Part I: The Confession

December 3, 2024. Las Vegas Convention Center. Matt Garman strode onto the stage at AWS re:Invent wearing his signature Amazon blue shirt, facing more than 50,000 developers, engineers, and IT executives gathered for the cloud industry's biggest annual conference. This was his first re:Invent as AWS CEO, and everyone knew what question hung in the air: Could Amazon Web Services—the $100 billion profit machine that funds Amazon's retail ambitions—compete in the AI era?

Six months into his tenure as AWS's third CEO, Garman did something unexpected. He admitted AWS had fallen behind.

"We didn't have some of the whiz-bangy things that you could get out there quickly," Garman told Fortune in a July 2025 interview, referring to the generative AI products that Microsoft, Google, and OpenAI had been shipping since late 2022. While Microsoft was integrating OpenAI's ChatGPT into Office 365 and GitHub, and Google was rushing out Bard, AWS had been... building infrastructure.

The confession was jarring. AWS—Amazon's crown jewel, accounting for more than 60% of Amazon's operating profit—was openly acknowledging it had lagged competitors in the defining technology shift of the decade. For a company that had dominated cloud computing for 18 years, controlling 30% of the $99 billion global market, this was an uncomfortable position.

But Garman wasn't finished. He argued the lag was "by design." AWS, he explained, had deliberately prioritized building the foundational infrastructure—the compute, storage, and AI chips—that would power generative AI at scale, rather than rushing out flashy chatbots. "Our customers want to build their own AI applications," he said. "They don't want generic chatbots. They want infrastructure that scales."

The question was whether this would be remembered as strategic patience or a costly miscalculation. Because while AWS had been building infrastructure, Microsoft's cloud business Azure had been growing 39% year-over-year in Q2 2025, and Google Cloud Platform had posted 32% growth. AWS? Just 17.5%.

Matt Garman—the former intern who became the first product manager of EC2, Amazon's revolutionary Elastic Compute Cloud—now had to prove that AWS's deliberate approach could win the AI infrastructure race. The stakes: not just AWS's market leadership, but Amazon's entire profitability model.

Part II: The Long Road from Intern to CEO

Matt Garman's journey to AWS CEO began improbably in 2005, when he was pursuing an MBA at Northwestern University's Kellogg School of Management. That summer, he landed an internship at Amazon—a company then known primarily for selling books online. AWS, Amazon's nascent cloud computing division, was just being conceived in Jeff Bezos's mind as a way to monetize Amazon's internal infrastructure.

Garman joined Amazon full-time in 2006, hired as one of AWS's first product managers. He holds a BS and MS in Industrial Engineering from Stanford University, credentials that would prove valuable as he helped design the systems that would revolutionize enterprise computing. Prior to Amazon, Garman worked at SideStep as Director of Product Management and at Riffage as a Product Manager, but it was at AWS where he would make his mark.

In 2006, AWS officially launched with two foundational services: Simple Storage Service (S3) for cloud storage and Elastic Compute Cloud (EC2) for on-demand computing. Garman became EC2's first product manager, leading product management in the critical early years when AWS was convincing skeptical enterprises to abandon their own data centers and trust Amazon's cloud.

The decision to join AWS in 2006 required vision. Cloud computing was unproven. Most CIOs thought the idea of renting computing power from Amazon—a retailer—was absurd. But Garman saw the future: a world where companies would no longer need to build and maintain expensive data centers, instead paying only for the compute and storage they actually used.

Over the next decade, Garman rose through AWS's ranks. In 2016, he became general manager of all AWS Compute services, overseeing not just EC2 but the entire portfolio of compute offerings that had proliferated as AWS expanded. This role gave him deep technical expertise in the infrastructure layer that would later become critical for AI workloads.

In 2020, Garman took on an even broader mandate: leading AWS's worldwide Sales, Marketing, Support, and Professional Services division. This shift from pure product work to commercial leadership positioned him as a well-rounded executive who understood both the technology and the business of cloud computing. AWS at this time was approaching $50 billion in annual revenue, and Garman's commercial organization was responsible for landing and expanding the Fortune 500 customers who were migrating entire IT infrastructures to AWS.

