Amazon is usually described as the world's biggest online store. That description misses most of what drives the company today. Its cloud business, AWS, earns most of the operating profit. Its advertising business may earn more than all of its retail. A bet of $18 billion on Anthropic is now worth about $339 billion on paper. And more than one million robots run its warehouses, making it what we call the secret robotics giant. At the 19 November 2026 close of $257.37 the stock was 9.4% below its August record, held back by the largest spending plan in its history.

-9.4%From the $284.02 record close of 3 August 2026 to $257.37 on 19 November
-$42.4BFree cash flow over the twelve months to September 2026
$339BValue of the Anthropic stake at its first-day close, on $18 billion invested
1M+Robots in more than 300 Amazon buildings

The backdrop is a market in the red. The neoclouds crashed as their debt financing dried up, the Democrats' midterm sweep raised fears about data centres, and Nvidia's chief executive Jensen Huang resigned. Each event hit the same nerve: the AI build-out suddenly looks very risky. Yet the numbers still look very good. Part of the demand is circular, with companies that profit from the build-out investing in their own customers, who then use the money to buy their products. But the demand is also real, and Amazon, through AWS the largest cloud and compute provider in the world, is the best place to test it.

This article follows the argument of the video: the midterm crash and the valuation, the Anthropic windfall, the businesses, AWS, the capex boom, the robots and how Amazon builds them, its blind spots, the efficiency machine, the fight against AI shopping agents, the competition at home and abroad, its deals, its strategy and the outlook. Market figures are anchored to the 19 November 2026 close of $257.37 unless another date is given; financial figures run to the third quarter of 2026.

What the valuation says now

At $257.37 Amazon trades at 20.8 times its trailing earnings per share of $12.37, cheaper than Microsoft at 26.0 and more expensive than Alphabet at 14.2. The number is misleading. About 40% of Amazon's trailing EPS comes from paper gains on its Anthropic stake. Without them, trailing EPS is $7.40 and the P/E is 34.8, 67% higher than reported and 34% above Microsoft. The same caution applies to Alphabet, whose reported earnings also include gains on Anthropic and SpaceX.[1][2]

The Midterm and Neocloud Crash

The Democrats swept the US midterm elections on 3 November 2026. The next day the market rallied: Amazon rose 6.1%, almost four times the S&P 500's 1.6%. The crash came on 5 November. The S&P 500 fell 4.4%, its worst day since April 2025, and the Nasdaq 100 fell 6.4%, its worst day since 2020. Amazon fell only 2.2%, about a third of the Nasdaq 100's drop. The neoclouds and chipmakers took the hit instead: Nvidia fell 9.3% and CoreWeave 9.8% that day.[1][4]

The bet behind the move, as Reuters and CNBC reported it, was that curbs on new data centre construction would protect the companies that already own capacity. Texas governor-elect Gina Hinojosa pledged an immediate moratorium on data centre approvals. Over the two weeks from 3 to 17 November Amazon rose 2.9% while the S&P 500 lost 3.0%. Microsoft lost 2.8%, Meta 0.4% and Alphabet was flat at plus 0.1%. Part of Amazon's edge was its own news: it rose 4.8% on the morning of 9 November, when JPMorgan put it on its Analyst Focus List.[4][1]

Investors are used to negative cash flow

Amazon's cash flow is worse than its peers'. Its free cash flow over the twelve months to September 2026 was minus $42.4 billion, while Alphabet and Meta stayed positive. Since 2021 Amazon had negative free cash flow in 11 of 23 quarters. From 2024 on, Amazon had four negative quarters, Alphabet only its last two, Meta only its last one and Microsoft none.[2][6]

Yet cash flow did not decide who fell most. Microsoft, positive in every quarter since 2024, fell 3.6% on 5 November, 1.6 times Amazon's drop. Meta, negative only in its latest quarter, fell 4.0%. Our reading is that investors are used to Amazon spending ahead of its cash, which makes the stock unusually resilient to bad cash flow news.

Part of the reason is the kind of business Amazon is. Microsoft, Alphabet and Meta are mostly software companies. Amazon is a big tech company with a very strong physical side: warehouses, deliveries, a very large staff and many locations, all of which need heavy capital spending. For a company like that, negative cash flow in a building phase is not a shock.

