Heavy-Weight A.I. Clash: Amazon vs Alphabet

Differentiated Approaches - Which Is The Better Option For Exposure to The AI Revolution?

---

Disclaimer & Disclosure: The author has a position in Amazon. This post is for informational purposes only and should not be construed as investment advice. Conduct your own due diligence and seek professional investment advice before making any investment decisions.

---

The Heavy-Weights In AI Today

The market is in a frenzy over Artificial Intelligence — the next great frontier brimming with investment potential. Since OpenAI’s launch of ChatGPT stunned the world, a wave of capital has flooded into the AI space. Every investor is searching for their angle, a way to profit from the rise of the AI economy. Yet surprisingly few companies are truly positioned to capture value across the entire chain.

Big tech market caps soar with the onset of the AI revolution

The AI economy, as it turns out, isn’t a single opportunity but rather a layered ecosystem where dominance at one level doesn’t guarantee success at another.

This makes the two companies that have assembled complete, end-to-end AI capabilities particularly compelling from an investment perspective.

When we examine the architecture of this new economy, four distinct but interconnected layers emerge.

• Computing of any kind needs the requisite hardware. Without high powered chips, AI simply wouldn’t be possible. So semiconductor manufacturing provides the physical backbone upon which all AI computation occurs. This is the foundation.

• The next layer is the AI platforms themselves, the large language models and machine learning systems that deliver intelligence - the AI brains.

• But these AI brains are worthless unless they are accessible and given purpose. So they live in the cloud - a vast network of distributed systems that provide the scale and accessibility modern AI demands.

• Finally, at the top sits the application layer, where AI capabilities translate into tangible business value through products and services that customers actually use and pay for.

The strategic significance of controlling multiple layers cannot be overstated. Each layer captures different economics, faces different competitive dynamics and provides different defensive moats.

Semiconductors offer high barriers to entry but require enormous capital investment. Cloud infrastructure benefits from massive economies of scale and switching costs. AI platforms demand extraordinary technical talent and vast computational resources for training. Applications require distribution, brand strength and deep understanding of customer needs.

So, at least theoretically, a company that successfully integrates across these layers can optimize the entire stack, capture more value and create competitive advantages that are extraordinarily difficult to replicate.

Analysis of major technology companies’ positioning across AI platforms, cloud infrastructure, semiconductors and AI-driven businesses

Only Alphabet (GOOG, GOOGL) and Amazon (AMZN) possess all four elements of the AI economy stack (4/4), making them uniquely positioned to capture value across the entire AI value chain. Microsoft comes close at 3.5/4, while pure-play specialists like NVIDIA, Anthropic, and OpenAI excel in single layers but depend on partners for complete solutions.

So let’s focus on Alphabet (Google’s parent company) and Amazon.

Alphabet: The Vertically Integrated AI Powerhouse

Alphabet’s position in the AI economy represents perhaps the most comprehensive vertical integration in technology history. The company’s advantages begin at the semiconductor level with its Tensor Processing Units, or TPUs. Unlike traditional CPUs or even GPUs, TPUs were designed from the ground up specifically for the matrix multiplication operations that dominate neural network training and inference. Google began developing these chips internally as early as 2015, recognizing that the economics of AI at Google’s scale demanded custom silicon.

The brilliance of the TPU strategy extends beyond mere cost savings, though those are substantial. By controlling its own chip design, Google can optimize the entire hardware/software stack in ways that companies dependent on third-party semiconductors simply cannot match. When Google’s AI researchers need specific computational capabilities, those requirements can flow directly into the next TPU generation. This creates a virtuous cycle where hardware informs software architecture and vice versa, with each generation becoming more efficient and capable than general-purpose alternatives.

Importantly, Google now makes TPUs available to external customers through Google Cloud Platform, transforming what began as an internal efficiency project into a competitive weapon and potential profit center. The pricing structure for TPU access consistently undercuts equivalent NVIDIA GPU capacity, sometimes dramatically so.

Although NVIDIA GPUs are considered the gold standard, for many workloads, particularly inference operations where a trained model processes new data, TPUs deliver superior price/performance ratios. In other words, they may not match NVIDIA’s cutting-edge GPUs for absolute performance on every possible workload, but for the vast majority of practical AI applications, they’re more than sufficient and considerably cheaper.