Then came 2021, and a moment that must have stung. Andy Jassy, AWS's founding CEO, was promoted to succeed Jeff Bezos as Amazon's overall CEO. Everyone expected Garman—the EC2 product manager who had risen to lead sales—to take over AWS. Instead, Amazon shocked observers by bringing back Adam Selipsky, who had left AWS in 2016 to run Tableau Software (later acquired by Salesforce).

Why did Amazon pass over Garman? Sources close to the decision told Fortune that Amazon wanted someone with fresh external perspective and proven CEO experience. Selipsky had run Tableau for five years and successfully navigated a $15.7 billion acquisition by Salesforce. Garman, while deeply knowledgeable about AWS, had never been a CEO. Amazon's leadership principles prize "learn and be curious" and "hire and develop the best"—and bringing back Selipsky sent a signal that even internal stars needed seasoning.

Garman remained loyal. He continued leading AWS's sales organization under Selipsky, never publicly complaining about being passed over. Those who worked with him describe an "unflappable" executive who kept his emotions in check. But the 2021 decision would prove consequential: it meant Garman would take over AWS not in a time of smooth sailing, but at a critical inflection point.

In May 2024, Adam Selipsky announced he was stepping down to spend more time with his family. This time, there was no external search. On June 3, 2024, Matt Garman became the third CEO of Amazon Web Services. He inherited a business with $107.6 billion in annual revenue, a 30% market share, and a looming problem: Microsoft and Google were using AI as a wedge to steal AWS customers.

Garman's first task: prove that the 19 years he'd spent learning AWS's business from the inside out hadn't blinded him to the AI revolution happening outside Amazon's walls.

Part III: The AI Crossroads—How AWS Fell Behind

To understand AWS's AI predicament, you need to understand what happened in November 2022. That month, OpenAI launched ChatGPT, and within five days, it had one million users. Within two months, 100 million. The world had discovered that large language models could do more than complete sentences—they could write code, analyze documents, answer complex questions, and fundamentally change how people worked.

Microsoft saw the opportunity instantly. The company had invested $1 billion in OpenAI in 2019, and CEO Satya Nadella had been watching the GPT models evolve. In January 2023—just two months after ChatGPT launched—Microsoft announced a $10 billion investment in OpenAI at a $29 billion valuation. More importantly, Microsoft announced it would integrate OpenAI's models across its product suite: Office 365, Bing, GitHub, Azure.

The Azure OpenAI Service launched, giving enterprise customers access to GPT-3.5, GPT-4, and DALL-E models through Azure's cloud infrastructure. Suddenly, Azure wasn't just a cloud provider—it was the only place enterprises could access the most advanced AI models in the world under enterprise licensing terms. Microsoft's GitHub Copilot, powered by OpenAI's Codex model, began writing code for developers. Microsoft 365 Copilot started drafting emails and PowerPoint presentations.

Google, terrified of losing search dominance, rushed out Bard in March 2023 despite internal concerns about accuracy. CEO Sundar Pichai declared a "code red" and reorganized the company around AI. Google DeepMind's Gemini models launched in December 2023, positioning Google as a credible AI competitor to OpenAI.

And AWS? AWS had Amazon Bedrock, announced in April 2023 but not fully available until September 2023. Bedrock wasn't a foundation model—it was a marketplace where customers could access third-party models from Anthropic, AI21 Labs, Cohere, Meta, Stability AI, and Amazon's own models. AWS also had Amazon SageMaker, its machine learning platform, but SageMaker was built for data scientists training custom models, not business users wanting ChatGPT-like capabilities.

The strategic difference was stark. Microsoft had an exclusive partnership with the AI leader (OpenAI) and was integrating AI directly into products 1.5 billion people used daily. AWS was offering a buffet of models and betting that enterprises would want to choose their own. Microsoft was vertical integration; AWS was horizontal infrastructure.

In the crucial 18-month window from November 2022 to mid-2024, perception solidified: Microsoft was the AI cloud leader, and AWS was playing catch-up. The market share data told the story. In Q2 2025, AWS revenue grew 17.5% year-over-year to $30.9 billion. Microsoft's Intelligent Cloud segment (primarily Azure) grew 39% to $29.9 billion. Google Cloud grew 32% to $13.6 billion.

Even more concerning: generative AI-specific cloud services grew 140-180% in Q2 2025, according to Synergy Research Group. This explosive growth was concentrated in Azure and Google Cloud, not AWS. Customers wanted to run AI workloads, and they were choosing Microsoft and Google to do it.

