two do not match perfectly, and the gap between them is where most of the confusion comes from.
Start with Synergy. In its Q2 data, according to AFP, AWS stayed number one by revenue, Microsoft held second place and Google ranked third. Those are the most recent third-party rankings our sources confirm. We did not find a verified Synergy or Canalys percentage for any recent quarter, so we do not print one. Synergy usually publishes its quarterly cut a few weeks after the earnings calls, and the Q3 numbers will likely land in November or December.
The direction matters more than the level. Google Cloud, at $30.7 billion, sits close to Azure at $33.2 billion, and its growth rate of 103% is far above Azure’s 48%. If growth rates hold even partway, the order of second and third place is a live question for 2027.
For a head-to-head on the two largest, read our AWS vs Azure breakdown. The short version of the market share story is simple. AWS keeps the biggest base, Azure is closing the gap on a large base, and Google Cloud is growing from a smaller base at a pace nobody else matched this quarter. The next sections show how each one got there.
Q3 2026 Scoreboard: Revenue, Growth, Margin and Run Rate
The table below pulls the headline numbers into one place. Reported figures come from the companies’ earnings releases as covered by CNBC, AFP and Constellation Research. Rows marked “calc.” are our arithmetic on those reported figures. Rows from Synergy and Stack Overflow are older and measure different things, so read them as context rather than as a score.
| Metric | AWS | Microsoft Azure | Google Cloud |
|---|---|---|---|
| Q3 2026 revenue | Not listed | $33.2B | $30.7B |
| Year-over-year growth | Not listed | 48% (constant currency) | 103% |
| Operating income | Not listed | Not disclosed separately | $11.12B |
| Operating margin (calc.) | Not listed | Not disclosed | 36.2% |
| Annualized run rate (Q3 x 4, calc.) | Not listed | Not listed | $123B |
| Share of big-three revenue (calc.) | Not listed | Not listed | Not listed |
| Size relative to AWS (calc.) | 100% | 72% | Not listed |
| Latest backlog disclosed | Not listed | Not confirmed | $561B (Q3) |
| 2026 capex guidance | Not listed | About $170B (Microsoft, per AFP) | $195B to $205B (Alphabet) |
| Parent free cash flow | Not listed | $18.6B (quarter) | -$9.07B (quarter) |
| Synergy Q1 2026 share | 28% | Not listed | 14% |
| Pro developers using it (Stack Overflow 2025) | Not listed | Not listed | Not listed |
Microsoft’s cloud story is harder to score because the broader Microsoft Cloud line blends Azure with Microsoft 365 commercial cloud, Dynamics and LinkedIn. We do not cite an Azure growth rate here because we could not confirm one from a source we trust. The recent signals are product-side: the Microsoft Copilot harness in Foundry Agent Service, announced October 28, 2026 in public preview, and the Azure Maia 300 inference platform.
Capacity is the quickest way to gauge the leader’s position. Amazon CEO Andy Jassy reportedly said on October 23, 2026 that much of AWS capacity for 2027, and some for 2028, was already sold. We could not tie that claim to a primary AWS or Amazon release, so treat it as unconfirmed. It would still suggest how much demand the largest provider believes it has.
AWS: Scale Leader Expanding in Models and Agents
The mix of what is growing matters. Jassy said core compute and storage grew at their fastest rates in several years, which counters the idea that all new cloud growth is AI training. Constellation Research reported that AWS Lambda crossed a $1 billion annual run rate. Jassy also said backlog for prior-generation Graviton instances more than doubled quarter over quarter after the Graviton 5 launch, a sign that cheaper Arm compute is a real buying driver. If you compare custom silicon across the three clouds, our look at Graviton5 against Cobalt 200 and Axion covers the chips in detail.
AWS also changed its stance on AI hardware. Jassy said Amazon intends to make Trainium available outside AWS late in 2027, copying the route Google took with its TPUs. During the quarter AWS announced plans to expand its Nvidia GPU portfolio, bought data analysis startup DuckLabs (the agreement was signed August 26, according to AWS), and signed deals with Meta and OpenAI, per CNBC.
For buyers, the AWS signal is capacity confidence. If a provider reportedly has much of its future capacity already sold, you can expect reservations and committed-spend discounts to remain the main route to price certainty. AWS says Savings Plans can reduce your bill by up to 72% compared to On-Demand prices in exchange for a usage commitment. We cover those levers in the pricing section. One data gap to flag: we could not confirm an AWS backlog figure from a source we trust, so we do not cite one.
