Amazon Has a $25B Chips Business. Wall Street Noticed.

The volatility crush arrived as expected. Amazon had priced a roughly 6.9% move into Thursday’s close. By Friday morning, the stock was up sharply in premarket trading. The straddle sellers collected. The directional call buyers collected more. But the real signal was not in the move itself. It was in what the results disclosed about a business line the options market had never properly valued.

The Signal

CEO Andy Jassy said AWS is “booming,” noting it was growing 36.7% year over year in Q2, its fastest growth in 18 quarters, and that Amazon’s AI and chips businesses each eclipsed run rates of more than $25 billion. That sentence carried a detail the market had been underweighting for months. Two separate $25 billion-plus run-rate businesses are embedded inside what Wall Street still mostly prices as a cloud and retail company.

Amazon has increasingly highlighted its in-house chips division, which includes the Trainium and Graviton brands, as a newer growth pillar for the company. Before Thursday’s report, that framing felt like corporate marketing. After Thursday’s numbers, it reads differently. Amazon has previously said that if its chips business were treated as a stand-alone vendor selling externally, its annual run rate would be about $50 billion. The options market has not yet priced that business the way it prices Nvidia or Broadcom, two names where the semiconductor multiple dominates the valuation.

Why It Matters

Ahead of earnings, options implied a roughly plus-or-minus 7% post-earnings move. The call buyers were speculating on the binary event. The more interesting positioning was what those strikes reveal about where the consensus ceiling was sitting before the report. The stock blew past the upper implied range. That kind of gap between the options-priced move and the actual move is exactly what happens when a catalyst contains information the market had not fully incorporated.

Amazon has cleared its implied move more often than not, but the precise hit rate varies by data source and lookback window. Over its last 15 earnings reports, the average earnings-day move has been around the mid-single digits to high-single digits. A double-digit overnight gap eclipses that range, and points to something structural rather than cyclical. The AI and chips run-rate disclosure was the structural piece.

The Company Behind the Signal

Amazon reported revenue of $200.6 billion for Q2 2026, up 20% year over year, with operating income of $27.5 billion, up 43%. AWS revenue growth accelerated for the fifth straight quarter, reaching 36.7% year over year. Amazon also pointed to strong contracted demand in its AWS backlog, which it said is growing triple digits year over year. That backlog deserves attention. A contracted revenue figure of that scale is not a cloud business in the conventional sense. It is a constrained infrastructure business with more demand than it can currently serve.

On the call, Jassy said Amazon now expects to spend approximately $220 billion in 2026 cash capital expenditures, up from its prior estimate of about $200 billion, and that even at that level it still will not have enough capacity to meet AI demand in 2026. He said he believes that dynamic will also be true in 2027, and added that demand for 2028 is “striking.” That is the framework a semiconductor analyst uses to describe a supply-constrained cycle, not the language of a cloud business managing churn and pricing pressure.

The customer anchor helps explain why the cycle is real. Anthropic has publicly described an agreement with Amazon that secures up to 5 gigawatts of capacity for training and deploying Claude, including new Trainium capacity. On the CPU side, Graviton5 is now in general availability, and Amazon says it delivers up to 25% better compute performance than Graviton4. Amazon also says Graviton is used by 98% of the top 1,000 EC2 customers.

The free cash flow picture creates the one legitimate tension. Amazon said trailing-12-month free cash flow swung from an $18.2 billion inflow to a $7.6 billion outflow. For a stock pricing a semiconductor franchise into the multiple, the capex bill matters as much as the revenue trajectory.

Market Expectations

Before Thursday’s report, the options market had priced a fairly symmetrical event. The July 31 expiration straddle implied a move of around 6.9%, creating an expected post-earnings range that traders typically frame as a roughly plus-or-minus band around spot. The stock’s opening trajectory Friday morning broke through the upper end of that range. That is not a coincidence. The earnings-day straddle was pricing the AWS beat the Street anticipated. It was not pricing the AI and chips run-rate disclosure, because those run rates, each crossing $25 billion, were not in any model that the consensus had built.

Now the question shifts. With implied volatility crushed after the event, near-term options on AMZN will be cheaper than they were 24 hours ago. The forward implied move reflects an earnings event that has already resolved. What the market has not yet resolved is whether a chips franchise embedded inside AWS should carry a different multiple than the rest of the business. That debate now begins.

Strategic Considerations

The volatility crush creates a condition worth understanding. IV built into Thursday’s expiration has now collapsed. Longer-dated options, particularly in the 30-to-60-day range, are likely pricing lower as realized volatility post-earnings tends to moderate. For readers watching the Trainium thesis, the tool that fits is a longer-dated call spread rather than an outright long call, for two reasons.

First, the $220 billion capex commitment means free cash flow pressure continues into 2027, and that limits how aggressively most valuation frameworks will expand the multiple near-term. A call spread structures the upside without requiring an aggressive re-rating in a compressed timeframe. Second, Amazon has said it is actively having conversations with customers interested in purchasing Trainium chips separately from the cloud, and that it could pursue that in the future. If Amazon begins selling Trainium directly, the way Nvidia sells GPUs, that is the catalyst that forces a semiconductor-style re-rate. That is not a Q3 story. It is a 2027 story, which argues for longer duration.

The primary risk is capital allocation. Amazon has indicated cash capex is rising materially year over year. If memory prices continue rising, as management said they contributed to the capex increase from roughly $200 billion to roughly $220 billion this year, AWS margins become harder to defend even as revenues accelerate. A margin miss on AWS in Q3 would be the fastest way to reverse the re-rating the market is starting to price.

The secondary risk is concentration. Anthropic has publicly discussed multi-year commitments tied to Trainium capacity. If anchor customers shift compute spend toward alternative silicon, the revenue commitments may be more durable than the growth trajectory they imply.

What to Watch

Three developments will either validate or challenge the Trainium re-rating thesis over the next 60 days. The first is any announcement of direct external chip sales. That is the structural catalyst that separates a $25 billion cloud-embedded business from a standalone semiconductor franchise. The second is Q3 AWS margin guidance. In Q2, third-party summaries of the release and call put AWS operating margin at about 39%, up roughly 650 basis points year over year. If that expansion continues into Q3, the argument for a higher multiple gets stronger. If capex absorption compresses it, the thesis pauses. The third is how the AWS backlog converts to quarterly revenue. Contracted demand is not the same as recognized revenue, and the pace of data center buildout determines when that backlog starts flowing through the income statement at scale.

The options market resolved its binary question Thursday night. The more interesting question, whether Amazon’s silicon business belongs in a semiconductor valuation conversation, is only now beginning to get asked.