What went wrong? Garman's answer—that AWS deliberately focused on infrastructure over flashy demos—contains truth but obscures uncomfortable realities. AWS didn't have a ChatGPT-like model to integrate into products. Amazon had no equivalent to Office 365 or Google Workspace where AI features could create obvious value. AWS's core business was renting compute and storage to other companies' applications, not building end-user productivity software.

Moreover, AWS's culture, inherited from Jeff Bezos, prized customer obsession and working backwards from customer needs. This approach works brilliantly for understanding existing needs but can create blindness to paradigm shifts. In 2022-2023, AWS teams were focused on optimizing EC2 instances, improving S3 storage economics, and expanding AWS's already-massive service portfolio of 200+ offerings. Generative AI? That was R&D, not something enterprises were asking for at scale.

By the time enterprises started asking—by mid-2023—Microsoft had a two-quarter head start and a superior positioning story: "We have the best AI models (OpenAI), integrated into software you already use (Office, GitHub), running on enterprise-grade infrastructure (Azure)."

Matt Garman inherited this strategic deficit in June 2024. He had six months to prepare for re:Invent, AWS's annual coming-out party. The December 2024 conference would be his chance to show the world that AWS could compete in AI. The question: How?

Part IV: The $125 Billion Infrastructure Bet

If Matt Garman was going to defend AWS's position, he needed to do what Amazon does best: outspend competitors and build infrastructure at scale. In February 2025, Amazon announced that its capital expenditure for 2025 would reach $100 billion, up from $83 billion in 2024. By October 2025, that estimate had risen to $125 billion for the full year—the highest capex spending of any technology company, exceeding Microsoft ($80 billion), Google ($75 billion), and Meta ($65 billion).

Where was that money going? More than 90% to AWS data centers, and approximately 90% of that for AI systems and supporting infrastructure. That meant roughly $86 billion in AI infrastructure spending in 2025 alone—nearly equivalent to AWS's total annual revenue in 2024.

The spending reflected a brutal reality: the AI infrastructure race requires unprecedented capital deployment. Training frontier AI models demands tens of thousands of NVIDIA H100 or Blackwell GPUs clustered together with ultra-fast networking. Inference—running those models for millions of users—requires different but equally expensive infrastructure optimized for speed and efficiency. AWS couldn't afford to lag in compute capacity; if customers couldn't get GPUs or AI accelerators from AWS, they'd go to Azure or Google Cloud.

But Garman faced a strategic choice: Should AWS simply buy NVIDIA chips like everyone else, or should it invest in custom AI accelerators that could provide cost and performance advantages but would take years to mature?

AWS chose both. In December 2024, Garman announced new EC2 P6 instances powered by NVIDIA's latest Blackwell GPUs, delivering up to 2.5 times faster compute than previous generation P5 instances. This ensured AWS customers would have access to the same cutting-edge NVIDIA hardware available on Azure and Google Cloud.

But the more interesting bet was Trainium, AWS's custom AI training chip. Announced at re:Invent 2024, Trainium2-based EC2 instances promised 30-40% better price-performance than GPU-based P5e instances. And Garman teased Trainium3, slated for late 2025, with EC2 Trainium3 UltraServers expected to be four times more performant than Trainium2.

AWS also pushed Inferentia, its custom chip for AI inference workloads. Inferentia2 offered up to 50% lower inference costs compared to GPU-based instances, with companies like Finch Computing claiming 80% lower costs and Dataminr reporting 9x better throughput per dollar after optimization.

The strategic logic: If AWS could convince customers to train and run models on Trainium and Inferentia instead of NVIDIA chips, AWS would capture more margin (since it designed the chips itself), reduce dependence on NVIDIA's supply constraints, and create switching costs (customers optimized for Trainium wouldn't easily move to Azure). Moreover, AWS could pass cost savings to customers, undercutting Azure and Google Cloud on price for equivalent performance.

The challenge: NVIDIA's CUDA software ecosystem had 20 years of developer mindshare, tooling, and optimization. Enterprises comfortable with CUDA faced real costs switching to AWS's Neuron SDK for Trainium and Inferentia. AWS needed not just competitive hardware but a compelling reason for customers to abandon the NVIDIA ecosystem they knew.