Windfall: The Anthropic Investment Drives Profit

Amazon has invested $18 billion in Anthropic since 2023. On 30 September 2026 it carried the stake at $238.7 billion, 13 times what it paid. From January 2025 to September 2026 the stake produced $80.6 billion of pre-tax gains, more than Amazon's whole 2025 operating income of $80.0 billion. The second quarter of 2026 alone brought $50.5 billion, after Anthropic's Series H round, 1.8 times Amazon's third-quarter operating income of $28.1 billion.[2]

Anthropic went public this week. It priced its IPO at $83 a share on 18 November, above the $70 to $78 range, and closed its first trading day at $97.17, up 17%, after opening at $111.50. At that close, Amazon's 3.48 billion shares, held plus still issuable from its notes, were worth about $339 billion, 19 times the money it put in and 42% above the September carrying value. Amazon's own stock fell 3.9% on the same day, from $267.74 to $257.37. How much of the rise reaches Amazon's fourth-quarter profit depends on accounting it has not yet disclosed.[5][1]

Amazon backed the other leading lab too: it bought $50 billion of OpenAI preferred stock in 2026, so $68 billion of cash sits in the two labs. Both are also among AWS's largest customers. Anthropic committed at least $110 billion to AWS over ten years, and OpenAI added $100 billion over eight years to an earlier $38 billion deal.[2][3]

Ownership

Amazon does not disclose what percentage of Anthropic it owns. Its notes can only convert while its fully diluted stake stays at or below a 12% cap, which Amazon chose not to waive for the IPO. Press estimates of its share differ and are not used here.

Business Units

Amazon had $716.9 billion of revenue in 2025, passing Walmart's $713.2 billion for the first time. Online stores, the goods and digital content Amazon sells itself, brought in $269.3 billion (38%). Third-party seller services, the fees from outside sellers, brought in $172.2 billion (24%), AWS $128.7 billion (18%), advertising $68.6 billion (9.6%), subscriptions such as Prime $49.6 billion (7%) and physical stores, mostly Whole Foods, $22.6 billion (3%). The Prime fee has stayed at $139 since February 2022.[2]

Amazon increasingly earns by charging others. Seller fees plus ads made up 33.6% of revenue in 2025, up 5 points from 28.6% in 2021, and advertising grew 120% in that time, from $31.2 billion to $68.6 billion.[2]

Data centres and AI are becoming a large part of Microsoft and Google, but Amazon is still mostly a retail and commerce company, and that is what people think it is. Google, known as a search company, faces a weakening search business. Amazon's core is intact, which is one reason investors see it as more robust and panic less when its cash flow turns negative or its data centre spending rises: AWS is a very good business, but a much smaller share of revenue, and the retail business is a very stable day job next to the build-out.

AWS: The Gift That Keeps On Making Profit

AWS is the gift that keeps on giving. When Amazon created it, it was the first to run that kind of business at scale and had almost no competition for years, which gave it the cash and the infrastructure to stay ahead. AWS earned $45.6 billion of operating income in 2025, 57% of Amazon's $80.0 billion total. In the third quarter of 2026 it was 63.3% of operating income on 22.8% of revenue. Its growth has accelerated for six quarters in a row, from a low of 12% in mid-2023 to 39.3% in the third quarter of 2026, when revenue reached $46.0 billion. The annual run rate is $184 billion, more than double the roughly $82 billion Amazon quoted in late 2022.[2][3]

The backlog of signed contracts not yet earned reached $578 billion, 3.6 times the last twelve months of AWS revenue. Amazon names only two customers in it. OpenAI committed $38 billion in November 2025 and $100 billion more over eight years in February 2026. Anthropic committed more than $100 billion over ten years in April 2026. If both deals are booked in full, at least $134 billion of the backlog's rise came from other customers, none named by Amazon.[2][3]

AWS is the incumbent and the largest cloud provider, with a target on its back from Google and Microsoft. Its share of cloud infrastructure spending fell 3 points in a year, from 29% to 26%, while Google rose from 13% to 16% and Microsoft slipped from 20% to 19%, according to Synergy Research. In the third quarter Azure grew 48% and Google Cloud 103%, both faster than AWS, which has grown the slowest of the three in every quarter since early 2023. Andy Jassy promises AWS growth of more than 40% in both 2027 and 2028.[7][3]

Aggressive Data Center Capex

Amazon guides 2026 capex to about $220 billion, 4.6 times the $48.1 billion of 2023. The third quarter alone took $75.9 billion of net capex, 37.6% of sales, and gross capex of $77.3 billion was 32% above the $58.7 billion analysts expected. The stock fell 5% to 7% after hours on the report, though it closed only 1.4% lower the next day. In the first nine months of 2026, 74.8% of Amazon's new property went to AWS, up 44 points from 30.5% in 2021. For 2027 Amazon promises "meaningfully more"; S&P Global projects $319 billion, 45% more than 2026.[2][3]