Thomas Kurian, CEO at Google Cloud recently stated, “Anthropic’s choice to significantly expand its usage of TPUs reflects the strong price-performance and efficiency its teams have seen with TPUs for several years.” He went on to say that, “We are continuing to innovate and drive further efficiencies and increased capacity of our TPUs, building on our already mature AI accelerator portfolio, including our seventh generation TPU, Ironwood.”

Looking beyond TPUs, Alphabet’s CEO, Sundar Pichai, has claimed a significant leap in Quantum computing, “Our Willow chip has achieved the first-ever verifiable quantum advantage. Willow ran the algorithm - which we’ve named Quantum Echoes - 13,000x faster than the best classical algorithm on one of the world’s fastest supercomputers... This breakthrough is a significant step toward the first real-world application of quantum computing, and we’re excited to see where it leads.”

Perhaps it’s only a matter of time before NVIDIA GPUs are displaced entirely. As competition heightens and the chip market matures, it would explain why Jensen Huang at NVIDIA is actively diversifying beyond traditional GPUs, evolving into a broad-based AI infrastructure and platform company.

Moving up the stack, Google Cloud Platform (GCP) itself has evolved from a distant third player behind AWS and Azure into a genuinely competitive cloud infrastructure provider. While still smaller than its rivals in overall market share, GCP has carved out particular strength in data analytics and machine learning workloads.

More particularly, the integration between GCP and Google’s AI capabilities creates natural synergies. Companies that want to build on Google’s AI models find the path of least resistance runs through GCP infrastructure. Those using GCP infrastructure discover that Google’s AI tools are deeply integrated and often superior to alternatives. Each layer of the stack feeds the next, which highlights the huge competitive benefit derived from being a full stack provider.

Google’s journey in terms of developing its own AI model is an interesting story.

Google stood at the absolute forefront of artificial intelligence research throughout the 2010s, possessing unmatched technical talent and revolutionary breakthroughs that would reshape the entire industry.

In 2017, eight researchers from Google Brain published “Attention Is All You Need,” introducing the transformer architecture, the foundational technology underlying virtually every modern large language model used today and enabling the entire AI boom.

Google had assembled the densest concentration of AI talent globally, including luminaries like Ilya Sutskever, Geoff Hinton and Dario Amodei. Yet despite creating the technology powering today’s AI revolution, Google hesitated to fully capitalize on its invention, primarily due to fears about cannibalizing its extraordinarily profitable search advertising business that generated tens of billions annually.

The company had even developed an internal chatbot called Mina in the late 2010s that was functionally similar to what would much later become ChatGPT, but leadership deemed it too risky to ship publicly, worried about both the business model implications and potential legal risks of disintermediating publishers.

Alphabet possessed all the necessary components to lead in Artificial Intelligence, even dominate in the field in the way it had with Google search, but failed to synthesize them into a competitive product due to organizational inertia and fear of cannibalizing its own golden-goose of a search business.

This strategic hesitation created a vacuum that competitors eagerly filled and triggered a devastating exodus of the very talent that had made Google the AI leader.

All eight authors of the transformer paper eventually left Google to pursue opportunities elsewhere, including moves to OpenAI and founding entirely new companies. The brain-drain accelerated when prominent researchers like Ilya Sutskever departed for OpenAI in 2015. Dario Amodei left Google Brain in 2016 to join OpenAI before later founding Anthropic in 2021 with other former OpenAI researchers. Even Noam Shazeer, the legendary engineer who had rewritten the transformer code and made it work, eventually left in frustration to found Character.AI after Google repeatedly declined to ship his chatbot products.

This dispersal of concentrated AI expertise amounted to Google sewing the seeds of the AI industry and giving away its insurmountable advantage. Multiple competitors came into being, with each taking crucial proprietary knowledge about how to build and scale large language models.

ChatGPT’s viral launch on November 30, 2022 forced Google into crisis mode, with CEO Sundar Pichai declaring a “code red” and bringing back founders Larry Page and Sergey Brin. ChatGPT reached 100 million users in a matter of weeks, the fastest product in history to hit that milestone and the very disruption that Alphabet feared was fast becoming reality.

Google executive Sissie Hsiao testified at an antitrust trial that since ChatGPT’s debut, Google had seen declines in certain search categories, and that Vidhya Srinivasan, the executive overseeing Google’s ads business, believed AI would ultimately cannibalize Search revenue.