Microsoft Azure: 48% Growth and the First Dollar Disclosure
Microsoft reported $92.3 billion in total revenue and $36.6 billion in profit for the quarter, ahead of the $90.6 billion revenue and $35.1 billion profit analysts expected, per AFP. Azure grew 48%, and the dollar figure of $33.2 billion is the number that lets outsiders finally size it against AWS. CNBC called it the first Microsoft earnings report to include Azure revenue in dollars.
Microsoft is pushing consumption-based Azure usage, with time-limited promotions scheduled to launch November 2, 2026 covering database modernization, Microsoft Fabric and Azure application-platform scenarios. On October 28, 2026 it also announced the Microsoft Copilot harness in Foundry Agent Service, in public preview. That pairing is Microsoft’s pitch against AWS and Google: the data platform and the agent platform sit side by side.
Capex tells the other half. Microsoft’s total capital expenditures reached $45.2 billion in the quarter, and Chief Financial Officer Amy Hood said around three-quarters went to short-lived assets such as memory chips and storage. She guided to more than $50 billion in the next quarter, bringing total spending to about $170 billion for the year, AFP reported. Microsoft also added 24 new data centers last quarter, according to Nadella on the call. Free cash flow still came in at $18.6 billion, which suggests Microsoft is funding more of its build from operations than its peers.
On products, the October news flow centers on Microsoft Foundry and the Foundry Agent Service. Microsoft announced the Copilot harness in Foundry Agent Service on October 28 in public preview, and its model router now covers 35 models. Microsoft also published a technical deep dive on Azure Maia 300, its chip for cloud-scale agentic inference, and said on October 29 it will expand Maia deployments to multi-gigawatt scale, according to Microsoft’s Azure blog. In hybrid, Azure Local now replaces Azure Stack HCI, and the matching partner specialization was renamed Azure Local for Hybrid & Sovereign Private Cloud.
Two warnings apply to Azure buyers. Microsoft reported scheduled price increases effective February 1, 2027 for certain virtual machines and storage, and it announced time-limited consumption promotions that start November 2, 2026. Both are covered in the pricing section. Microsoft also published Azure announcements on October 19 and 20 describing expanded controls for digital sovereignty and sovereign AI, which matters if that is on your shortlist. Our sovereign cloud comparison of AWS, Microsoft and Google goes deeper on that angle.
Google Cloud: 103% Growth, 36% Margin and a TPU Business
Google Cloud is the quarter’s headline. Revenue of $30.7 billion was up 103%, against analyst expectations of $29.2 billion, and operating income reached $11.12 billion at a margin of about 36%. Alphabet attributed the result to enterprise AI infrastructure, enterprise AI solutions, core Google Cloud Platform services and TPU system sales. That last phrase is new: Alphabet now sells complete TPU systems to outside customers, and the revenue lands inside the Cloud segment.
Google Cloud CEO Thomas Kurian said on the call that the unit now has 18 product lines with more than $1 billion each in annual revenue, that nearly 90% of the Fortune 100 use Gemini Enterprise, and that nearly 100 customers use TPUs. CEO Sundar Pichai said Gemini 4 Argon saw developer usage in its first week ramp 3.5 times faster than Gemini 3, even with capacity constraints. Pichai also said the Gemini app has 1 billion users. Those are management statements from the earnings call as relayed by Constellation Research.
The backlog is the other eye-catching number. Google Cloud reported about $561 billion, up from $514 billion three months earlier. That is roughly 18 times the quarter’s revenue. A multiple that high means contracted demand will take longer to convert, which is good for visibility and a risk if capacity does not arrive on time.
The market did not cheer. Alphabet shares fell about 13% after the report, the biggest one-day drop since the company went public in 2004, with roughly $482 billion in market value erased. Free cash flow turned negative $9.07 billion as quarterly capex hit $51.4 billion, and Alphabet’s CFO Anat Ashkenazi warned cloud margins could face near-term pressure because Google relies on outside computing capacity. Alphabet held 2026 capex guidance at $195 billion to $205 billion and said spending will rise significantly in 2027.