The money is borrowed as well as earned. Long-term debt rose 163% in nine months, from $68.8 billion at the end of 2025 to $180.9 billion in September 2026, after about $115 billion of bonds and loans in dollars, euros, Swiss francs, Canadian dollars and pounds. Interest expense was $1.69 billion in the third quarter, 3.1 times a year earlier. Jassy's answer to the doubters: "We're not investing approximately $200 billion in capex in 2026 on a hunch."[2][3]

Like its peers, Amazon has bought back no stock since 2022 and its share count has risen 7.5% since the end of 2020: negative cash flow has to be financed somehow, with debt or with new shares.[2]

That spending meets a political climate turning against it. An anti-AI sentiment is getting louder on both sides of the political spectrum, and if both sides agree, curbs on data centres become likely. "Data centre" has become a toxic word for something that has existed for decades. The next test comes with Amazon's fourth-quarter report, expected in February 2027: if the market crashes further, regulation tightens and Huang does not return to Nvidia, the backlogs and the capex could slow. Our guess is that they will not, because what happened in the market is not about the technology or the demand. Demand is very strong, the infrastructure is heavily used, and AI already automates so much knowledge work that it will need ever more data centres. The real risk is financial: neoclouds that build data centres entirely on debt, large companies handing out guarantees and revenue booked before it is received. If that is rooted out, the market will turn very red, but the hyperscalers, Amazon among them, have the cloud businesses to come through it.

The Secret Robotics Company

Amazon runs more than one million robots in more than 300 buildings, roughly three times the 350,000 Morgan Stanley estimated for 2021. It calls itself "the world's largest manufacturer and operator of mobile robotics", and we found no company with a larger disclosed fleet: JD.com's 3 million is a five-year purchase plan. Yet Amazon reports no robotics segment and no robot savings.[8]

Most of the fleet are drives that carry shelves and packages. On top come arms: Sparrow and Robin pick and sort, Cardinal lifts heavy packages and Vulcan can feel what it touches. The arm fleet doubled in 2026 from an undisclosed base. A generative AI model called DeepFleet directs the traffic of the whole fleet; Amazon claims it makes robot travel 10% more efficient, but has published no before-and-after data.[8]

That coordination is an expertise of its own. Most of the attention in robotics goes to single machines, a humanoid, Tesla's Optimus or a self-driving car, and how well each one behaves. Orchestrating a fleet of a million robots is a different problem: coordinating a group, not an individual, and it produces a great deal of data that can train better coordination models.

Outside the warehouse the record is weaker. Prime Air made "hundreds of thousands" of drone deliveries by August 2026, far below an internal target of one million for the year. In Richardson, Texas it flies about 150 orders a day, 15% of the 1,000 its approval allows. Drones crashed in Texas and England in 2026, and Amazon dropped its drone plans for Italy in December 2025.[8]

That slow progress makes sense. The AI a drone needs is far simpler than a humanoid's, but every delivery meets edge cases: wind, bad weather, birds, people trying to catch drones with nets, or a package that does not land in the right spot. Road transport is already very efficient without those risks, so drones matter little to Amazon's business and could even make delivery more expensive.

The data advantage is real but narrow. Amazon says it has "literally billions of hours of robot navigation data". Physical Intelligence trained its robot model on about 10,000 hours and Figure on about 500. Those numbers compare unlike data: Amazon's is mostly robots driving around warehouses, while the rivals collect hand skills. Amazon does not sell or share its robot data.[8]

Supply Chain Wizard, Not Technology Leader

Amazon was never known as a manufacturer. It runs warehouses, packs and ships, and excels at retail, but production lines are the main business of companies like Tesla, not of Amazon. Yet Amazon does build its robots. It designs and assembles its core robots at a 350,000-square-foot plant in Westborough, Massachusetts, announced in 2019 as a $40 million investment and opened in 2021. By late 2022 its plants in North Reading and Westborough had built more than 500,000 robots, according to Amazon Robotics executive Scott Dresser, about half of today's fleet. Its lines can build up to 1,000 robots a day, and two new plants are coming: a "multibillion-dollar" plant in Austin, Texas, announced in August 2026 without an exact amount, and a $100 million-plus plant in Greenwood, Indiana, running by 2028.[9]

