Amazon Google Cloud Race Intensifies as AI Spending Jumps Again

Amazon raised its 2026 capital spending plan to $220 billion, despite already committing more than any major rival to new computing capacity. The increase sharpens the Amazon Google contest for AI infrastructure customers, while putting Amazon’s future cash returns under greater scrutiny.

The timing matters because Amazon Web Services is finally accelerating alongside the spending. AWS revenue rose 37% year over year during the second quarter, its fastest growth in 18 quarters. Amazon says demand still exceeds the capacity it can install.

Google and Microsoft are reporting the same basic constraint, but their financial signals differ. Google Cloud grew faster, while Microsoft avoided another large spending increase. Amazon now must show that its unmatched investment scale produces durable cloud growth, not merely larger bills.

Amazon Added Another $20 Billion to Its Buildout

Amazon’s new spending plan turns a strong cloud quarter into a much larger bet on future demand.

Amazon now expects approximately $220 billion in cash capital expenditures during 2026. That is $20 billion above the estimate management provided in February and far above the $128 billion spent during 2025.

Capital expenditures, usually shortened to capex, cover long-lived assets such as servers, chips, networking equipment, and data centers. Amazon’s total also includes investments in robotics and its satellite business, so it is not a pure AI figure.

AWS remains the main destination, according to CEO Andy Jassy. Amazon needs more computing capacity for conventional cloud workloads, model training, AI inference, and its own growing software services.

The company’s quarterly results supplied an unusually strong argument for the increase. AWS revenue reached roughly $42.2 billion during the April through June quarter, rising 36.7% from one year earlier.

That growth accelerated sharply from 28% in the previous quarter. It also exceeded the recent growth rates that had fueled concerns about AWS losing momentum to Microsoft Azure and Google Cloud.

Amazon said its AI and custom-chip businesses each surpassed annualized revenue run rates of $25 billion. A run rate annualizes current performance, rather than reporting completed full-year sales.

Jassy also cited more expensive memory as the main reason for raising the spending forecast. AI servers require large amounts of high-bandwidth and conventional memory, making component inflation meaningful at Amazon’s scale.

The increase does not mean Amazon can immediately satisfy every customer. Data center projects require land, electricity, cooling systems, network connections, permits, and specialized equipment before servers begin generating revenue.

Jassy said Amazon still expects demand to exceed available capacity throughout 2026. He expects the imbalance to continue during 2027, while describing already visible demand for 2028 as striking.

That statement captures the central tension. Amazon is not building solely because executives expect an eventual AI boom. It says existing customers are already asking for more capacity than AWS can supply.

However, customer interest and profitable utilization are not identical. Capacity ordered today must remain useful after newer chips, models, and software architectures change the economics of AI computing.

The strongest part of Amazon’s case is therefore not the headline spending figure. It is the combination of accelerating AWS revenue, existing commitments, and constrained supply.

The weakest part is the long gap between writing checks and collecting returns. Buildings can support several server generations, but some expensive computing equipment depreciates much faster.

Amazon’s $220 billion plan raises the stakes on both sides. Faster AWS growth justifies building more, yet the enlarged buildout creates an even higher standard for future growth.

The Amazon Google Spending Race Has a Revenue Signal

The Amazon Google rivalry is no longer defined only by promised infrastructure because both companies are now reporting sharp cloud acceleration.

Alphabet reported that Google Cloud revenue grew 82% year over year in its second quarter. Revenue reached $24.8 billion, while its cloud backlog rose to $514 billion.

Backlog represents contracted revenue that has not yet been recognized. It offers visibility into demand, although the timing and profitability of that future revenue can vary.

Google also increased its 2026 capex forecast to between $195 billion and $205 billion. Its previous range was between $180 billion and $190 billion.

The company said the overwhelming majority of that spending will support technical infrastructure. That category includes servers, networking equipment, and data center construction.

Google CEO Sundar Pichai said the company remained supply constrained even after its capacity additions. Google also reported that approximately nine million developers use its models each month.

Its first-party model APIs were processing about 22 billion tokens per minute, up from 16 billion one quarter earlier. Tokens are the small text or data units that AI models process.

These operating figures help connect infrastructure to activity. They do not reveal the margin earned on every AI request, but they show that expensive equipment is supporting measurable usage.

Google’s cloud momentum creates a direct challenge for Amazon. AWS remains larger, but Google Cloud is adding revenue at a much faster percentage rate.

Google can also spread infrastructure across several businesses. The same technical foundation supports Cloud customers, Gemini, Search features, YouTube tools, security services, and internal model research.

