Amazon Is Suddenly In Trouble

Added: 2026-05-26

[00:00:00]Amazon's CEO, Andy Jasse, got on an earnings call and told Wall Street that Amazon would spend $200 billion on AI infrastructure this year, which is up from 131 billion the year before and up from 83 billion the year before that. 3 years of spending, each one nearly doubling the last with no ceiling in sight. The stock dropped immediately, down over 10% within weeks, and investors weren't reacting to the number. They were reacting to what the number implied, that Amazon is pouring money into AI faster than any company in history, and nobody can explain what they're getting back for it. Amazon's free cash flow dropped 77% in a single year. In 2025, capital expenditure consumed almost every dollar the company earned, and the centerpiece of all that spending is a $33 billion bet on Anthropic, the AI lab behind Claude.

[00:00:55]Anthropic spent 2 6 billion on AWS cloud services in just the first 9 months of 2025. Amazon invests in Anthropic.

[00:01:05]Anthropic pays that money right back to Amazon as cloud bills. It's a financial loop and both sides count it as growth.

[00:01:13]Amazon's profit share is worth exactly nothing if anthropic never turns a profit. And just to hedge the bet, Amazon also committed up to $50 billion in open AI. So now Amazon is bankrolling both sides of the same war, hoping at least one of them figures out how to make money before the music stops. Now let's talk about Alexa, the product that was supposed to justify all of this spending decades ago. Amazon has poured over $25 billion into the Alexa and Echo ecosystem since 2017. A Wall Street Journal investigation obtained internal documents showing the devices division lost money every single year. For context, Plus, one of the most famous product failures in tech history, lost about 2 billion. Microsoft's Windows phone, lost roughly 7 billion. Alexa makes both of those look like rounding errors. So, Amazon's answer was Alexa Plus, the AI powered rebuild that would finally, after a decade and $25 billion, make Alexa smart. But when it launched on March 31st, 2025, the key features they had advertised were simply missing.

[00:02:22]No GrubHub ordering, no grocery ordering, no family member recognition, no kids story generation, no browser version. Only 300 of Alexa's 100,000 third-party skills worked with the new system. At launch, a Fortune investigation interviewed more than a dozen current and former Alexa employees. What they described was organizational chaos. a decentralized structure where teams competed against each other on overlapping problems.

[00:02:50]Managers who churned every few months, engineers who waited weeks to access their own training data because of dysfunctional privacy guardrails. The AGI team had accumulated roughly 1 million fine-tuning data points while Meta had 10 million for Llama 3. They were outgunned by their own inefficiency. Then there was Rufus, Amazon's AI shopping assistant, launched in July 2024. Rufus hallucinated product specs, recommended non-TVs when customers searched for TVs, invented features that didn't exist, and gave glowing reviews of competitors while trashing the sellers who were actually paying Amazon for advertising. Amazon sellers started reporting that AI generated review summaries were misrepresenting their products and destroying their sales. Rufus was also trivially easy to jailbreak. users figured out how to make it solve robotics math problems instead of recommending toothpaste. Less than two years after launch on May 2026, Amazon shut down Rufus entirely. But none of that compares to what happened inside Amazon's own engineering organization.

[00:03:56]In November 2025, Amazon sent an internal memo mandating that 80% of engineers use Kira, their in-house AI coding tool, on a weekly basis. It was tracked as a corporate OKR. About 1,500 engineers pushed back internally, saying they preferred other tools, but the management ignored them. In December 2025, an engineer at AWS used Kirro to fix a bug in Cost Explorer. That's the dashboard that tells you how much money you're spending on cloud services. Not exactly missionritical, but Kirro evaluated the situation and decided the correct approach was to delete the entire production environment and rebuild it from scratch. It took the team 13 hours to get it back online.

[00:04:41]Amazon's public statement called it an extremely limited event caused by user error. They mandated the tool. They tracked adoption as a corporate metric.

[00:04:51]And when their mandated tool deleted production, they blamed the engineer who used it. Then on March 2nd, it happened again. Amazon's other AI coding tool Q developer pushed bad code straight into the live retail platform. 120,000 orders vanished overnight. 1.6 million website errors hit customers across the board.

[00:05:11]People were trying to buy everyday items and Amazon was showing them delivery dates that made no sense. 3 days later, on March 5th, another outage. This one killed 99% of orders across all of North America. 6.3 million orders lost in a single day. Amazon.com effectively ceased to exist as a functioning store for 6 hours. Amazon called an emergency meeting. An internal briefing note explicitly cited Gen AI assisted changes as a contributing factor in what they described as a trend of incidents with high blast radius. And then according to the Financial Times, the reference to AI's role was deleted from the document before the meeting took place. They identified the problem and then erased the evidence that they had identified it. Their fix was to impose a 90-day safety reset requiring senior engineers to review all AI generated code before deployment on 335 critical customerf facing systems. But Amazon had just fired 30,000 corporate workers in 4 months explicitly to make the company leaner for AI. And now the solution to AI breaking everything is to put humans back in the loop. And while Amazon was dealing with this internally, it was quietly losing the one race that actually matters. AWS is still the largest cloud provider in the world. But the gap is closing, and it's closing specifically because of AI. Google Cloud grew 36% in 2025. Microsoft Azure grew 39% in Q4 alone. And AWS is growing slower than both. The problem is structural. Microsoft has Open AAI integrated directly into Office, the most widely used productivity suite on Earth. Google has Gemini embedded into Workspace and Search. Amazon has Bedrock and a set of custom tranium chips that even Anthropic apparently doesn't fully trust. In October 2025, Anthropic signed a deal to access up to 1 million Google Tpus. That is Amazon's single biggest AI investment, choosing a competitor's hardware. That is not a vote of confidence. The Financial Times projected that by 2029, Microsoft's cloud revenue would overtake AWS entirely. And then there's the human cost, which might be the most disturbing part of all of this. Leaked internal documents published by the New York Times revealed Amazon's plan to automate 75% of warehouse operations and eliminate up to 600,000 jobs by 2033.

[00:07:46]The documents showed that managers were instructed to avoid words like automation, AI, and robot in public communications and replace them with euphemisms like advanced technology.

[00:07:57]They were also told to sponsor local parades and toys for tots drives in the communities where the jobs would be disappearing. Meanwhile, their robotic warehouses have an injury rate nearly double the industry average. And the reason is simple. The robots move faster. So, Amazon expects the humans to keep up. Workers at robotic fulfillment centers are expected to pick and scan 400 items per hour. At traditional warehouses, that number is about 100.

[00:08:26]Human bodies were not built for that.

[00:08:28]Amazon's AI surveillance systems track worker productivity down to the second.

[00:08:33]Their algorithms automatically generate warnings and terminations without supervisor review. a 63-year-old Army veteran named Steven Normandin. It became a viral story. After being fired by Amazon's algorithm with no explanation and no human ever reviewing the decision, the system decided he was underperforming and that was the end of his employment. MIT Nobel laurate Darren Atimoglu told the New York Times, "Nobody else has the same incentive as Amazon to find the way to automate. Once they work out how to do this profitably, it will spread to others, too." One of the biggest employers in the United States will become a net job destroyer, not a net job creator. Amazon is a $2 trillion company, and it is bleeding out. There is a very real difference between a company that is investing in the future and a company that is hemorrhaging into a hole it dug for itself. Amazon convinced itself that if it just spent enough money fast enough, AI would transform everything it touched. And AI did transform everything it touched, but just in the wrong