The rest comes from suppliers and partners. Amazon builds "working with local suppliers" that it does not name. The Sparrow arm is an off-the-shelf Fanuc arm with Amazon's own gripper and AI. It bought only five Digit humanoids from Agility, for $1.05 million. The frontier is bought or licensed: Kiva in 2012, Covariant's robot models in 2024, Fauna and Rivr in 2026, stakes in Agility and Skild AI, and Nvidia's robot platform.[9][8]

This is the supply chain at work. Tesla tries to own its value chain and buy as little as possible from others to protect its margins. Amazon does not care whether it makes a part itself or buys a perfectly good robotic arm at scale from a strong supplier, as long as the efficiency and the return on investment are there.

The China dependency

The supply chain behind the store leans on China. Morgan Stanley estimates that about 25% of the cost of the goods Amazon sells itself comes from China, two and a half times the 10% average of the online retailers it tracks. Raymond James estimates that 30% of Amazon's sales and 15% of its ads come from China. Among the top 10,000 sellers on Amazon.com, the Chinese share rose 13.4 points, from 42.5% in 2020 to 55.9% in 2026, according to Marketplace Pulse, although US sellers still made 65.3% of that group's sales. 60% of new US sellers in 2025 were Chinese. Since February 2024 Amazon's own filings say China-based sellers and suppliers account for "significant portions" of its business.[11][10][2]

The Blind Spot: Most Robots, but Low-End

Most of the fleet are simple drives from the Kiva line. Vulcan, Amazon's best arm, handles about 75% of item types, so a quarter still needs people. Two ambitious projects were dropped: the Scout delivery robot in 2022 and the Blue Jay multi-arm system in 2026. Amazon has released no general robot foundation model; DeepFleet only directs fleet traffic.[8]

That is the price of sourcing everything: Amazon is the leader in applied robotics, robots that do real work, but it builds little expertise in the most advanced robots and their intelligence. The blind spot is its data. Data is becoming a moat for companies, and Amazon has a great deal of it, just as Uber has driving data it does not use for its own self-driving models. Amazon would be a natural leader in AI models for robots in warehouses and factories. A busy warehouse is confined, its parameters are known and cameras are everywhere, yet a robot that can walk through it would have to make countless small decisions from visual input. It is also very quiet about how useful its robots really are.

Google is further ahead in robot intelligence. Google DeepMind's Gemini Robotics 2 controls a humanoid's whole body, and its 2.5 version reportedly reached 96% success sorting car parts in a Mercedes-Benz trial, a proof of concept rather than a rollout. Amazon's own AI models lag as well: it began winding down its top Nova models in July 2026, and its AI chief Peter DeSantis said they "haven't been at the very frontier".[14][3]

In the long run it would be a mistake to sleep on robots. They will be one of the next technology waves, even if it takes ten years, alongside others such as 3D printing that can make almost anything, which may be much further away.

Efficiency Machine: Pragmatic Automation to Cut Costs

The stores business got much leaner. The operating margin of the two store segments rose 5.2 points, from 1.6% in 2021 to 6.8% over the twelve months to September 2026. Fulfillment costs fell 1.2 points of sales, from 16.0% to 14.8%, worth about $9.6 billion a year. Amazon says its cost to serve each unit fell four years in a row, though it has given no number since 2023. Its newest systems store items up to 75% faster (Sequoia) and process orders up to 25% faster (Shreveport), by Amazon's own count.[2][3]

Headcount barely moved while the robot fleet tripled. Amazon has 1,638,000 employees, a record but only 2% more than the 1,608,000 of 2021. Nvidia, which has been worth more than Amazon, employs about 42,000. Revenue per employee in the stores business rose 53%, from $253,000 to $388,000, and operating income per employee rose 579%, from $3,900 to $26,500. Amazon gives no split of its staff; GeekWire estimates that about 1.2 million of 1.5 million work in warehouses, where robots bring shelves to people and arms take the heavy and high work. The New York Times reported plans to avoid 600,000 US hires by 2033; Amazon calls its people and robots "one team".[2][8]

Amazon did this the smart way. Rather than firing people and saying robots now do the work, it keeps hiring, just more conservatively, and lets robots take on part of the work. It never says how much of the efficiency comes from robots, which avoids the fear around layoffs.