Product news kept pace. Google Cloud’s update feed shows Gemini 3.5 Flash generally available on Google Distributed Cloud air-gapped as of October 20, and Claude Opus 5.5 and Sonnet 5.5 available in Google Antigravity the same day, per the Google Cloud updates tracker. Google also introduced the Gemini agent on October 8, with project-level spend caps and smart routing. For a direct duel with the leader, see our AWS vs Google Cloud analysis.
Backlog, Capex and Cash: Who Is Funding the Capacity
Market share follows capacity, and capacity follows capital. The three providers are spending on a scale that has no real precedent in enterprise technology. The full-year figures in the table rest on different bases: Amazon’s cash capex, Alphabet’s calendar-year guidance and Microsoft’s figure as AFP reported it from the CFO, so treat them as order-of-magnitude markers rather than numbers to add up into a precise total.
| Item | Amazon (AWS) | Microsoft (Azure) | Alphabet (Google Cloud) |
|---|---|---|---|
| 2026 capex outlook | Not listed | About $170B (AFP summary of CFO guidance) | $195B to $205B |
| Latest quarterly capex | Not confirmed | $45.2B | $51.4B |
| Free cash flow | Not listed | $18.6B in the quarter | -$9.07B in the quarter |
| Backlog | Not listed | Not confirmed | $561B (Q3), up from $514B |
| Cloud operating margin | Not listed | Not disclosed | 36.2% |
| Funding signal | Not listed | Funds more of its build from operations | Has raised external capital through equity and debt |
The pattern is clear. Microsoft is the only one of the three reporting positive free cash flow this quarter, which gives it more room to wait for returns. Amazon and Alphabet are borrowing and issuing stock to keep building. That does not mean they are in trouble. It means the economic rent from AI capacity now depends on margins holding as depreciation catches up with spending. Exante’s market note made the same point: if cloud operating margins withstand rising depreciation, capacity owners keep a fair share of the value they fund. If margins narrow as new capacity reaches the income statement, the returns look more like a utility’s.
What should a cloud buyer take from this? First, sold-out capacity is a negotiating factor. Jassy reportedly said on October 23 that much of AWS’s 2027 capacity, and some of 2028, was already sold, though we could not tie that to a primary Amazon release, so treat it as unconfirmed. Second, short-term capacity for older training instances has sold at a premium compared with the first half of the year, which Jassy mentioned on the call. Third, the large backlogs at AWS and Google mean a few huge AI customers are locked in for years. That supports supply, but it can crowd out smaller customers when GPU or TPU inventory is tight.
For a practical read on how this flows to your bill, our GPU price comparison across AWS, Azure and Google Cloud shows the instance-level spread. Memory shortages also matter: Exante’s note says DRAM and NAND scarcity pushed Amazon to raise its capex outlook, and Microsoft’s CFO said three-quarters of its capex went to short-lived assets like memory and storage. Hardware inflation is already visible in provider budgets, and it may reach customer price lists in 2027, as the Azure announcement below suggests.
Developer Adoption: What the Stack Overflow Survey Adds
Revenue measures what companies pay. Developer surveys measure what engineers actually pick up and use. The Stack Overflow Developer Survey is a useful cross-check on that, but our verification run could not confirm its cloud usage figures, so we do not quote them here. You can read the data directly at Stack Overflow’s 2025 technology survey.
That gap matters for hiring and for tooling. A smaller developer base means fewer engineers who already know the platform, fewer third-party tutorials, and a thinner marketplace of integrations. Google is attacking that gap with developer products. It launched a Connected Apps Platform on October 28 and now offers Anthropic’s newest models inside Antigravity. Microsoft attacks from the enterprise side through Foundry and Fabric, while AWS leans on the familiarity of its older services, from S3 to Lambda.
Note the age of the survey. It reflects 2025 behavior, so it predates the agent and model-router launches that changed how teams buy cloud AI in 2026. We include it because it is the best large-sample developer data available and because its ranking has been stable year over year. If you need a fresher signal, look at job postings and certification demand in your own market. The Synergy numbers cited earlier give a spending view, the Stack Overflow numbers give an engineering view, and the earnings give a financial view. When all three agree on the order, as they do for first place, you can trust the order. When they disagree, as they do on Google’s rate of change, you should ask which definition matters for your project.
