Here is the feature article for your publication.
*
**Amazon's Paper Profits from Anthropic Hide a Much More Expensive Reality**
At first glance, Amazon’s third-quarter earnings report looked like a blockbuster. Profits surged 38% to $21.2 billion, a figure that sends a clear signal of dominance. But peel back a single layer of the financial statements, and the story changes completely. The engine driving that headline number wasn't a boom in e-commerce or a surge in its cloud business. It was an accounting entry, one that led to headlines declaring that Amazon’s Anthropic investment boosts its quarterly profits by $9.5B.
A staggering $9.5 billion pre-tax gain materialized on the income statement, courtesy of Amazon’s stake in the AI startup Anthropic. This wasn't cash. No assets were sold. This was the financial equivalent of your home’s Zillow estimate going up. After Anthropic secured new funding at a stratospheric $183 billion valuation, accounting rules (specifically, a process known as a 'mark-to-market' adjustment) required Amazon to write up the value of its investment. You’re richer on paper, but your bank account balance hasn’t changed.
To put that $9.5 billion paper gain in context, Amazon Web Services—the company’s historical profit machine and the undisputed engine of its growth for a decade—generated $11.4 billion in actual operating profit during the same period. A non-cash, non-operating gain from a venture investment nearly equaled the quarterly profit of one of the most successful business divisions in modern history. I've reviewed countless quarterly reports, and seeing a non-operating, non-cash gain presented so prominently feels like a deliberate piece of financial theater designed to distract from a much more complicated reality.
The Anatomy of a Paper Gain
Let’s be clear: holding a valuable stake in a company like Anthropic is a strategic coup. Amazon has committed up to $8 billion to the AI firm, positioning itself as a key partner in the race against Microsoft’s OpenAI and Google’s own AI ventures. This is a smart, defensive, and potentially lucrative long-term play. The problem isn't the investment itself; it’s the narrative the resulting paper gain allows the company to spin.
While Wall Street might fixate on the shiny, valuation-driven profit number, the operational metrics tell a story of immense, and growing, pressure. This is where the real game is being played, not in the footnotes of an investment portfolio but in the concrete and silicon of new data centers and the salaries of a new kind of engineer.
The AI arms race is phenomenally expensive, and Amazon is paying the price of admission in cash. The company just lit up "Project Rainier," an $11 billion data center complex built to run Anthropic’s models. Imagine the hum of hundreds of thousands of Amazon's proprietary Trainium 2 chips, all consuming immense power, all representing a massive capital outlay. That single project’s cost is greater than the entire paper gain from the Anthropic stake. And it’s just one piece of a much larger spending spree. Amazon’s total capital expenditures hit $35.1 billion in the third quarter alone, a 55% increase from the year prior.
This spending is already eating into the profitability of Amazon’s crown jewel. While AWS sales rose a healthy 20% to $33 billion, its operating income grew by less than half that—just 9.6%, to be exact. That discrepancy between revenue growth and profit growth is the key data point. It’s a bright red flag signaling significant margin compression. The cost of building the future of AI is hollowing out the near-term profitability of the cloud business that is supposed to pay for it. The most telling number of all? Amazon’s free cash flow, the lifeblood of any company, plummeted 69% over the past year.
The Cost of the Real Game
The infrastructure buildout is only half of the expense equation. The other half is human capital—and not just any kind. AI companies are in a frantic hiring race for what some are calling The new hot job in AI: forward-deployed engineers.
Pioneered by the secretive data intelligence firm Palantir (where they were famously sent to military bases in Iraq and Afghanistan), FDEs are a hybrid of elite coder and high-touch consultant. They don't just build software; they embed within a customer’s organization to understand their unique problems and then customize AI models on-site. They are the human bridge between a general-purpose AI model and a specific, revenue-generating business application.
This is "product discovery from the inside," as one Palantir executive put it. It’s about ensuring the fantastically expensive AI models actually do something useful enough for a Fortune 500 company to pay for. Anthropic announced it would grow its team of FDEs and other applied AI specialists fivefold this year. OpenAI expects to have about 50 in its European division alone in 2025. This isn't a cheap or scalable talent pool. These are highly specialized, highly compensated engineers who represent a massive and ongoing operational expense.
This two-pronged spending—billions on silicon, millions on salaries—is the real story of Amazon's AI strategy. The company is placing a monumental bet that by providing the infrastructure (AWS) and the partner (Anthropic) with its embedded FDEs, it can lock in the next generation of enterprise AI customers. CEO Andy Jassy says they are monetizing the capacity as fast as they build it, but the numbers suggest otherwise. The margins are shrinking, and the cash is burning. So the unavoidable question becomes: what is the true return on investment for this strategy, and how long can Amazon sustain this level of cash burn before investors start ignoring the paper gains and focusing on the collapsing free cash flow?
The Real Price Is Measured in Margins, Not Markups
Let's cut through the noise. The market is celebrating a phantom profit generated by venture capital markups while systematically ignoring the operational reality. Amazon is engaged in a brutal, capital-intensive war with Microsoft and Google, and it is costing them dearly in the metrics that actually matter. The $9.5 billion gain from Anthropic is a distraction. The real story is the 55% surge in capital spending, the 69% drop in free cash flow, and the compressing margins at AWS. That is the true price of entry into the AI big leagues, and it’s a bill that can't be paid with paper.