What the numbers cannot prove

Ads, which need almost no warehouse staff, and the 2023 redesign of the US network into eight regions also lifted these margins. Amazon discloses no robot savings, so the robots' share of the gain cannot be isolated.

Defending Against Agents

E-commerce is Amazon's main business, and every marketplace is now threatened by AI agents that shop for the user. ChatGPT, Google, Perplexity, Meta's Muse and Microsoft's Copilot can all buy on a shopper's behalf, and an agent never sees a sponsored listing. That puts Amazon's ads at risk: J.P. Morgan expects about $83 billion of ad revenue in 2026, roughly a third of Amazon's operating income. Adobe counted a 693% rise in AI traffic to US retail sites over the 2025 holidays. So far the threat is small: agents bring only about 0.4% of Amazon's traffic, according to J.P. Morgan.[13]

Amazon's answer is a wall. Its robots.txt file blocks 99 AI crawlers and agents, including ChatGPT, Google, Perplexity and Claude. It blocked Meta's Muse from 20 September 2026. It sued Perplexity over its Comet browser in November 2025 and won an injunction in March 2026, but an appeals court threw it out in August, ruling that the user, not Perplexity, accessed Amazon. At the same time Amazon pushes its own agents: Rufus has more than 300 million users and adds nearly $12 billion of sales a year, and Alexa for Shopping more than 400 million users, by Amazon's count. Asked on the October call whether outside agents hurt Amazon's ads, Jassy said: "The answer is no."[13][3]

Suing will not hold the line. Music and film companies went after piracy tools, then after individuals, and failed because nobody can sue everyone. Agents are tools that people use, and anyone can already let Claude or Codex operate their browser. Of all the big platforms, Amazon is still in the best position: it runs a marketplace with a very strong grip on it, and Microsoft's operating system looks more exposed to disruption than Amazon's store. With agents at 0.4% of its traffic the threat is small today, but in about two years it may be the central question: a headless Amazon, where an agent, not a person, does everything from finding the product to the purchase. That would disrupt every marketplace, Uber included.

Competitive Frenzy

Few companies fight on as many fronts. In e-commerce Amazon faces Walmart, Temu, Shein and strong local leaders abroad; its Temu rival Haul had about $2 billion of US sales against about $30 billion for Temu, 15 times smaller, according to Marketplace Pulse. In cloud it faces Microsoft and Google. In retail, Walmart: without AWS, Amazon's $588 billion of 2025 sales were more than $100 billion short of Walmart's $713 billion, and Walmart now sells inside ChatGPT and Gemini, while Amazon is in neither.[10][2][13]

In ads it faces Google, Meta and Walmart. Amazon's ads equalled 31% of Google Search's ad revenue and 31% of Meta's in the third quarter of 2026. Walmart's ads were about $6.4 billion, a tenth of Amazon's, but grew 46% in a year. In streaming, Prime Video and MGM face Netflix, which also sells ads through Amazon. In chips, Amazon's own Trainium, with more than $225 billion of commitments, competes with Nvidia, while Amazon still buys 2 million more Nvidia GPUs. Amazon Now offers quick delivery in more than 450 cities, Leo satellites take on Starlink and Zoox robotaxis take on Waymo. Even its chips come from an acquisition: Annapurna Labs, bought in 2015, designs Trainium and Graviton, so far used only inside AWS.[2][3][15]

These are not direct and indirect competitors in one industry but very different markets, and in several of them, such as streaming, where Prime Video is also a producer of expensive, high-end content, and satellites, Amazon is not the top contender.

Local E-Commerce Wars

Amazon runs 23 country stores and offers Prime in 27 countries. It opened eight new countries in five years: the Netherlands, Saudi Arabia, Sweden, Poland, Egypt, Belgium, South Africa and Ireland. Where a strong local marketplace already existed, Amazon arrived late and stays small. In Poland it has about 6% after four years, while Allegro dominates, according to the OECD, and Allegro even raised its seller fees from 10.2% to 12.3% while Amazon competed. In the Netherlands bol sells about €3.1 billion, three times Amazon's estimated €1 billion. In South Africa Amazon has 1.6% of tracked online spending, against 22.5% for Takealot, 14 times as much.[12]

In India Flipkart had about 48% of e-commerce in fiscal 2023, according to Bernstein, and both it and Amazon's Indian marketplace still lose money. In Latin America MercadoLibre grew revenue 39% to $29 billion in 2025. Where Amazon is strong, it is very strong: in Germany it has about 60% of online sales, according to the German competition authority, and Amazon says the UK, Japan and France are all profitable.[12]

