Amazon's AI Dilemma
Image Source | Internet (Please contact for removal if infringing) Partially AI-generated

On October 7, 2026, during Amazon Prime Day—a day that should have been a shopping frenzy—an internal Amazon Slack channel with nearly 37,000 members was flooded with a different kind of news.

A large number of employees suddenly found their access permissions revoked, with notices of position eliminations appearing in their personal inboxes. Some employees posted about the situation, while others inquired about severance plans. The channel quickly descended into chaos.

This was not the first wave of layoffs. From October 2025 to October 2026, in less than a year, Amazon laid off over 30,000 corporate positions, accounting for nearly 10% of its 350,000 white-collar workforce. This marked the largest layoff in the company's 30-year history.

More intriguingly, at the same time as the massive layoffs, Amazon announced that its 2026 capital expenditures would soar to approximately $200 billion, with the vast majority allocated to AI infrastructure.

This amount is nearly 1.5 times the $131 billion in capital expenditures from the previous year and nearly 37% higher than Wall Street expectations.

By cutting jobs on one hand and spending heavily on the other, what is this globally astute company calculating?

How Were 30,000 Employees Laid Off?

The pace of these layoffs is worth examining closely.

On October 28, 2025, Amazon's Head of Human Resources sent a memo to all employees announcing the elimination of approximately 14,000 corporate positions, affecting core departments such as AWS cloud services, Devices & Services, Advertising, Prime Video, and Human Resources.

Three months later, on January 28, 2026, a second round of layoffs occurred, with 16,000 positions cut at once. The impact further expanded to AWS's Bedrock team, Redshift data warehouse platform, and ProServe consulting department, covering regions including the U.S., U.K., and India.

The two rounds added up to exactly 30,000. But this was not the end. Amazon then shifted to "small-batch, normalized" continuous reductions: the Robotics department laid off at least 100 white-collar positions in March, and the AGI department initiated layoffs in July.

By Prime Day in October, the retail department's Stores division laid off fewer than 1,000 more employees.

Notably, over 70% of the laid-off positions were concentrated in mid-level management roles at the L5 to L7 levels.

Internally, Amazon had explicitly required teams to increase the ratio of "individual contributors to managers" by at least 15% by the end of the first quarter of 2025. In plain terms: cut out the middle layer and let workers directly report to decision-makers.

This is directly related to over-expansion during the pandemic. Amazon's total workforce swelled from about 1.3 million in 2020 to 1.576 million by the end of 2025, a roughly 21% increase over five years.

Founder Bezos, in a rare direct response during an October 8 interview, explained: "During COVID-19, people stayed home and placed a surge of orders. It was an incredibly high-pressure period, with the entire team working desperately, but our workforce did expand excessively."

He defined the layoffs as necessary corrections rather than strategic contractions.

Before saying "layoffs save money," one must first do the math: laying off people itself costs a lot.

Amazon's Q3 2025 financial report contained a key data point: $1.8 billion in estimated severance costs, involving North America, International, and AWS business segments.

By the end of 2025, Amazon's cumulative severance expenditures had reached approximately $2.7 billion. According to Business Insider estimates, the average cost of laying off one Amazon employee is about $60,000.

So how much was saved? According to Cybernews, the first round of 14,000 layoffs alone is expected to save the company up to $3 billion annually, with cumulative savings from this strategy reaching $12.6 billion by 2027.

$3 billion in annual savings, compared to $200 billion in AI capital expenditures, is not even a fraction.

The truth behind the continuous layoffs is that the cost savings from layoffs are just a drop in the bucket. What Amazon is truly doing is switching the company's overall resource structure from "supporting people" to "supporting computing power." The money saved from layoffs is a signal telling Wall Street: this company is seriously making trade-offs.

What Is Amazon Anxious About?

Since the AI boom, what is driving Amazon's continuous layoffs?

The throne of cloud computing is loosening.

AWS is Amazon's profit engine. In 2025, AWS contributed $128.7 billion in revenue, accounting for about 18% of the company's total revenue but more than 60% of operating profit.

If this business falters, the entire Amazon will shake.

But competitors are closing in. In Q4 2025, AWS revenue grew 24% year-over-year, already the fastest growth in 13 quarters. However, Microsoft Azure grew 39%, and Google Cloud grew 48% in the same period.

Amazon AWS's CEO admitted: "We're growing from a significantly larger base, generating more incremental revenue than others, but our growth rate is indeed being caught up."

More troubling is that Microsoft has locked in hundreds of billions of dollars in Azure cloud spending through its investment in OpenAI, while Google Cloud is rapidly seizing market share with its AI services.

If AWS does not accelerate its AI computing power deployment, its first-mover advantage may be gradually eroded.

At the AI model level, Amazon lacks its own ace.

This is Amazon's most awkward spot. It is the absolute leader in cloud infrastructure but has always been a "distributor" in the core area of large models.

It mainly provides third-party models like Anthropic's Claude and Meta's Llama on the Bedrock platform. Its self-developed Nova model is roughly equivalent to the mid-to-low end of Claude's capabilities.

This is like opening the city's largest supermarket, where the shelves are stocked entirely with other brands, and your own private-label products go unsold.