Microsoft’s subsequent $10 billion investment in OpenAI and integration of GPT into Bing search compounded Google’s nightmare scenario.

The company found itself in the uncomfortable position of playing catch-up in a field it had pioneered, watching as OpenAI capture the public imagination, while Anthropic built a reputation for safety-focused AI development.

Alphabet’s belated response felt reactive rather than visionary, a stark contrast to its historical position as the industry’s innovation leader. In February 2023 it rushed the launch of Bard, powered by its Lambda model, but the launch proved disastrous. The product delivered factually incorrect information in its own demo video, causing Google’s stock to drop 8% that day.

Bard was clearly inferior to ChatGPT, revealing just how far behind Alphabet had fallen in a field it had pioneered.

However, Google has mounted a formidable comeback, leveraging its vast resources, infrastructure, and remaining technical talent to re-establish itself as a serious competitor in the AI race.

In mid-2023, Sundar Pichai took the bold step of merging Google Brain and DeepMind into a single unified organization, ending years of internal competition and duplication. He also mandated that the company would standardize on one model, Gemini, across all products and services, fundamentally changing how Google approached AI development. It also brought back some of the talent it had previously lost, reuniting some of the most talented engineers in AI.

The results have been remarkable: Gemini models have achieved strong performance, with Gemini Pro outperforming ChatGPT in several benchmarks, especially in code and reasoning tasks. The company now claims 450 million monthly Gemini users.

Google’s comprehensive AI stack spanning custom TPU chips, massive cloud infrastructure, world-class research capabilities, and billions of users means it remains firmly in the game.

The company’s integrated approach - embedding Gemini across Google Workspace, Search, Pixel devices and other products, provides distribution advantages that pure-play AI businesses can’t match.

It benefits from better unit economics, an ability to generate $50 billion in annual Google Cloud revenue from its AI offerings, the opportunity to capitalize on distribution through Google search - which is de-facto the front door to the internet, it remains a leader in computational resources with some of the best AI research talent, plus Google has unparalleled access to training data with the ability to amortize training costs across trillions of inference tokens.

In short, today it possesses all four critical elements of the AI stack: foundational models, cloud infrastructure, custom semiconductors, and AI-driven applications at scale - plus scale economies that others will struggle to compete with.

While the early stumbles allowed competitors to seize initiative and mindshare, Google’s fundamental strengths in infrastructure and remaining talent density allowed it to claw its way back to the top.

Alphabet also has another huge advantage over other AI model providers. In contrast to others, many of which struggle with profitability, it has a robust self-sustaining profitable business which will ensure it remains formidably positioned as the AI economy evolves.

What distinguishes Google’s approach at the platform layer is the tight integration with its existing product ecosystem. Gemini isn’t offered merely as a standalone API or chatbot competitor. Instead, it’s being woven throughout Google’s entire product portfolio.

Google Search alone represents one of the most AI-intensive applications on earth, having incorporated machine learning into its ranking algorithms for years before the current AI boom. Every search query benefits from sophisticated natural language understanding, personalization, and prediction. The advertising system that monetizes this search traffic employs AI extensively to match ads to queries, predict click-through rates, and optimize bidding strategies. This isn’t speculative future revenue, it’s a proven, highly profitable business that AI makes continuously better.

Beyond search, Google operates a portfolio of AI-native or AI-enhanced businesses that few competitors can match.

YouTube is the world’s second most visited website, just behind Google, with between 74.8 and 77.9 billion monthly visits, beating all other social platforms. Alphabet owns them both. In fact, YouTube is either the second or third largest social media network globally; only Facebook (and by some measures WhatsApp) surpasses its user base. YouTube’s recommendation system, which drives the majority of viewing time on the platform, represents one of the world’s most sophisticated machine learning applications.

Then there are the other services: Gmail’s spam filtering, smart compose features and automatic categorization all rely heavily on AI. Google Maps uses machine learning to predict traffic patterns, suggest routes, and estimate arrival times with remarkable accuracy. Google Photos employs computer vision to organize and search images. Android leverages AI for everything from voice recognition to battery management.