AI Model Access: Bedrock vs Foundry vs Google’s Model Garden
October 2026 turned into a month of model-catalog land grabs. Every cloud now sells OpenAI and Anthropic models alongside its own, so the choice is less about which models exist and more about how each platform routes, meters and governs them. If you are weighing agent platforms specifically, our AgentCore, Foundry and Vertex AI agent builder comparison goes through the controls one by one.
| Dimension | AWS | Microsoft Azure | Google Cloud |
|---|---|---|---|
| Main AI platform | Amazon Bedrock and Bedrock AgentCore | Microsoft Foundry and Foundry Agent Service | Gemini Enterprise Agent Platform |
| Newest agent launch | Strands Box and Nova 2.5 Sonic (Oct. 13 roundup) | Copilot harness in Foundry Agent Service, public preview (Oct. 28) | Gemini agent with routing and spend caps (Oct. 8) |
| Model routing | Cross-Region inference profiles (for example GLM 5.3, Oct. 5) | Model router, 35 models | Smart routing and project-level hard spend caps |
| Third-party models in October | GPT-6.1 Terra generally available (Oct. 26), Claude Haiku 5.5, GLM 5.3 | Foundry router pool (see text for the model list) | Claude, OpenAI, Gemma and Mistral integrations (Oct. 20 to 23) |
| Custom AI silicon | Trainium, Graviton 5 | Maia 300 (technical deep dive Oct. 12) | TPU systems sold to outside customers |
| Air-gapped or sovereign angle | Claude Haiku 5.5 on AWS GovCloud (US) | Azure Local replaces Azure Stack HCI | Gemini 3.5 Flash GA on Google Distributed Cloud air-gapped (Oct. 20) |
On AWS, the newest headline is OpenAI GPT-6.1 Terra reaching general availability on Amazon Bedrock on October 26. AWS says the model balances capability, speed and cost for everyday workloads, and that its input and output prices sit at least 50% below GPT-6.1 Sol and GPT-5.6 Terra, per the AWS announcement for GPT-6.1 Terra. AWS also added Claude Haiku 5.5, which it describes as roughly 75% cheaper than Claude Haiku 4.5 for most tasks and the first Haiku model with effort controls.
On Azure, the model router is the story. According to Microsoft’s documentation, the router pool added OpenAI GPT-6 Astra and xAI’s Grok 4.6 in an October update. An August update brought GPT-5.6 Sol, Terra and Luna along with Claude Opus 4.8. That is a single-source listing, so check the Foundry model documentation before you plan around a specific model. For price points, our Microsoft Foundry pricing breakdown lists the tiers.
On Google Cloud, the October move was breadth. Google announced OpenAI models in its Model Garden on October 21 (updated October 27) and, according to that announcement, named GPT-6.1 Sol with usage appearing on one Google Cloud invoice, drawing down eligible commitments. On October 23 it offered both Anthropic-managed and Google-managed deployment of Claude models, with the same pricing and billing for either option. Google also pointed to more than 200 first-party, third-party and open models on its platform. The first-party flagship is Gemini 4 Argon, announced October 13, which Constellation Research reports is seeing strong early demand.
The takeaway: model availability is converging, so differentiation moves to routing, spend control, data location and the chips underneath. AWS leans on breadth of tiers, Azure on Foundry plus Fabric and Copilot, and Google on its own models and TPU economics.
Pricing Signals: Savings Plans, Token Tiers and the 2027 Azure Hike
Cloud list prices are hard to compare across providers because instance families, regions and discount programs differ. The sources we verified do not provide matched late-October prices for standard CPU compute, object storage and egress on all three clouds, so we do not invent a three-way price grid. What we can compare are the pricing mechanisms and the changes announced in the last few weeks.
| Pricing item | AWS | Microsoft Azure | Google Cloud |
|---|---|---|---|
| Commitment discounts | Savings Plans up to 72% vs On-Demand | Savings plan up to 65% on select compute | Flexible Savings Plans (rate not confirmed in our sources) |
| Spare capacity | Spot up to 90% vs On-Demand | Not confirmed in our sources | Not confirmed in our sources |
| Frontier model tokens | GPT-6.1 Terra priced at least 50% below GPT-6.1 Sol and GPT-5.6 Terra | See Foundry pricing by model | Introductory $2 input and $10 output per 1M tokens through Nov. 30, then $4 and $20 |
| Inference tiers | Flex and Batch 50% below Standard, Priority 75% above | Not confirmed in our sources | Standard, flexible, dedicated capacity and model-specific rates |
| Agent runtime | AgentCore v1: $0.0895 per vCPU-hour, $0.00945 per GB-hour. v2: $0.1276 and $0.0169 | Defender Workload Units billing in preview, no price confirmed | Not confirmed in our sources |
| Scheduled changes | None confirmed | Increases effective Feb. 1, 2027, incl. 25% for v1 and v2 VMs | Introductory token price ends Nov. 30, 2026 |
| Promotions | Up to $200 Free Tier credits for new customers | Consumption promotions start Nov. 2, 2026 | Not confirmed in our sources |
Azure’s announced increase is the most important item for budget owners. Long View Systems’ October licensing roundup reports that price increases for certain Azure virtual machines and storage services take effect February 1, 2027, including increases in specified European regions and a 25% global increase for v1 and v2 virtual machines. These are future prices, not current ones. If you run older VM generations on Azure, the cheapest response is usually to move to a current generation before the date, since the increase targets v1 and v2 sizes specifically.