In the long run, Amazon is unlikely to "get a share" of the most profitable models themselves.

So Amazon made a bold decision: in April 2026, it announced additional investment in Anthropic, raising the total from $8 billion to $33 billion, with subsequent investment plans capped at $65 billion, and committed to providing at least 10 gigawatts of computing power support.

In exchange, Anthropic promised to spend over $100 billion on AWS over the next decade.

The essence of this deal is Amazon exchanging computing power for models, binding core players in the AI era with infrastructure.

It does not pursue being the best at making models but ensures that all the best models run on its cloud.

The e-commerce base is also being eroded.

What is easily overlooked is that Amazon's e-commerce business is also under pressure. According to the International Post Corporation's 2025 Cross-Border E-Commerce Consumer Survey, Amazon's global market share in cross-border e-commerce dropped from 26% in 2022-2023 to 24% in 2025, while Temu caught up to 24% in the same period.

With slowing e-commerce growth, the need for massive AI investments, and intensifying cloud computing competition, Amazon's choice is essentially only one: concentrate limited resources on the most critical battlefields.

Mid-level management and process coordination roles that do not directly generate competitiveness became the first to be "optimized."

The $200 Billion Bet

Understanding the anxiety helps explain why Amazon dares to bet $200 billion despite profit pressures.

From a technological standpoint, Amazon's AI strategy is shifting from "selling others' computing power" to "building its own tools."

Self-developed chips. Amazon's Trainium series AI chips have delivered over 1.4 million units, with the latest Trainium 3 offering a 40% improvement in cost-effectiveness. By mid-2026, its production capacity is "almost fully booked."

The CEO explicitly stated on the earnings call that Trainium is already a "business generating billions in annualized revenue and achieving triple-digit growth."

This means Amazon is gradually reducing its reliance on NVIDIA GPUs and bringing the cost structure of AI computing power under its control.

Organizational integration. In December 2025, Amazon merged its large model R&D team (Nova) with its self-developed chip team (Trainium) under a unified organization led by a 27-year infrastructure veteran.

The logic behind this move is clear: optimize every line of model code for peak performance on self-developed chips, achieving deep hardware-software coupling and lowering unit computing costs.

Locking in major clients. Beyond Anthropic's $100 billion commitment, Amazon secured a $38 billion cloud computing deal with OpenAI.

Meanwhile, AWS's backlog surged 40% in Q4 2025 to $244 billion, with a weighted average contract term of 3.8 years. These numbers indicate very high revenue visibility for the coming years.

After announcing the $200 billion capital expenditure plan, Amazon's stock price plunged more than 11% after hours, erasing over $200 billion in market value.

Investors are concerned: How will you manage free cash flow after spending so much? How long is the payback period?

Amazon's free cash flow (rolling 12 months) in Q4 2025 was just $11.2 billion, while capital expenditures were more than ten times that.

This gap is widening. Amazon even issued $25 billion in bonds to finance AI data center construction.

In fact, if we zoom out, Amazon's choice represents a collective shift across the tech industry.

According to layoff tracking platform Layoffs.fyi, the global tech industry laid off approximately 245,000 employees in 2025. The top four U.S. tech giants, including Amazon, Microsoft, Alphabet, and Meta, are expected to spend over $630 billion in total capital expenditures in 2026.

Meanwhile, about 55,000 layoffs in the U.S. in 2025 were explicitly attributed to AI replacement.

This is not simply "AI taking jobs" but a structural shift in capital allocation.

Meta nearly doubled its AI-related spending while laying off employees, Oracle bluntly stated that "AI implementation drives reductions," and Microsoft laid off about 15,000 while Azure growth surged to 39%.

Some analysts describe this phenomenon as "prosperity layoffs": company performance is not poor, stock prices are even rising, but the organization is being rewritten.

The growth model supported by large-scale hiring over the past decade is failing, replaced by "fewer people + stronger AI tools + more expensive computing power."

What does this mean for ordinary workers? Over 70% of the 30,000 laid off by Amazon were mid-level managers.

These individuals were not incompetent, but their work—coordination, approval, reporting, process management—is precisely what AI is best at replacing.

Amazon is already using AI agents for data analysis and project coordination, completing tasks in minutes that previously required teams days to finish.

The truly secure positions are shifting toward two ends: one end is core technical personnel who can define AI capability boundaries, and the other is frontline business personnel who can directly reach customers and understand real needs.

The middle layer of "upward communication, downward coordination" roles is becoming the most dangerous zone.

Amazon's moves this round are essentially a story about "trade-offs."

It has abandoned the organizational bloat from the pandemic era, the comfortable positioning as a pure "AI distributor," and even temporarily sacrificed cash flow performance that would please investors.

In exchange, it gains a leaner organization, a set of self-developed chip-plus-model foundational capabilities, and an entry ticket to the AI era's computing infrastructure.

Whether the $200 billion bet will pay off, no one knows.

But one thing is certain: in the AI race, not betting is the biggest risk of all.

Amazon's choice may not be correct, but at least it is not wasting time in hesitation.

What makes Amazon's layoffs worth pondering is that they reveal how organizations are being reshaped in the AI era.

Seeing this direction is more useful than anxiety.