Let’s not forget about Waymo, perhaps the most ambitious and technically challenging AI application any major technology company has undertaken. Autonomous vehicles represent AI’s ultimate real-world test: computer vision, sensor fusion, prediction, planning and control all operating in real-time with safety-critical consequences. Waymo has accumulated more autonomous driving miles than any competitor and now operates genuine robotaxi services in multiple cities, generating actual revenue from paying customers. While the path to full commercialization remains long and capital-intensive, Waymo’s lead in this space is substantial and the total addressable market for autonomous transportation is measured in trillions of dollars.

This embedded approach means Google can monetize its AI capabilities not just through direct API access but through enhanced functionality across products that already generate substantial revenue.

The strategic coherence of Alphabet’s AI stack creates multiple reinforcing advantages. Custom TPUs make AI more economical, enabling more aggressive deployment across Google’s products. More AI deployment generates more data and more insights about what works, informing both model development and chip design. Better AI capabilities make Google’s products more useful, driving more usage and more revenue. More revenue funds more AI investment. Each layer supports and enhances the others in ways that would be extraordinarily difficult for a less integrated competitor to replicate.

The only other company on the planet with a full AI stack is Amazon, so how does its model compare to that of Google?

Amazon: The Open Platform Full Stack Strategy

Amazon’s approach to the AI economy differs fundamentally from Alphabet’s vertical integration, yet proves equally comprehensive in its coverage of all four critical layers. Where Google builds walls around its ecosystem, Amazon constructs marketplaces. Where Google optimizes for integration, Amazon optimizes for choice. These differences reflect not just strategic preference but the distinct DNA of each company and the different businesses from which they emerged.

At the semiconductor layer, Amazon’s AWS has developed its own custom chips: Trainium for training AI models and Inferentia for running inference workloads. Like Alphabet’s TPUs, these chips were born from economic necessity. When you operate cloud infrastructure at AWS’s scale, and your customers increasingly demand AI capabilities, the cost of relying entirely on third-party semiconductors becomes prohibitive. By designing custom silicon optimized for specific workloads, AWS can offer better price-performance ratios to customers while improving its own unit economics.

While Trainium isn’t a direct replacement for Nvidia’s high-end GPUs, it doesn’t have to be: it offers a viable, lower-cost solution that is perfectly suited for many AI training needs. According to AWS, Trainium offers comparable performance while reducing costs by 25%.

Another key advantage AWS offers is accessibility. By promoting Trainium, the company enables customers to experiment with AI training and inference workloads without long wait times or premium pricing for Nvidia’s high-demand GPUs.

The Trainium and Inferentia chips represent a substantial engineering achievement and a meaningful strategic asset. They provide AWS customers with alternatives to NVIDIA GPUs that, for many use cases, deliver comparable performance at significantly lower cost. Training a large language model on Trainium instances might cost 30-40% less than equivalent NVIDIA capacity, while inference on Inferentia can be even more economical. For enterprises watching their AI budgets carefully - which is to say, virtually all of them - these cost savings translate directly into competitive advantage for AWS.

Enterprises used to working with Nvidia’s ‘Compute Unified Device Architecture’ (CUDA) need to think about the cost of switching to a whole new platform like Trainium, but once the switching cost issue has been overcome - and the discounted chip price ought to help make that happen - one needs to ask, “Is Amazon about to disrupt yet another industry?”

Critically, however, Amazon doesn’t force customers onto its custom chips. AWS continues to offer extensive NVIDIA GPU capacity alongside its proprietary options. Customers can choose based on their specific requirements, performance needs and budget constraints. This optionality reflects Amazon’s fundamental platform philosophy: provide the best tools and let customers decide what works for their situation. It’s the same approach that made AWS dominant in cloud infrastructure generally.

Speaking of cloud infrastructure, AWS remains the unquestioned market leader despite strong competition from Microsoft Azure and Google Cloud. The advantages of incumbency in cloud computing are substantial and self-reinforcing. AWS has more data centers in more regions than anyone else. It offers more services, hundreds of them, addressing increasingly specialized needs. Its tooling and ecosystem are more mature than either of its competitors. Its sales and support organization is most developed. Most importantly, the installed base of applications running on AWS creates enormous switching costs. Migration to another cloud provider is technically possible but operationally complex, risky and expensive enough that most companies won’t undertake it without compelling reason.