Enter Anthropic. In November 2024, Amazon invested an additional $4 billion in Anthropic (bringing total investment to $8 billion), with Anthropic naming AWS as its primary cloud provider and committing to train its foundation models on Trainium. In October 2025, AWS and Anthropic opened Project Rainier—an $11 billion data center complex in New Carlisle, Indiana, featuring half a million Trainium2 chips entirely devoted to Anthropic's workloads. The facility consumes 2.2 gigawatts of power across 30 buildings once complete, making it one of the largest AI-specific data centers in the world.

Anthropic wasn't just a customer—it was a showcase. If Anthropic, one of the world's leading AI labs, could train frontier models like Claude on Trainium instead of NVIDIA GPUs, it would validate AWS's chip strategy and give other enterprises confidence to make the switch. Claude's success would become Trainium's success.

But then came another twist. In November 2025, OpenAI—Microsoft's exclusive cloud partner—announced a $38 billion compute deal with AWS. OpenAI, which had been exclusively tied to Azure, was adopting a multi-cloud strategy and would begin running inference workloads on AWS infrastructure. This was a seismic shift: the AI leader was hedging against Microsoft, and AWS had just landed the industry's most prestigious customer.

Garman's infrastructure bet was coming together. AWS would offer the broadest compute options (NVIDIA, Trainium, Inferentia), the most capacity ($125 billion in spending ensured that), and partnerships with both leading AI labs (Anthropic and OpenAI). The question: Would this be enough to reverse AWS's slowing growth and recapture momentum from Microsoft and Google?

Part V: Amazon Bedrock and the Multi-Model Strategy

While Microsoft bet on exclusive OpenAI integration and Google bet on its own Gemini models, AWS pursued a different strategy: Amazon Bedrock, a marketplace offering access to dozens of foundation models from multiple providers. By late 2024, Bedrock featured more than 100 models including Anthropic's Claude, Meta's Llama, Cohere's Command, AI21 Labs' Jurassic, Stability AI's Stable Diffusion, and Amazon's own Nova models.

The Bedrock strategy reflected AWS's philosophical stance: customers want choice, not lock-in. Rather than forcing enterprises to adopt a single model family (as Microsoft did with OpenAI or Google with Gemini), AWS would let customers choose the model that best fit their use case, switch models easily, or even use multiple models for different tasks.

At re:Invent 2024, Amazon CEO Andy Jassy unveiled Amazon Nova, AWS's first family of frontier foundation models. Nova included six models spanning text, image, and video generation capabilities, with performance competitive to GPT-4o and Claude 3.5. Nova's existence addressed a key weakness: AWS no longer had to rely solely on third-party models; it could offer proprietary models while maintaining the multi-model marketplace.

Garman emphasized Bedrock's enterprise features: Retrieval-Augmented Generation (RAG) for grounding models in company data, Agents for multi-step task automation, Guardrails for content filtering, and Model Distillation for creating smaller, cheaper models from larger ones. Bedrock Intelligent Prompt Routing could automatically direct queries to the optimal model based on complexity, balancing performance and cost.

Perhaps most importantly, Bedrock Prompt Caching could reduce compute costs by up to 90% and latency by up to 85% for repeated queries—a huge advantage for production AI applications serving millions of users. These features targeted the gap between demo chatbots (easy to build) and production AI systems (hard to build reliably at scale).

By mid-2025, tens of thousands of customers had adopted Bedrock, including major enterprises across financial services, healthcare, media, and technology sectors. AWS claimed Bedrock was growing faster than any AWS service in history—though Amazon didn't break out specific revenue figures, making it impossible to verify against Azure OpenAI Service's performance.

The multi-model strategy had clear advantages. Customers nervous about OpenAI's governance (after the November 2023 Sam Altman firing drama) or uncomfortable with Google's advertising business model could choose Anthropic's Claude or Cohere's models. Enterprises in regulated industries could select models with specific compliance certifications. Companies worried about model obsolescence could diversify across multiple providers.

But the strategy also had weaknesses. Microsoft's tight OpenAI integration meant Azure customers got new GPT capabilities immediately upon release, often with Azure-specific optimizations. Bedrock, by contrast, had to wait for model providers to make new versions available, creating lag time. Moreover, Microsoft could offer bundled pricing—GitHub Copilot, Office 365 Copilot, and Azure OpenAI Service as a package—while AWS sold cloud services à la carte.

