Amazon Stock 14% Below Peak: Why AWS, AI Capex and Trainium Will Decide AMZN’s Outlook

Amazon Stock is having a very Wall Street problem: the business is getting stronger, but the stock is being punished for the price of that strength. After hitting a record $278.56 on May 5, AMZN fell nearly 14% in eight weeks, even as AWS posted its fastest growth in almost four years, operating margin hit a record 13.1%, and AWS backlog rose to $364 billion.

So why is Amazon stock down? Because investors are no longer asking whether Amazon can win in AI. They are asking how much it must spend to win. With 2026 capex expected near $200 billion and trailing free cash flow collapsing to $1.2 billion from $25.9 billion, the market’s concern is simple: when does this AI spending machine start throwing cash back?

Let's break down why the standard cloud growth rate comparison of AWS with Azure and Google Cloud gives investors the wrong frame, what a $20 billion custom chip business growing triple digits inside AWS actually means for Amazon stock valuation, and what the bull and bear cases on this stock look like heading into Q2 results on July 29.

Why Amazon Stock is Down Despite AWS Growth

AWS generated $37.6 billion in revenue in Q1 2026, up 28% year over year. The last time the segment grew this fast was Q1 2022, when the business was roughly half its current size. CEO Andy Jassy said on the April 29 earnings call that growing 28% on a $150 billion annualized run-rate base is "very unusual for a business of this scale."

That context matters more than the percentage. AWS's 28% growth added approximately $8.3 billion in incremental quarterly revenue. Google Cloud grew 63% in the same quarter, but from a $12.3 billion base, adding approximately $7.5 billion. Despite a growth rate more than twice as fast, Google Cloud's absolute dollar addition was smaller than AWS's. Azure grew 40% to approximately $26.3 billion. All three are supply-constrained and investing heavily.

Source: Amazon, Microsoft, Alphabet Q1 2026 earnings releases, 

The margin column is where AWS's structural advantage becomes clear. At 37.7%, AWS is generating operating profit at a rate that Google Cloud (at roughly 17%) cannot yet match, and Azure is not disclosed separately enough to compare precisely.

Yet Amazon stock fell 2-3% after-hours on April 29. Investors were focused on the other number: $44.2 billion in capital expenditure in a single quarter, up 77% year on year.

Amazon’s Hidden AI Valuation Problem

This is the same structural mispricing Indian investors saw after the HDFC-HDFC Bank merger. For nearly 18 months, the market struggled to value a housing finance business and a bank inside one stock. Amazon faces a similar problem, but across three layers. We call it The Nesting Doll Trap. Here is how the trap works. 

Currently, Amazon’s valuation problem is that investors are not looking at one business. They are looking at three businesses stacked inside each other:

• Layer 1: Amazon overall: A $181.5 billion quarterly revenue machine, still largely valued as a mega-cap commerce and cloud platform.
• Layer 2: AWS: A $150 billion annualized cloud business, growing 28%, with a 37.7% operating margin. This is where most investors stop.
• Layer 3: AI engine inside AWS: This includes Trainium AI accelerators, Graviton CPUs, Nitro networking chips, and Bedrock. Together, they show that AWS is not just renting cloud capacity; it is building its own AI infrastructure and services stack. The custom silicon business crossed a $20 billion annualized run-rate in Q1 2026, while Bedrock token volumes in Q1 exceeded all previous years combined and customer spend rose 170% quarter-on-quarter.

That is the Nesting Doll Trap. Three businesses. Three growth stages. Three multiples that would apply if each traded separately. But because they nest inside each other, the market prices all three at the multiple appropriate for the outermost layer: a large-cap cloud company.

If Amazon’s custom chip business is valued on a standalone basis, the math becomes interesting. Amazon has disclosed that Graviton, Trainium and Nitro have crossed a $20 billion annual revenue run-rate and are growing at triple-digit rates. Applying a 15x forward revenue multiple (conservative versus high-growth semiconductor peers, given triple-digit growth) would imply a standalone value of roughly $300 billion.

At present, the chip business has no dedicated line item in Amazon's P&L. It is invisible in the reported financials, which is why The Nesting Doll Trap persists.

Why OpenAI and Anthropic Matter for AWS

Here is a number that does not appear in any AWS coverage published after Q1 earnings, and it is relevant.

• OpenAI has expanded its AWS commitment to roughly $138 billion over seven years, or about $20 billion annually. 
• Anthropic has committed $100 billion over ten years for Trainium compute, or about $10 billion annually. 

Together, these two customers alone represent committed annual spend equal to nearly 20% of AWS’s current $150 billion annualized revenue run-rate. That concentration cuts both ways. 

• The risk is dependency: two capital-intensive AI companies account for a large share of committed demand.
• The signal is stronger: The world’s two most important AI labs, despite having every cloud option available, have locked multi-year infrastructure commitments into Trainium. 

Amazon Management says Trainium2 offers about 30% better price-performance than comparable GPU instances, while Trainium3 improves that by another 30–40%. In his April 2026 shareholder letter, CEO Andy Jassy said the chip business could reach a $50 billion annual revenue run-rate if Amazon sold externally to third-party data centers. 