It has been suggested by some that the only reason Google Cloud and Microsoft Azure have such robust market share is that most large enterprises with disaster recovery plans like to have redundancy built in to their operating model. In other words, AWS is the primary cloud provider and the other two supply back up capacity in case of emergencies.

Amazon’s cloud leadership provides the foundation for Amazon’s AI strategy. Every AI model, regardless of who developed it, requires computational infrastructure to train and run. By controlling that infrastructure and making it world-class, AWS positions itself to capture value from the AI boom regardless of which specific models or applications become dominant. It’s a classic platform play, reminiscent of how Microsoft captured value from the PC revolution by controlling the operating system layer rather than betting on specific applications.

The AI platform layer is where Amazon’s strategy becomes particularly distinctive and fascinating. Rather than competing primarily with its own models, AWS introduced Amazon Bedrock. This is a model-agnostic platform that functions as an AI marketplace. Through Bedrock, customers can access Anthropic’s Claude models, Meta’s Llama, Cohere’s command models, AI21 Labs’ Jurassic, Stability AI’s image generation models and others, all through a unified API and billing relationship. It’s a deliberate choice to prioritize customer choice over proprietary lock-in.

This doesn’t mean Amazon has abandoned building its own models. The company recently launched its Nova family of models, offering text, image and video generation capabilities at competitive price points. But Nova exists within the Bedrock ecosystem as one option among many, not as the forced default. Amazon is essentially saying to customers: “We’ve built capable models if you want them, but if you prefer Anthropic or Meta or someone else, we’ll make that easy too.” It’s Amazon’s tried and tested strategy - it worked fantastically well for AWS and is likely to prove a winning formula in relation to AI models. It’s all about making the lives of customers easier rather than forcing them into any single solution.

The Bedrock strategy carries both risks and advantages. The obvious risk is commoditization: if customers can easily switch between models, Amazon captures less value and faces more competition. But the advantages are equally significant. By supporting multiple models, Amazon increases the total addressable market for its infrastructure and lowers the barrier to AI adoption. Customers worried about vendor lock-in or betting on the wrong AI model can experiment freely, knowing they’re not trapped. This optionality itself becomes a competitive advantage, particularly in enterprise sales where hedging and flexibility are highly valued.

Moreover, the Bedrock approach positions AWS to win regardless of how the AI model landscape evolves. If Claude becomes the dominant enterprise AI, Amazon benefits through its strategic investment in Anthropic and Bedrock’s hosting relationship. If Llama or another open-source model prevails, Amazon still provides the infrastructure. If Amazon’s own Nova models prove competitive, excellent, but if they don’t, the company’s AI strategy doesn’t collapse. This portfolio approach reduces risk while maintaining upside across multiple scenarios.

At the application layer, Amazon’s AI deployment is pervasive but often less visible than Google’s because it’s deeply embedded in operational systems rather than consumer-facing products. The company’s fulfillment centers represent perhaps the world’s most sophisticated implementation of AI-powered robotics and logistics optimization. Thousands of autonomous robots navigate warehouse floors, collaborating with human workers in a carefully choreographed dance optimized by machine learning algorithms. Inventory positioning, shipping route optimization and demand forecasting are all powered by AI systems processing vast amounts of data to shave seconds and pennies from millions of transactions.

Alexa, while struggling to achieve profitability as a standalone business, represents a massive deployment of voice AI and smart home technology in hundreds of millions of devices worldwide. The natural language processing, intent recognition and integration across thousands of third-party services required to make Alexa function represents a substantial AI achievement and creates valuable network effects even if direct monetization remains elusive.

Amazon’s retail recommendation systems drive billions of dollars in incremental sales annually through sophisticated collaborative filtering and personalization algorithms. The dynamic pricing systems adjust prices millions of times daily based on supply, demand, competitor pricing and individual customer propensity to purchase. The sponsored product advertising business uses machine learning extensively to match products with search queries and optimize ad placement. Throughout the shopping experience, AI operates invisibly, nudging customers toward purchases and optimizing Amazon’s economics.

AWS itself increasingly serves as an application layer where AI enhances the product. Services like Amazon Forecast, Amazon Personalize, Amazon Rekognition for image analysis, Amazon Textract for document processing and dozens more offer pre-built AI capabilities that customers can incorporate into their own applications without building models from scratch. These managed AI services lower the barrier to AI adoption and create additional revenue streams beyond raw infrastructure.