The real test: Would enterprises value Bedrock's flexibility more than Azure's simplicity? Garman was betting that as AI matured beyond the "ChatGPT novelty" phase into production deployments, enterprises would want the control and choice that Bedrock provided. Microsoft was betting that enterprises would prize tight integration and single-vendor support.

By late 2025, the verdict remained unclear. What was clear: AWS was no longer conceding the AI infrastructure narrative to Microsoft.

Part VI: The Wartime CEO—Leadership Under Pressure

Those who know Matt Garman describe him as "unflappable"—someone who maintains composure under pressure and doesn't let emotions dictate decisions. But they also note a "self-confidence that can at times come across as arrogance," born from 19 years of success at AWS. One source close to Amazon characterized Garman as a "wartime" leader, suggesting that AWS needed a more aggressive posture to compete in AI.

In his first six months as CEO, Garman demonstrated that aggressive posture through organizational changes and public positioning. In June 2024, shortly after taking over, he announced a streamlined leadership structure reducing bureaucracy and speeding decision-making. AWS had grown to 200+ services and tens of thousands of employees; Garman's mandate was to move faster.

Garman's public comments revealed a CEO willing to say what others wouldn't. In July 2025, he bluntly told enterprises that replacing junior developers with AI would be "the dumbest thing I've ever heard." This comment, made at a time when many companies were experimenting with AI coding assistants, sparked controversy. Garman clarified that AI would augment developers, not replace them, and that critical thinking and soft skills would matter more than ever.

The comment reflected Garman's broader philosophy about AI: tools that amplify human capabilities beat tools that attempt full automation. This aligned with AWS's product strategy—Bedrock, SageMaker, Q Developer (AWS's coding assistant)—all positioned as augmentation rather than replacement.

In another revealing moment, Garman told CNBC in August 2025 that "critical thinking is going to be the most important skill going forward" as AI handles routine tasks. He predicted employers would prioritize hiring workers with creativity and adaptability over narrow technical skills. This people-centric messaging differentiated AWS from competitors emphasizing headcount reduction through AI.

But Garman's most important leadership test came in defending AWS's strategic choices. When asked why AWS lagged in AI product releases, Garman could have blamed his predecessor Adam Selipsky or claimed the gap was smaller than it appeared. Instead, he owned it while reframing the narrative: "We didn't have some of the whiz-bangy things that you could get out there quickly. But our customers don't want whiz-bangy demos. They want infrastructure that works at scale."

This framing—infrastructure over demos, substance over flash—played to AWS's strengths while implicitly critiquing competitors. Microsoft might have ChatGPT integrated into Office, but could Azure scale to AWS's level of reliability and breadth? Google might have Bard, but did enterprises trust Google's commitment to cloud given its history of killing products?

The "wartime" mentality showed in AWS's deal-making. The $38 billion OpenAI cloud contract, announced in November 2025, was a major coup. OpenAI had been exclusively on Azure; breaking that exclusivity required AWS to offer compelling economics and capabilities. Similarly, the $8 billion Anthropic investment with dedicated Trainium infrastructure demonstrated AWS's willingness to spend aggressively to lock in strategic relationships.

Garman also pushed AWS's go-to-market teams to emphasize AI use cases in customer conversations. AWS re:Invent 2024 featured hundreds of sessions on building production AI applications, with Garman emphasizing that "the future value companies will get from AI will be in the form of agents"—autonomous systems that complete multi-step tasks. AWS launched Bedrock Agents and SageMaker Unified Studio to make agent development easier, positioning AWS for the next AI wave beyond chatbots.

By late 2025, early signs suggested Garman's leadership was working. AWS revenue growth re-accelerated to 20% in Q3 2025 (up from 17.5% in Q2), driven by AI workload adoption. Operating income in Q3 reached $10.4 billion on $33 billion revenue—a 31.5% margin that remained industry-leading despite massive infrastructure investments.

But challenges remained. Microsoft's Azure kept growing faster than AWS (though from a smaller base). Google Cloud was winning key AI deals and positioning itself as the preferred infrastructure for non-OpenAI models. And NVIDIA, with its chips powering all competitors, was capturing enormous value from the AI infrastructure boom—$3.3 trillion market cap by late 2025, surpassing both Amazon and Alphabet.

