Is AI Really a Commodity? What Investors Need to Know

The Commodity Argument Is Going Viral — And It's Half Right

Over 1.1 million people saw Gary Marcus's tweet arguing that AI is a commodity with no moat, no clear winners, and no pricing power. Steve Eisman — the investor who famously shorted the 2008 housing bubble — echoed the same view publicly: AI models are interchangeable, users switch constantly, and the hundreds of billions being spent by Meta, Google, Microsoft, and Amazon will ultimately produce returns no better than commodity margins.

It's a compelling argument. And it's also incomplete.

Understanding why requires looking beyond the product itself — at how commodities have historically been packaged, distributed, and monetised into durable, high-margin businesses. For anyone learning how stocks work for beginners or trying to understand how to invest for beginners in stocks, the AI commodity debate is one of the most instructive real-world case studies playing out in markets right now.

Let's break it down.

Why the Bears Have a Point: AI Models Are Converging

The bearish case isn't baseless. Right now, there are more than 100 AI models publicly available, spanning offerings from OpenAI, Google, Anthropic, Meta, Mistral, xAI, and dozens of others. Performance benchmarks show these models constantly leapfrogging each other — OpenAI leads one month, Google's Gemini the next, then Grok. No single provider holds a sustained technical lead.

From a classical economics standpoint, this looks textbook commodity-like:

• Low switching costs: Users can move between ChatGPT, Gemini, and Claude in minutes.
• Undifferentiated core product: A large language model is a large language model. The underlying architecture across providers is remarkably similar.
• Intense competition: Every major tech company and dozens of well-funded startups are racing toward the same destination.
• Price pressure: OpenAI, Google, and Anthropic have all cut API prices significantly over the past 18 months as competition intensifies.

Steve Eisman also raises a structural concern worth taking seriously: the hyperscalers are transforming from asset-light businesses — the model that made them extraordinarily profitable — into capital-intensive ones. Meta, which generated enormous free cash flow by spending very little on physical infrastructure relative to revenue, is now projecting capex in the range of $60–65 billion for 2025 alone. Microsoft, Google, and Amazon are on similar trajectories. That shift compresses the free cash flow that shareholders previously captured.

For investors, both concerns are legitimate inputs. The question is whether they lead to the right conclusion.

How Stocks Work: The Commodity Packaging Principle

Here's where the commodity argument breaks down — and where understanding how stocks work for beginners becomes genuinely useful. The value of a business is rarely found in the raw product. It's found in the system built around that product.

Consider Amazon Web Services. Before AI became the dominant market narrative, AWS was already one of the most profitable divisions in corporate America. Its flagship product, S3 object storage, is — by definition — a commodity. Storing data on a server is not a proprietary capability. Yet AWS commanded operating margins of approximately 30% on storage and cloud infrastructure revenue that exceeded $60 billion annually before AI-driven demand accelerated growth further.

How? AWS did not sell storage. It sold:

• Global availability across dozens of regions and availability zones
• Security and compliance frameworks trusted by regulated industries
• Lifecycle policies and versioning that made data management automated
• Thousands of integrated developer tools that created switching costs organically
• Service-level agreements that enterprises could stake their operations on

The commodity — raw storage — became the foundation. The economic value came from everything layered on top of it. Crucially, very few companies could replicate the full stack. Google Cloud, Microsoft Azure, and AWS hold roughly 65% of global cloud infrastructure market share between them, despite the underlying technology being theoretically replicable.

Three Non-Tech Examples That Prove the Model

This pattern — commodity input, premium service output — is not unique to technology. Three examples illustrate why the commodity label alone does not determine a business's economics.

Spotify: Spotify does not own a single song in its catalogue. It licenses music from labels, the same music available on Apple Music, YouTube Premium, Amazon Music, and a dozen other services. By the commodity logic applied to AI, Spotify should be a zero-margin business. Instead, it generated over $3 billion in net income in its most recent trailing 12-month period, with free cash flow accelerating. The moat is not the music — it is the recommendation algorithm, the playlist curation, the social features, the user interface, and 15 years of listening data that makes discovery better on Spotify than anywhere else. Users pay for the experience of accessing the commodity, not the commodity itself.

Netflix: A significant portion of Netflix's library is licensed content — shows and films available on other platforms or purchasable outright. Yet Netflix's market capitalisation reflects a durable business because its discovery engine, user profiles, content recommendation system, and consistent product experience create retention that competitors struggle to match. The packaging is the product.

Texas Roadhouse: Steak is about as close to a pure commodity as a consumable gets. There are no barriers to entry. Any restaurant can serve steak. Yet Texas Roadhouse consistently generates return on invested capital of 17–20%. For context, fully commoditised industries typically produce ROIC of 8–12%. Texas Roadhouse earns double commodity returns by selling consistency, service quality, atmosphere, and operational execution — the system around the steak, not the steak itself.

The pattern is clear: commoditised inputs can underpin premium businesses when the service layer creates genuine, defensible value.

Applying This to AI: Where the Real Moat Lives

For investors trying to understand how to invest in stocks for beginners, the AI investment thesis is best understood through this lens. The AI model itself — the LLM — is increasingly commoditised. Marcus and Eisman are correct about that. But the model is not the product that Google, Microsoft, Meta, or Amazon are ultimately selling.