AWS AI chief Peter DeSantis confirmed early discussions to make Trainium available outside AWS. Whether that converts in 2026 or 2027, the internal business is already material: Trainium commitments total $225 billion in backlog, and reported Trainium3 demand is largely sold out.

Source: Amazon Q1 2026 earnings release and earnings call transcript | ARR = annualized revenue run-rate.

Amazon’s Cash Flow Looks Broken

Amazon’s trailing free cash flow has fallen from $25.9 billion to just $1.2 billion, mainly because property and equipment purchases rose by $59.3 billion year-on-year. Analysts expect the pressure to continue: 

Morgan Stanley estimates negative $17 billion in 2026 free cash flow, while Bank of America expects a $28 billion deficit. But the business itself is still generating cash. Trailing operating cash flow rose 30% year-on-year to $148.5 billion. 

The issue is that Amazon is taking that cash and more, and putting it into roughly $200 billion of capex, mostly for data centers, servers, and AI infrastructure. That distinction matters.

Jassy’s argument is that Amazon has done this before. The original AWS data center build and the 2013–2018 fulfilment expansion both suppressed free cash flow during the investment phase, then produced years of margin and cash flow expansion once capacity was monetized.

The bear case is not that this playbook never worked. It is that AI infrastructure may be a different beast: larger, more expensive, more competitive, and potentially slower to monetize. That is why Amazon’s stock is being punished despite strong operating numbers. Investors are not questioning whether Amazon can spend. They are questioning how quickly that spending turns back into cash.

Analysts Are Still Bullish on AMZN Stock

The analyst consensus on Amazon's direction is unusually unified for a stock at this valuation level. According to S&P Global data updated as of June 26, 2026, 67 analysts covering AMZN have a consensus "Strong Buy" rating with an average 12-month price target of approximately $313, implying roughly 30% upside from current levels.

Source: TheStreet, Benzinga, MarketBeat and S&P Global

Key Risks for Amazon Stock Through 2027

Risk 1: No capex payback timeline - Amazon may spend ~$200B on capex in 2026, but has not given a clear FCF recovery date or capex ceiling. If AWS growth slows while quarterly capex stays near $44–50B, the cash-flow story weakens. Debt is already up from $65.6B to $119.1B YoY.

Risk 2: EPS is noisy because of Anthropic  Q1 net income of $30.3B included $16.8B in Anthropic gains. That makes headline EPS less reliable. For Q2, watch operating income guidance of $20–24B, not EPS.

Risk 3: AWS backlog is concentrated - OpenAI and Anthropic represent about $30B in annual AWS commitments, equal to ~20% of AWS’s $150B ARR. Strong demand signal, but also a concentration risk if either customer slows spending.

Is Amazon Stock Cheap After the Pullback?

Amazon trades at approximately 28x forward earnings, which is not cheap by any traditional measure. But the question worth asking is: cheap relative to what? Goldman Sachs values AWS alone at 9x forward revenue. At $150 billion ARR, that implies roughly $1.35 trillion for AWS in isolation.

Amazon stock is cheap on a sum-of-the-parts basis, but not cheap on free cash flow today. That is the tension.

Together, these pieces imply roughly $2.25–2.5 trillion of value versus Amazon’s current market cap of about $2.56 trillion. That means Amazon is not obviously “cheap” at the parent level, but the market may be assigning very little standalone value to its AI chip and Bedrock layers.

Hence, the best cloud bet question has a more nuanced answer. If you want the highest growth rate, Azure at 40% and Google Cloud at 63% are both running faster. If you want the best operating economics, AWS at 37.7% operating margin is materially ahead of Google Cloud's 17%. What Amazon offers is the most asset-heavy, most vertically integrated cloud infrastructure play with $225 billion in signed commitments, where the real upside is in the two businesses sitting beneath AWS that the current price largely ignores. 

Whether that is worth the FCF wait is ultimately a question of how long you are willing to hold, not whether the underlying businesses are strong.

AMZN Outlook: Bull & Bear Case, Investor Watchlist

Amazon is not a cloud company trading at a cloud company multiple. It is three businesses, all reported as a single segment. The market's inability to price each layer separately creates a structural discount that is larger than the 14% gap from the all-time high alone suggests.

Amazon Bull case: AWS growth remains above 25% through 2026 as enterprise AI workloads move from pilot to production. The $364 billion signed backlog plus Anthropic's separate $100 billion deal converts to revenue over the next 2-3 years at margins that validate the capex. The chip business begins generating external revenue by 2027-2028, creating a new earnings line the market can price at semiconductor multiples. The Nesting Doll Trap gradually closes as Amazon either reports silicon as a separate segment or external chip sales provide a standalone pricing anchor.

Bear case: Free cash flow stays negative through 2027. Long-term debt climbs past $150 billion, requiring equity or debt issuance that dilutes the share count. AWS growth decelerates to the low-20s as Azure's GPT-5 enterprise integration proves stickier than expected or Google Cloud's TPU cost advantage captures incremental AI training workloads. The Anthropic mark reverses in one or more quarters, creating headline EPS misses that obscure operational performance.

Amazon reports Q2 earnings on July 29. The two numbers that matter most are AWS growth rate and Q2 operating income versus the $22 billion midpoint of guidance. A beat on operating income while capex stays manageable would confirm that the margin expansion from Q1 was structural, not seasonal.