Garman's next test: proving that AWS's infrastructure advantages—global scale, broadest service portfolio, cost optimization, operational excellence—would matter more than Microsoft's AI integration or Google's model sophistication as enterprises moved from AI experiments to production deployments.

Part VII: The Future of Cloud AI—Unanswered Questions

As 2025 draws to a close, Matt Garman has successfully stabilized AWS's position in the AI infrastructure race. The company that was perceived as lagging in early 2024 now has credible offerings across foundation models (Amazon Nova), model marketplace (Bedrock), custom chips (Trainium/Inferentia), and strategic partnerships (Anthropic, OpenAI). AWS's $125 billion infrastructure spending ensures it won't be capacity-constrained, and its 30% market share remains the industry's largest.

But several fundamental questions remain unanswered, and their resolution will determine whether Garman's tenure is remembered as a successful defense of AWS's dominance or the beginning of a slow decline:

Can AWS's multi-model strategy beat Microsoft's OpenAI integration? Bedrock's flexibility appeals to enterprises that want choice and control, but Microsoft's tight GPT integration offers simplicity and rapid feature deployment. As AI moves from experimentation to core business processes, will enterprises prioritize flexibility or ease of use? The answer may vary by industry, company size, and technical sophistication—but if a clear winner emerges, it will reshape cloud market share.

Will Trainium gain meaningful adoption beyond Anthropic? AWS's custom chip strategy depends on customers beyond Anthropic adopting Trainium for training and Inferentia for inference. Early customers include Snap, Qualtrics, and Autodesk, but these remain a small fraction of AWS's customer base. If enterprises stick with NVIDIA's CUDA ecosystem due to familiarity and tooling maturity, AWS's multi-billion-dollar chip investment may fail to deliver sufficient returns or competitive differentiation.

Can AWS maintain margins while competing on AI infrastructure? The AI infrastructure race requires massive capital expenditure—AWS is spending $125 billion in 2025, nearly 120% of annual revenue. While Q3 2025's 31.5% operating margin remained strong, sustaining margins while building data centers and subsidizing customer migrations will be difficult. If Azure or Google Cloud undercut AWS on price to gain share, a margin war could erode the profitability that makes AWS Amazon's crown jewel.

Who wins the AI agent era? Garman believes "the future value companies will get from AI will be in the form of agents"—but so does everyone else. OpenAI is building agents, Microsoft is building agents, Google is building agents, and hundreds of startups are building vertical-specific agents. If the agent platform war mirrors the mobile OS war (two winners: iOS and Android), which platforms will dominate? AWS's Bedrock Agents face competition from LangChain, AutoGPT, Microsoft Semantic Kernel, and Google's Vertex AI Agent Builder.

Can AWS catch up on AI developer mindshare? Microsoft's GitHub Copilot has millions of developers writing code with AI assistance daily, creating habitual usage and data flywheel effects. Google's NotebookLM and AI Studio attract AI researchers and enthusiasts. AWS's Q Developer (its coding assistant) and SageMaker target similar audiences but lack the same cultural momentum. Developer mindshare translates to enterprise adoption as developers-turned-executives choose familiar platforms. If AWS remains perceived as the "infrastructure plumbing" while competitors own "AI innovation," that perception gap will compound over time.

What happens when OpenAI launches its own cloud? OpenAI's $38 billion AWS deal and multi-cloud strategy suggest the company is positioning for infrastructure independence. Sam Altman has discussed multi-trillion-dollar data center investments and sovereign AI infrastructure. If OpenAI eventually builds its own global cloud infrastructure—perhaps in partnership with Oracle or others—it could bypass both Microsoft and AWS, selling AI capabilities directly to enterprises. This would fundamentally disrupt cloud economics and force AWS to compete on infrastructure commodities (compute, storage) rather than differentiated AI services.

These questions don't have definitive answers in late 2025. What's clear is that the cloud AI infrastructure race is far from over, and the competitive dynamics remain fluid. AWS's advantages—operational excellence, customer relationships, service breadth, global scale—remain formidable. But Microsoft's AI integration and Google's model leadership pose existential threats to AWS's growth trajectory.