What they are selling — and what generates recurring, high-margin revenue — includes:

• Cloud infrastructure and compute: Running AI workloads requires massive, globally distributed data centre networks. Building that infrastructure takes years and tens of billions of dollars. It cannot be replicated overnight.
• Enterprise integration: Microsoft's Copilot is embedded into Office 365, used by over 400 million commercial users. Replacing it requires replacing the entire productivity stack — a switching cost measured in years, not months.
• Proprietary data advantages: Google's search index, YouTube's video corpus, and Meta's social graph represent training data assets that no startup can replicate. Data quality and scale compound over time.
• Developer ecosystems: AWS, Azure, and Google Cloud each have thousands of integrated third-party tools, certifications, and enterprise relationships that create network effects independent of any single AI model.
• Trust and compliance: Regulated industries — banking, healthcare, government — will not run sensitive workloads on unproven infrastructure. The hyperscalers carry compliance certifications and audit histories that take years to build.

None of this means the AI investment thesis is risk-free. The capex buildout is real, it is large, and it will pressure near-term free cash flow across all four hyperscalers. If AI adoption plateaus or enterprise spending pulls back — as Uber's CEO recently indicated when the company burned through its annual AI budget in a single quarter — the returns on this infrastructure spend could disappoint.

But the conclusion that AI is a commodity with no moat conflates the model with the platform. The model may be commoditised. The platform is not.

What This Means for Stock Investors Right Now

For those learning how buying stocks work for beginners, the current AI market debate offers a practical masterclass in how to evaluate a business beyond surface-level narratives.

A few analytical frameworks worth applying:

1. Separate the product from the platform. When analysts argue that AI has no moat, ask: which layer are they talking about? The model layer is increasingly commoditised. The infrastructure, integration, and data layers are not.

2. Watch capex-to-revenue ratios, not capex in isolation. A company spending $60 billion in capex on $200 billion in revenue is a different business than one spending $60 billion on $50 billion in revenue. The hyperscalers have the revenue base to absorb this investment — but the ratio deserves monitoring each quarter.

3. Follow enterprise adoption metrics, not consumer buzz. Consumer AI usage is noisy and price-sensitive. Enterprise contracts, cloud backlog growth, and seat expansion in tools like Microsoft 365 Copilot are more durable indicators of whether AI is generating real economic value.

4. Assess switching costs honestly. The ease with which an individual user switches from ChatGPT to Gemini is not the same as the ease with which a Fortune 500 company migrates its entire cloud infrastructure. These are categorically different switching cost environments.

5. Monitor free cash flow conversion. As capex ramps, the companies that maintain healthy free cash flow conversion despite higher spending will demonstrate that the investment is disciplined. Those that see free cash flow collapse without corresponding revenue acceleration should attract greater scrutiny.

The Bottom Line

The commodity argument against AI is intellectually serious and partially correct — but it draws the wrong conclusion. AI models, like storage, music, steak, and licensed television content before them, can be commoditised at the product level while generating premium economics at the platform level. History shows this repeatedly.

The genuine risks — capital intensity, uncertain ROI timelines, and potential enterprise spending fatigue — are worth monitoring closely. But dismissing the entire AI investment thesis because the underlying model lacks a moat is like arguing Spotify has no business because music is free on YouTube.

The service layer is where value lives. And in AI, the service layer is owned by very few companies with very large balance sheets.

This article is for informational purposes only and does not constitute financial advice. Always consult a qualified financial professional before making investment decisions.

Frequently Asked Questions

How do stocks work for beginners in the context of AI investing?

When you buy a stock, you own a fractional share of a company's future earnings and assets. In the context of AI, this means you are buying exposure to a company's ability to monetise AI — not just the AI technology itself. The key question is not whether AI is impressive, but whether the companies building it can convert that capability into durable, growing revenue and profit. Understanding how stocks work for beginners starts with this distinction: you are investing in business models, not technologies.

Is AI actually a commodity with no competitive advantage?

Partially. The AI model layer — the large language models themselves — is becoming increasingly commoditised, with dozens of providers offering comparable performance. However, the platforms built around those models, including cloud infrastructure, enterprise software integration, proprietary data assets, and developer ecosystems, retain significant competitive advantages. The commodity argument is correct about the product but underestimates the value of the platform.

Why are tech companies spending so much on AI infrastructure?

Companies like Google, Microsoft, Meta, and Amazon are investing tens of billions annually in AI infrastructure — data centres, custom chips, and networking — because they believe AI will become the foundational layer of cloud computing. The logic is similar to why AWS invested heavily in data centre infrastructure in the early 2010s: the upfront capital cost is large, but the recurring, high-margin revenue from renting that infrastructure to millions of businesses justifies the spend. Whether that bet pays off at current spending levels is the central debate in markets.

How should beginners think about investing in stocks during volatile AI-driven markets?

Volatility is a normal feature of markets undergoing structural change, not a signal to exit. For beginners learning how to invest in stocks, the more useful discipline is separating short-term price movements — driven by earnings expectations, interest rate changes, or macro data — from long-term changes in business fundamentals. Monitoring metrics like revenue growth, free cash flow, and return on invested capital gives a clearer picture of whether a business is becoming more or less valuable over time, independent of what the stock price is doing on any given week.