That integration improves potential utilization. It also makes the economics harder to isolate because Alphabet does not disclose a separate income statement for every AI product.

Amazon has its own shared-use advantage. AWS infrastructure supports external customers, Amazon services, custom silicon, and partnerships with model developers.

The distinction lies in where each company begins. Amazon leads the cloud infrastructure market, while Google combines a smaller cloud position with a much larger consumer distribution network.

First-quarter market estimates placed AWS at 28% of worldwide cloud infrastructure spending. Microsoft held 21%, while Google held 14%, according to industry estimates reported in market share data.

That lead gives AWS an enormous installed base. Enterprises already run databases, applications, storage systems, and security tools inside Amazon’s environment.

AI workloads can expand those existing accounts. A company training a model may also purchase storage, networking, analytics, databases, and identity services from the same provider.

Google approaches the opportunity from another direction. Its custom Tensor Processing Units, Gemini models, data products, and security portfolio form an integrated alternative to AWS.

The Amazon Google contest is therefore not a simple race to own the most graphics processors. Each company wants AI demand to pull customers deeper into its broader cloud platform.

The latest numbers suggest both strategies are attracting demand. They do not yet establish which company will convert that demand into superior long-term returns.

Microsoft Makes the Spending Comparison Harder

Microsoft shows that cloud acceleration can impress investors without announcing another dramatic increase in planned investment.

Microsoft reported that Azure and other cloud services revenue rose 43% during its fiscal fourth quarter. That performance exceeded the growth range management had forecast one quarter earlier.

The company also said annual Azure revenue surpassed $100 billion for the first time. Microsoft Cloud revenue exceeded $214 billion for its full fiscal year.

Its Azure results arrived one day before Amazon reported its quarter. They gave investors another benchmark for evaluating the relationship between spending and growth.

Microsoft did not pair the results with a comparable increase in its calendar-year spending plan. That contrast helped reinforce a market preference for visible revenue acceleration without another large capex surprise.

The comparison requires care because the companies use different reporting periods and accounting presentations. Microsoft also uses finance leases extensively for data center assets.

Still, Microsoft confronts the same physical limits. Management has repeatedly said demand exceeds available cloud and AI capacity, even as new infrastructure comes online.

Microsoft expected more than $40 billion in quarterly capex during its fiscal fourth quarter. Its calendar-year 2026 expectation stood near $190 billion earlier in the year.

The company said higher component costs accounted for a meaningful portion of that plan. It also expected supply constraints to persist through 2026.

Microsoft’s advantage is its enterprise software distribution. Azure infrastructure connects directly with Microsoft 365, GitHub, security products, databases, and industry applications.

Microsoft 365 Copilot reached more than 30 million paid seats, according to the company. That gives Microsoft a route for monetizing AI above the infrastructure layer.

Amazon has a different software profile. AWS provides a wide set of developer and enterprise services, but Amazon lacks Microsoft’s dominant workplace software bundle.

Google occupies a middle position. It sells enterprise infrastructure and productivity software, while also distributing AI through consumer products with enormous audiences.

These differences matter because AI infrastructure margins face pressure from depreciation, energy use, and frequent hardware replacement. Providers want software revenue that can absorb those costs.

A cloud customer might rent raw accelerator capacity for model training. That can generate large revenue, but it may offer less differentiation than a complete application or managed platform.

Each provider is therefore trying to climb the same stack. Amazon promotes Bedrock, SageMaker, custom Trainium chips, and managed agents alongside basic infrastructure.

Google combines its cloud platform with Gemini models, Vertex AI, custom TPUs, Workspace, and security products. Microsoft integrates Azure with OpenAI models, Copilot, GitHub, and its business applications.

This is why quarterly cloud growth alone cannot settle the contest. Revenue composition, contract duration, depreciation, and software attachment all influence the eventual return.

Microsoft’s quarter increases pressure on Amazon because Azure accelerated without another major revision to spending guidance. It also pressures Google, whose faster growth accompanied a larger capex forecast.

Amazon must now prove that its additional $20 billion buys revenue that would otherwise remain unavailable. Merely matching competitors’ infrastructure expansion will not be enough.

More Capacity Does Not Guarantee Better Returns

The central risk is not weak AI interest, but whether today’s infrastructure can earn enough before its most expensive components become outdated.

Amazon’s spending increase landed alongside a deterioration in free cash flow. The company reported a $7.6 billion outflow for the 12 months ending June 30, compared with an $18.2 billion inflow one year earlier.