Matt Garman inherited AWS at a crossroads and has executed well to keep the company competitive. But "competitive" isn't the same as "winning," and AWS's slowing growth relative to Azure and Google Cloud suggests the company is in a battle for position, not consolidating a lead.

Conclusion: The Intern's Ultimate Test

Matt Garman's journey from 2005 intern to 2024 CEO of a $100 billion business is remarkable—a testament to Amazon's promote-from-within culture and Garman's technical and commercial acumen. But the real test isn't his rise; it's what comes next.

AWS built its dominance in an era when cloud computing was about migrating on-premise workloads to elastic infrastructure. Enterprises needed compute, storage, databases, and networking—and AWS offered the broadest, most mature portfolio. The competition was data centers and legacy IT vendors, not fellow cloud providers.

The AI era is different. Competition comes from Microsoft (with its OpenAI partnership and productivity software integration), Google (with its frontier models and search expertise), and NVIDIA (capturing enormous value through chip sales to all players). The workloads aren't lift-and-shift migrations but new AI-native applications that didn't exist three years ago. And customers aren't just IT departments; they're business leaders demanding AI solutions that generate revenue and reduce costs.

In this new landscape, AWS's traditional advantages matter less. Service breadth? Every major cloud provider now offers hundreds of services. Global infrastructure? Microsoft and Google are spending tens of billions to match AWS's footprint. Operational excellence? Azure's uptime and reliability have reached parity with AWS in most regions.

What matters now is AI ecosystem positioning: Which cloud makes it easiest to build, deploy, and scale AI applications? Which offers the most advanced models? Which provides the best developer experience? Which has the momentum among AI researchers, startups, and enterprises?

Garman's challenge is to prove that AWS's infrastructure-first approach—Bedrock's multi-model marketplace, Trainium's cost advantages, SageMaker's flexibility—will win against Microsoft's vertical integration and Google's model sophistication. It's a bet that enterprises will choose best-of-breed infrastructure over bundled simplicity, and that AWS's customer obsession will outpace competitors' AI innovation.

So far, the evidence is mixed. AWS's Q3 2025 growth re-acceleration to 20% suggests the company is fighting back successfully. The $38 billion OpenAI deal proves AWS can compete for premier customers. Amazon Nova's launch shows AWS can build credible foundation models, not just resell third-party ones. And the $125 billion infrastructure investment ensures AWS won't be outspent or outcapacitated.

But Microsoft and Google aren't standing still. Azure's 39% growth and Google Cloud's 32% growth demonstrate that customers are voting with their wallets, and many are choosing alternatives to AWS for new AI workloads. If this trend continues, AWS's market share—30% today—could erode to the mid-20% range by 2027, fundamentally changing the cloud industry's power structure.

Matt Garman has a finite window—perhaps 18-24 months—to prove AWS's AI strategy works. If by mid-2027, AWS growth remains in the high teens while Azure and Google Cloud sustain 30%+ growth, Amazon's board will face difficult questions about whether a different strategy or leader is needed. If AWS growth re-accelerates to the mid-20% range and operating margins hold above 30%, Garman will have successfully navigated the transition and secured AWS's dominance for another decade.

The stakes couldn't be higher. AWS generates more than $40 billion in annual operating income—60% of Amazon's total operating profit. That cash funds Amazon's retail ambitions, content production for Prime Video, Alexa development, satellite internet through Project Kuiper, and everything else Amazon does. If AWS stumbles, all of Amazon feels the impact.

Twenty years ago, Jeff Bezos famously bet that Amazon could become the world's cloud computing leader by offering its internal infrastructure as a service. It worked spectacularly, creating a business more profitable than retail despite being unrelated to Amazon's core e-commerce DNA.

Now Matt Garman faces an equally consequential bet: that in the AI era, infrastructure will still matter more than applications, that choice will beat integration, and that AWS's scale and customer obsession will overcome its late start in generative AI.

Will Garman's 19-year apprenticeship at AWS—from intern to EC2 product manager to compute GM to sales leader to CEO—prove sufficient preparation for defending a $100 billion profit machine in the most disruptive technology shift since cloud computing itself?

The answer will define not just Matt Garman's legacy, but the future of cloud computing—and Amazon's ability to remain one of the world's most valuable and influential technology companies.