Acquisitions and Divestments

Amazon's deals follow its bets. In robotics and AI it paid a $94 million fee to call off iRobot, the Roomba maker, in January 2024, then chose licence-and-hire deals over takeovers with Adept and Covariant in 2024, and bought Fauna (the Sprout humanoid) and Rivr (delivery robots) in 2026. In health it bought One Medical's clinics for $3.5 billion in 2023 and closed Amazon Care and the Halo wristband. In media it paid $8.6 billion for MGM in 2022. In satellites a $10.9 billion deal for Globalstar is pending, closing in 2027. In retail it closed Kindle in China, Fresh in the UK and, in January 2026, its US Go and Fresh stores.[2]

The most painful bet was Rivian. Amazon put about $2 billion into the electric van maker from 2019 to 2021 and never sold. At the end of 2021 the stake was worth $15.6 billion; on 19 November 2026 it was worth about $1.79 billion, 88.5% below the peak and 12% below the cash Amazon put in. Because Amazon books it at market value, the 2022 crash cost $12.7 billion and caused Amazon's first annual loss since 2014.[2][16]

The Strategy

Amazon's strategy has three legs. It wants to lead in AI compute through aggressive capex. It wants to lead in robotics with the most robots in real use. And it partners with suppliers and providers, from Covariant and Agility to Nvidia, to get the benefits of automation without inventing everything itself. In AI models it holds stakes in the model makers, Anthropic and OpenAI, while still building its own: Jassy expects six to eight equally good models within a few years, and "one of them will be ours".[3] If Anthropic and OpenAI win, Amazon wins with them.

The big hyperscalers are converging on the same business: Microsoft, Amazon and Google are all becoming cloud providers, the role the neoclouds were built for. Amazon's difference is robotics, where it is ahead in real use, except for the blind spot of high-end robot intelligence.

Strategic Blind Spots

The biggest risk is agents. If AI agents do the shopping, they could kill the ads business and weaken the store itself, though there is no sign of it in Amazon's numbers yet, unlike the pressure AI puts on Google Search. The stakes are high: at the roughly 60% margin estimated by Wedbush and Bernstein, Amazon's $68.6 billion of ads would earn more than all of its retail. Blocking is not a lasting option, because agents can already use the shopper's own browser, which is exactly how the appeals court saw the Perplexity case.[17][13]

How will ads and e-commerce look with agents? Amazon is testing its own answer: people who click sponsored answers in Rufus buy 48% more often and spend 21% more, by Amazon's count. Walmart chose the opposite path and opened its store to approved agents in October 2026. And Amazon has not yet turned its billions of robot hours into world-class robot intelligence, while Google leads there. On 29 March 2027 the FTC's monopoly trial against Amazon begins, with structural remedies on the table.[13][3]

For Amazon

AWS growing 39% with a $578 billion backlog, ads at an estimated 60% margin, an Anthropic stake worth about $339 billion, more than a million robots, a stores margin up from 1.6% to 6.8% and a stock that held up in the midterm crash.

Against

Free cash flow of minus $42.4 billion, about $220 billion of capex and rising, debt up 163%, cloud share slipping to Microsoft and Google, no frontier model, AI agents aimed at its ads, and a P/E of 34.8 without Anthropic.

Outlook

The next report, for the fourth quarter of 2026, is expected around 4 February 2027; Amazon has not confirmed the date. It guides fourth-quarter sales of $240 billion to $247 billion and operating income of $28.5 billion to $33.5 billion. The number investors will watch most is the first capex figure for 2027, after the stock fell 4.1% the day after the February 2025 report and 5.6% after the February 2026 report, which brought the $200 billion plan. That report will also carry the first Anthropic gain after the IPO.[3][1]

In the short term the market may stay red for a while, and Amazon may suffer more with it. AI carries much of today's economic gains, so if it slows, the economy slows. But Amazon is in a uniquely good position either way. If the build-out is held back, it already owns the capacity as the market leader. If the build-out accelerates again, it has a robust business outside AI that keeps getting better and the means to invest more aggressively than anyone. In the long term Amazon should do very well.

Our view: Amazon is an AI and cloud company funded by a retail business that has finally become profitable. AWS and, by outside estimates, ads make the operating profit; Anthropic makes much of the reported profit; and the robots make the stores leaner, though by how much Amazon does not say. The risks are the size of the spending and the arrival of AI agents at the front door of its store.