Free cash flow generally measures operating cash after purchases of property and equipment. It can fall quickly when a company accelerates construction and server purchases.

Amazon’s overall business remains capable of generating substantial operating income. Yet the cash-flow reversal shows that its AI and logistics buildout has an immediate financial cost.

Alphabet faced similar scrutiny after reporting strong revenue. Heavy spending pushed its quarterly free cash flow below zero, despite Google Cloud’s rapid expansion and solid performance elsewhere.

An independent earnings analysis noted the contrast between accelerating AI demand and pressure on Alphabet’s cash generation. Investors reacted cautiously even though revenue exceeded expectations.

These results undermine the simplest bullish argument. Strong demand does not automatically prevent near-term financial strain.

They also weaken the simplest bearish argument. The providers are not spending while cloud revenue stagnates. AWS, Azure, and Google Cloud all reported rapid growth.

The harder question concerns the useful life of the assets. Data center buildings can operate for decades, while servers and accelerators require replacement much sooner.

Microsoft previously said roughly two-thirds of a quarterly capex total consisted of shorter-lived assets, primarily CPUs and GPUs. Those components need faster monetization than land or buildings.

Amazon describes its spending mix differently, but it faces the same technical reality. AI customers often seek the latest accelerator generation because performance improvements can lower per-token costs.

Custom chips offer one possible defense. Amazon’s Trainium accelerators aim to reduce dependence on Nvidia while giving AWS more control over hardware economics.

Google has pursued that strategy for years with its TPUs. Microsoft also develops first-party accelerators and server processors while continuing to purchase equipment from Nvidia and AMD.

Custom silicon can improve costs when utilization is high. It can also create development risk, customer compatibility concerns, and another inventory cycle that management must forecast correctly.

Energy presents another constraint. A completed server order cannot generate revenue without sufficient electricity, cooling, networking, and regional approvals.

Providers have announced data center projects across multiple continents, but some locations face grid delays and community opposition. Construction speed alone does not guarantee usable computing supply.

Customer concentration adds uncertainty. Large model developers can sign enormous commitments, yet their spending depends on fundraising, product adoption, and future model economics.

A backlog can provide confidence, but not every contract carries the same margin or schedule. Some arrangements also involve strategic investments flowing between cloud providers and AI companies.

That circularity deserves attention. A cloud provider can invest in a model developer, which then commits to purchasing computing capacity from the same provider.

The arrangement can accelerate platform development and secure major customers. It can also make reported demand harder to compare with ordinary enterprise consumption.

Amazon has expanded relationships with Anthropic and OpenAI, while Microsoft remains closely connected to OpenAI. Google develops its leading models internally and supplies outside developers through Cloud.

None of these approaches is inherently unsound. Each creates different dependencies that quarterly cloud growth rates do not fully disclose.

The prudent reading is therefore conditional. Amazon’s spending looks justified while contracted demand, capacity utilization, and AWS revenue continue rising together.

The case weakens if capacity becomes available faster than customers can deploy it. It also weakens if pricing falls faster than hardware efficiency improves.

Amazon says it plans capacity against visible demand rather than broad speculation. Investors still need several quarters of evidence before treating the entire $220 billion plan as validated.

Enterprise Buyers Gain Leverage and New Risks

The infrastructure race gives enterprises more computing choices, but it also raises the cost of selecting the wrong technical and commercial path.

When providers lack capacity, customers face longer deployment schedules and limited access to preferred chips. More construction should gradually reduce those bottlenecks.

Competition can also lower unit costs. Amazon, Google, and Microsoft are optimizing model serving, developing custom accelerators, and improving data center efficiency.

Yet the largest savings often require deeper platform commitment. A customer may obtain better economics by using a provider’s native chips, managed databases, model platform, and security services together.

That choice can increase switching costs. Applications built around one provider’s proprietary accelerator or managed agent system may require significant work to move elsewhere.

The risk is not limited to infrastructure. Data pipelines, identity controls, observability systems, and model evaluation workflows can all become tied to one cloud.

Enterprises should therefore separate temporary capacity decisions from long-term architecture decisions. Renting available GPUs is different from making one provider central to an entire AI operating model.

Multicloud strategies can preserve bargaining power, but they add operational complexity. Teams must manage networking, security policies, data movement, and different platform interfaces.

Data-transfer costs can further limit portability. Moving large training datasets or model outputs between clouds may create delays and additional charges, even without formal lock-in.

AI agents make the choice more consequential. An agent can call models, retrieve company information, use software tools, and execute tasks across several systems.