Microsoft is also dangling carrots. Its partner announcements say consumption promotions begin November 2, 2026 for database modernization, Microsoft Fabric and Azure app-platform scenarios. The reported discounts reach up to 30% on eligible Azure database pay-as-you-go usage, up to 20% on selected Fabric capacity and up to 15% on app PaaS and data services. Those figures come from a single partner-facing source, and eligibility rules apply, so ask your account rep for the terms.
Google’s token pricing sets a different kind of deadline. According to Google’s announcement, Gemini 4 Argon costs $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4 and $20 from December 1. If you are prototyping an agent, you have about a month to burn through experiments at half the standard rate. Google also emphasized project-level spend caps on Gemini workloads, which answers a common finance-team worry about runaway inference bills.
On AWS, the main pricing story is tiering. Bedrock now sells Standard, Flex, Priority and Reserved inference, and the official page says exact prices vary by model, provider, modality and region. Pair that with Savings Plans on EC2, and the lever set is broad but demands active management. A FinOps tool helps here, and our cost tooling comparison of Cost Explorer, Azure Cost Management and Google billing shows what each native console tracks.
Real-World Examples: Who Is Betting on Which Cloud
Customer stories show where each provider is winning. All of the examples below come from the earnings coverage, company pages and announcements we reviewed, and we attribute each to its source.
Apple on Google TPUs. On the earnings call, Google Cloud’s Kurian named Apple as a public reference customer scaling pretraining of its foundation models on the latest generation of TPUs. That is a major validation of Google’s decision to sell TPU capacity outside its own services.
Citadel Securities and Jump Trading on Google. Kurian also listed capital markets firms Citadel Securities and Jump Trading, plus Caltech and frontier model companies Mirendil and Physical Intelligence, among the new TPU customers. Finance firms move slowly on infrastructure, so their presence signals confidence in latency and reliability for compute-heavy research.
Meta and OpenAI on AWS. CNBC reported that AWS signed deals with Meta and OpenAI during the quarter. The details were not published in the coverage we reviewed, but the pairing shows AWS winning top-tier AI labs for infrastructure, not only enterprise IT.
Western Sydney University on Azure. Microsoft’s own page says the university took control of its modernization with Azure and saved approximately 30% in costs. This is a vendor-published case study, so treat the savings as the customer’s reported outcome rather than an independent benchmark.
August AI on Azure. Microsoft highlights a healthcare companion built on Azure that supports 3.5 million customers. Again vendor-published, but it shows Azure’s pull in regulated sectors where Microsoft’s compliance catalog matters.
AWS Lambda at $1 billion. Constellation Research reported that Lambda now carries a $1 billion annual run rate. That is a customer-behavior example at scale: serverless is no longer a niche.
Fabric and Foundry at Microsoft customers. Microsoft Fabric and Microsoft Foundry are both current Azure products. For a data-centric company already on Microsoft 365, using them together is the shortest path to a governed agent.
Taken together, the examples show a split. Google is winning AI-native compute buyers and quant-heavy finance. AWS holds the broadest base and the biggest AI labs. Microsoft wins where the buyer already lives in its software stack. None of these patterns means a single winner for your workload. They do suggest which provider will have the most references and partner depth in your industry.
Use-Case Recommendations: Which Cloud for Which Job
Share numbers tell you who is big. They do not tell you which cloud fits a given workload. These recommendations rest on the product signals above and on the pricing mechanics we could verify. Test each against your own architecture before you commit.