Last, but by no means least, while Alphabet has Waymo, Amazon has Zoox - its own robo-taxi service which is fully operational. Unlike Waymo, which retrofits conventional cars with its technology, the Zoox vehicles are custom made, based on the design of horse-drawn carriages to provide a better user experience. (Learn more about Zoox and to watch a video exploration of the service within this post).

The strategic coherence of Amazon’s approach, while different from Alphabet’s, is equally compelling. AWS’s infrastructure dominance gives Amazon first access to customer AI workloads and insight into emerging needs. The Bedrock marketplace ensures AWS remains relevant regardless of which models win. Custom chips improve economics for both Amazon and its customers. The vast AI deployment across Amazon’s own operations provides continuous feedback about what works at scale, informing both chip development and service offerings. And the financial resources generated by AWS and retail fund continued investment across the entire stack.

Why Isn’t Microsoft An AI Heavy-Weight?

The Microsoft (MSFT) story is often told as one of relentless innovation and visionary leadership, but the truth is far more complicated, and in many ways, more revealing. For decades, Microsoft has built its empire not on invention, but imitation. Its history reads like a catalog of reactions to the success of others.

MS-DOS, the product that launched the company, was based on QDOS - literally “Quick and Dirty” - software acquired from a third party which amounted to a borrowed foundation that set the tone for what would follow.

Windows, the crown jewel of Microsoft’s early dominance, lifted its graphical user interface concepts from Apple and Xerox. When Netscape ruled the web, Microsoft responded by acquiring Spyglass Mosaic to create Internet Explorer. Even its most familiar features, like the Windows Task Manager, mirrored Apple’s “Force Quit” long before anyone used “Ctrl+Alt+Del” as shorthand for frustration.

This pattern repeated across decades. Microsoft Office, now a near-universal workplace staple, didn’t start as a pure in-house innovation. It evolved from acquisitions and ideas lifted from WordPerfect and Lotus 1-2-3. When Apple redefined portable music with the iPod, Microsoft tried to follow with the Zune. When Google dominated search, Microsoft launched Bing. When Sony and Nintendo ruled gaming, Microsoft countered with the Xbox. And when the world moved to smartphones, Microsoft was once again late to the party. It paid $7.2 billion to acquire Nokia in a last-ditch attempt to compete with iOS and Android. The deal ended in disaster: 18,000 layoffs and one of the worst M&A deals in corporate history.

Even in the modern era, Microsoft’s strategy has remained more adaptive than original. Its Surface tablets arrived years after Apple had defined the category with the iPad. Teams only appeared once Slack had revolutionized workplace communication. The Edge browser now runs on Chromium, Google’s open-source foundation, after Microsoft abandoned its own browser engine. The company bought its way into social networking through LinkedIn and into software development communities through GitHub. Its cloud platform, Azure, was late to market, following Amazon’s AWS and Google Cloud’s early advances.

Perhaps most striking of all is Microsoft’s belated arrival to the artificial intelligence revolution. Despite being a software powerhouse with 40 years of experience, spending tens of billions on AI research over more than a decade and with over 1,500 dedicated researchers, it found itself lagging behind OpenAI, a startup founded as recently as 2015. When ChatGPT launched, CEO Satya Nadella reportedly vented in frustration, demanding to know, “Why do we have Microsoft Research at all?” The response was typical of Microsoft - “if you can’t beat them, join them” - it made huge investments in OpenAI and will now own 27% of that business - an expensive admission that Microsoft, the world’s premier software house, had missed the future it should have owned.

This says a great deal about the culture of the company. It is incredible how it has managed to dominate and that it still ranks among the Magnificent 7. Its success says as much about the power of scale, persistence and strategic imitation as it does about innovation. However, make no mistake, Microsoft is no challenger to either Alphabet or Amazon. They will continue to lead, and in all likelihood, Microsoft will continue to follow in their shadow, several steps behind.

The Investment Case

From an investment perspective, both Alphabet and Amazon present intriguing opportunities rooted in their comprehensive AI capabilities, but the risk-reward profiles differ in interesting ways.