If that workflow depends heavily on one provider’s identity, data, and orchestration services, moving it later becomes more difficult. The infrastructure decision becomes an application architecture decision.

The Amazon Google race offers buyers a useful comparison. AWS begins with broad infrastructure adoption and a large partner network. Google emphasizes its integrated models, data systems, TPUs, and security capabilities.

Microsoft adds a third route built around Azure and widely deployed workplace software. Smaller providers can compete through specialized accelerator access, lower complexity, or particular regional capacity.

Buyers should evaluate total workload economics rather than headline compute availability. That includes utilization, storage, networking, software services, engineering effort, and migration costs.

They should also test workloads on more than one architecture before committing. A model that performs efficiently on one accelerator may behave differently on another.

Procurement teams need clearer contract protections as commitments grow. Capacity reservations, performance expectations, data portability, and termination conditions deserve the same attention as unit pricing.

Knowledge workers face a related information problem. Product names, model versions, capacity claims, and enterprise commitments now change across every earnings cycle.

Maintaining a structured record of decisions can reduce confusion. A searchable AI knowledge base can connect vendor claims with tests, contracts, and internal deployment evidence.

That discipline matters because provider announcements describe strategic direction, not guaranteed customer outcomes. Every enterprise still needs to validate performance within its own data, security, and workflow constraints.

The spending race should improve access over time. It will not remove the need for architectural caution.

Three Signals Will Test the Amazon Google Thesis

The next phase will be judged by revenue conversion, capacity utilization, and the cost of turning AI demand into durable cash flow.

The first signal is AWS growth during Amazon’s next two earnings reports. The latest quarter established that the cloud business can accelerate while operating at enormous scale.

Another quarter above its recent trend would strengthen Amazon’s claim that new capacity is monetized quickly. A sharp slowdown would make the added $20 billion harder to defend.

Operating income matters alongside revenue. AWS can grow rapidly while delivering weaker incremental returns if depreciation, energy, and component expenses rise too quickly.

Investors should compare revenue growth with the AWS operating margin, not treat either figure alone as decisive. Stable margins during expansion would provide stronger evidence of disciplined investment.

The second signal is whether Google Cloud maintains unusually high growth while Alphabet brings new capacity online. Google’s 82% increase creates a demanding comparison for future quarters.

Continued backlog expansion would support the view that Google has enough contracted demand to absorb its spending. Slower backlog growth would suggest the current surge was partly concentrated in exceptional deals.

Google’s cloud operating margin also deserves attention. Improving profitability while revenue expands would show that infrastructure and software utilization are offsetting higher depreciation.

Falling margins would not automatically invalidate the strategy. However, they would clarify the near-term cost of pursuing AWS and Microsoft at this pace.

The third signal is whether reported supply constraints ease during late 2026. Amazon, Google, and Microsoft all say customer demand currently exceeds capacity.

If newly installed equipment is immediately allocated, their spending forecasts gain credibility. High utilization would show that construction is addressing an existing shortage.

If providers begin offering aggressive discounts or reporting excess capacity, the interpretation changes. That would suggest supply is catching demand faster than expected.

Hardware availability will provide an early clue. Shorter waiting periods for accelerators could indicate successful capacity additions, weaker demand, or both.

Contract behavior will offer another clue. Rising multiyear commitments would reinforce confidence, while shorter commitments could signal customer caution about model and hardware changes.

The Amazon Google contest will also be shaped by custom silicon adoption. More external use of Trainium or Google TPUs would improve each provider’s control over costs.

Customers will decide whether those chips provide enough performance and software compatibility. Provider benchmarks alone cannot answer that question.

Microsoft’s response remains an important control case. Azure growth and spending discipline will influence how investors judge similar decisions at Amazon and Alphabet.

If Microsoft sustains cloud acceleration without raising its plan again, Amazon’s higher spending faces more scrutiny. If Microsoft also raises spending, the industrywide capacity shortage looks more convincing.

The central question is no longer whether companies want AI computing. Their reported growth, usage, and backlogs already demonstrate substantial demand.

The question is whether infrastructure built during 2026 will stay heavily used across several hardware generations. That outcome determines whether today’s capex becomes a durable advantage or a prolonged drag.

Enterprise buyers should track these signals alongside their own workload results. Which provider is delivering predictable capacity, transparent economics, and portable architecture for your most important AI systems?

The next earnings cycle will offer the first answer. The longer answer will emerge as the Amazon Google buildout moves from construction spending to sustained customer use.