1. Frontier-model pretraining or large inference fleets: Google Cloud. TPU supply, a reported 80% price-performance gain for TPU 8i versus the prior generation (a single-source Google claim), and named customers like Apple make this the strongest case. Watch the capacity warnings from Alphabet’s CFO.
2. General enterprise workloads with a deep service catalog: AWS. Core compute and storage grew at their fastest rates in years, Savings Plans reach 72%, and the developer base is the largest. This is the default for teams that want the most mature options in every category.
3. Microsoft-centric enterprises: Azure. If you run Microsoft 365, Dynamics, SQL Server and Fabric, Azure’s Foundry, Fabric and Copilot integration shortens the path to production. Plan for the February 1, 2027 price changes on older VM generations.
4. Multi-model AI applications with spend controls: compare all three. AWS offers Flex and Priority tiers, Azure offers a 35-model router, and Google offers project-level hard spend caps. Pick on governance fit, because the model lists overlap. For databases behind those apps, our managed database cost comparison shows where the price gap sits.
5. Hybrid and air-gapped deployments: Azure Local or Google Distributed Cloud. Azure Local has replaced Azure Stack HCI, and Google put Gemini 3.5 Flash on its air-gapped distributed cloud on October 20. AWS GovCloud (US) now carries Claude Haiku 5.5 for regulated U.S. workloads.
6. Serverless-first teams: AWS. Lambda at a $1 billion run rate and the new Lambda Web Functions announced in the October 19 roundup make it the most active serverless platform in our sources.
7. Cost-sensitive startups: start with credits, then price by workload. AWS lists up to $200 in Free Tier credits for new customers. Google and Microsoft run their own promotions, including Azure’s November 2 consumption offers. Credits favor whoever you start with, so model costs at month 13, not month 1.
Where two clouds fit, run a two-week pilot on each with realistic traffic. Benchmarks published by vendors rarely match how your queries behave, and the real costs often come from egress, support plans and idle capacity.
Migration Guide: Moving Workloads Between Clouds
Cloud share shifts slowly because migration is hard. Jassy argued on the call that about 85% of global IT spend is still on premises, which implies more growth ahead for all three providers, but moving between clouds remains a project measured in months. This guide covers a practical sequence that works for most application workloads. For the cloud category overview, see our cloud computing coverage.
Step 1: Inventory and Classify
List every workload, its data volume, its dependencies and its uptime requirement. Group them into three buckets: lift-and-shift VMs, containerized services and managed-service-dependent apps. Managed databases, queues and serverless functions carry the highest switching cost because each cloud’s API is different. Record egress volume for each workload, since data leaving a provider is where unexpected bills appear.
Step 2: Pick the Target and Prove It
Run a proof of concept on the target cloud with one non-critical service. Measure latency, throughput and cost under production-like load for at least two weeks. Check regional availability for the specific services you need. Check committed-spend rules too, because Savings Plans and equivalents are provider-specific and do not transfer.
Step 3: Move Data First, Then Compute
Object storage usually moves first because it has the simplest interface. A tool such as rclone can sync between S3, Azure Blob Storage and Google Cloud Storage. The example below copies a bucket from AWS to Google Cloud with checksums and parallel transfers. Replace the remote names with the ones you configured.
rclone sync s3-source:my-bucket gcs-target:my-bucket \
--checksum --transfers 32 --checkers 16 \
--progress --log-file migration.logRun the sync several times before cutover so the final delta is small. Verify with a count and a checksum comparison, not with file size alone. For databases, use the provider’s migration service or native replication, then switch reads before writes.
Step 4: Cut Over, Monitor and Decommission
Lower DNS TTLs a few days ahead, freeze writes briefly, run the final sync and point traffic at the new environment. Keep the old environment read-only for at least one billing cycle as a rollback path. Once costs and error rates look stable, delete the old resources, cancel the old commitments if the contract allows, and document the new topology. Moving between clouds is expensive enough that most teams should only do it for a clear reason: capacity access, a specific AI model, a price change like Azure’s 2027 increase, or a sovereignty requirement.