Alphabet’s more integrated approach offers potentially higher margins if its AI models achieve or maintain technical leadership. The company’s existing profit engines in search and advertising are already AI-enhanced, meaning AI improvement translates directly into financial improvement in businesses that already generate enormous cash flows. The technical talent density at Alphabet remains extraordinary and the company’s research contributions to AI continue at the highest level. Waymo, if it achieves large-scale commercialization, could create an entirely new business measured in hundreds of billions of dollars. The valuation multiple on Alphabet remains relatively modest compared to its growth prospects, partially due to regulatory concerns and questions about whether the company can maintain search and advertising dominance in an AI-native world.

The risks center largely on execution and competitive dynamics. Can Alphabet maintain its technical edge in AI models against well-funded competitors? Will regulatory pressure force the breakup of Google’s advertising business or otherwise constrain the company? Can Google successfully transition its search business to accommodate AI-powered answers without cannibalizing its lucrative advertising model? Will Waymo’s capital requirements and timeline to profitability strain patience and resources? These questions introduce uncertainty but also create the valuation discount that makes the opportunity interesting.

Amazon’s diversified approach offers perhaps lower individual business risk but potentially lower margins in AI services due to the marketplace model. The company’s infrastructure advantage is substantial and durable, less dependent on winning the AI model race. AWS’s growth continues to accelerate and operating margins in that business remain highly attractive. The retail business, while lower margin, continues to gain market share and improve efficiency through AI deployment. Amazon’s balance sheet and cash generation capability support continued investment across multiple AI initiatives without requiring perfect execution on any single bet.

The risks here involve competitive pressure on AWS from Microsoft and Google, potential margin compression as cloud computing matures and questions about whether Amazon’s more distributed AI strategy results in slower progress than competitors’ focused approaches. The Bedrock marketplace, while strategically sound, essentially invites competition onto Amazon’s platform, which could create complex dynamics if, for example, Anthropic or another provider begins to view AWS as much platform as partner.

Summing Up

What makes both companies particularly attractive is that their AI investment theses don’t require believing AI will completely transform every industry overnight. Both are already profitable, cash-generative, market-leading businesses where AI makes existing operations better. AI isn’t a speculative bet on a distant future, it’s a present-day competitive advantage being deployed across billions of dollars of current revenue. The optionality that AI creates for entirely new businesses like autonomous vehicles or AI-native enterprise software represents an extra layer of cream on an already tasty cake.

The comprehensive nature of both companies’ AI stacks creates competitive moats that are exceptionally difficult to challenge. A potential competitor would need to simultaneously achieve breakthrough performance in semiconductor design, build hyperscale cloud infrastructure, develop state-of-the-art AI models and create compelling applications, all while competing against incumbents with decade-long head starts, deeper pockets and existing customer relationships. The barriers to replicating what Alphabet and Amazon have built are measured not just in billions of dollars but in years of accumulated learning, talent and infrastructure.

Perhaps most compelling is that investors don’t have to choose between these approaches. Alphabet’s integrated model and Amazon’s marketplace strategy represent different but complementary ways of capturing value from the AI economy. Both can succeed and, indeed, both are succeeding already. The AI economy is large enough and expanding fast enough to support multiple winners, particularly winners as well-positioned as these two.

At the beginning of this post was a chart that demonstrated how the valuations of big tech firms have risen following the launch of ChatGPT as capital has chased the AI boom.

The chart below shows a striking contrast. Over the same period, Alphabet’s stock (orange line) has surged 175%, while Amazon’s (blue line) has gained just 98%. Yet, both companies have grown revenue by roughly 31% over the same period. The real difference lies in profitability: Amazon’s operating margins have expanded significantly as high margin AWS makes an increasingly larger contribution to what used to be a low margin e-commerce driven business. As such cash flow from operations has climbed 2.6x. At Alphabet, operating cash flow has risen only 1.4x, with margins largely unchanged. This may suggest that the market may still be undervaluing Amazon’s potential as a genuine AI play.

Share price change since the launch of ChatGPT marked the start of the AI revolution

In an era where much AI investment focuses on speculative pure-plays with uncertain business models and distant profitability, Alphabet and Amazon offer something rare: comprehensive AI capabilities embedded within proven, profitable, scaled businesses that are already benefiting from AI deployment today while maintaining massive optionality on AI-driven opportunities tomorrow. That combination of present cash flow and future potential, all underpinned by technical leadership across the complete AI stack, makes both companies worthy of serious consideration for investors seeking exposure to artificial intelligence’s transformation of the economy.