Pros and Cons of Each Cloud
Every provider has clear strengths and real drawbacks. These summaries draw on the Q3 filings, the product announcements and the pricing mechanics above.
| Provider | Pros | Cons |
|---|---|---|
| AWS | Broadest service catalog, deep commitment discounts (Savings Plans up to 72%, Spot up to 90%) | Tiered inference pricing demands active management |
| Microsoft Azure | 48% growth on a $33.2B base, positive free cash flow ($18.6B), Foundry plus Fabric plus Copilot stack, strong hybrid story with Azure Local | Scheduled price increases from Feb. 1, 2027, profitability not broken out, more complex licensing, capex guidance above $50B next quarter |
| Google Cloud | 103% growth, 36.2% margin, $561B backlog, TPU sales to outside customers, Gemini 4 Argon, hard spend caps | Smallest of the three, negative free cash flow |
A few caveats sit behind that table. AWS’s Q3 backlog was not confirmed in our sources, so its apparent backlog disadvantage compares Q2 with Q3. Microsoft’s operating margin for Azure is not disclosed, so we cannot rank it against AWS and Google on profitability. Google’s segment includes Workspace, so its revenue growth overstates infrastructure-only momentum. These caveats do not change the direction of the story, but they should stop anyone from using a single number as a scorecard.
Also consider the nature of each risk. AWS’s risk is financial and competitive: huge spending, then a faster-growing rival. Azure’s risk is customer-facing: price rises and licensing complexity. Google’s risk is operational: it must deliver capacity on time, with the backlog and the stock market both watching. If your workload is mission-critical and you cannot tolerate capacity hiccups, the largest base is reassuring. If you want to ride the fastest-moving AI platform and can handle some delivery risk, Google’s numbers are hard to ignore.
Verdict: Reading the Cloud Market Share Data
AWS keeps first place in cloud market share by a clear margin, and nothing in Q3 2026 changes that. Its quarterly revenue stays well ahead of Azure’s $33.2 billion and Google Cloud’s $30.7 billion. Anyone who writes AWS off because of Google’s 103% is misreading the base effects.
The more interesting read concerns second place. Azure at $33.2 billion leads Google Cloud by $2.5 billion, and Google is growing more than twice as fast. By simple extrapolation, which is a rough tool and not a forecast, the order of second and third could flip within a few quarters. That would hinge on Google converting its $561 billion backlog and keeping TPU sales strong. Microsoft has the cash flow advantage and the installed software base to defend. We would not bet against either.
Our recommendation by buyer type is straightforward. Teams that want the safest and broadest default should choose AWS. Microsoft shops should lean on Azure, with a plan for the February 2027 price changes. AI-heavy teams with flexible architecture should price Google Cloud seriously, especially while the Gemini introductory token rate runs through November 30. Multi-cloud is a rational answer when you need access to specific models or chips, but it raises operating cost, so keep it deliberate.
Two data gaps limit this verdict. We have no verified Synergy or Canalys share figures for Q3 2026, and we have no verified Q3 backlog for AWS or Microsoft. Both will probably appear within weeks. When they do, update the share table first, because the next Synergy release will show whether Google’s jump translates into a measurable share gain on the market-wide definition. Until then, the best evidence says AWS leads, Azure follows closely, and Google Cloud is closing the distance faster than anyone.
FAQ: Cloud Market Share and the Big Three
Which cloud provider has the largest market share in 2026? AWS. AFP, citing Synergy’s Q2 data, reported that AWS remained number one by revenue, with Microsoft second and Google third.
How big is Azure compared with AWS now? Azure revenue was $33.2 billion in the quarter, and AWS remained the larger of the two.
Is Google Cloud really growing 103%? Alphabet reported Google Cloud revenue of $30.7 billion, up 103%. Part of the jump comes from TPU system sales and AI infrastructure demand. The segment also includes Workspace subscriptions, so it is not a pure infrastructure comparison.
Will Azure prices rise? Long View Systems reports price increases for certain Azure virtual machines and storage services effective February 1, 2027, including a 25% global increase for v1 and v2 virtual machines. These are future changes, and details vary by region and product.
Which cloud is best for AI workloads? It depends on the job. Google has TPUs and Gemini 4 Argon, AWS has the broadest model tiers on Bedrock including GPT-6.1 Terra, and Azure has Foundry’s 35-model router and Copilot integration. Compare governance and spend controls before price.
What is the biggest risk when choosing a cloud now? Capacity and cost. All three providers are spending hundreds of billions of dollars, and Alphabet’s CFO warned that margins could face near-term pressure. Negotiate committed-spend terms with clear exit clauses.
When will new market share data arrive? Synergy Research typically publishes its quarterly cloud infrastructure share a few weeks after the earnings calls. Check for the Q3 2026 release in November or December before you finalize multi-year contracts.
