亚马逊AGI组织裁员，AWS的AI估值锚要换了吗？

TL;DR

• Amazon confirms job cuts in its AGI organization, while stating that large AI models remain a priority.
• The market is divided on whether this adjustment signifies a downgrade of in-house model efforts or a refocusing on customer-funded projects.
• Related assets: AMZN, AWS, Anthropic, AI cloud infrastructure chain.

On July 22, Amazon confirmed that some positions within its AGI organization were eliminated, while also stating that the company continues to build large AI models, calling it one of its most important tasks. According to a report cited by Reuters, Amazon explained the adjustment as a move to concentrate resources on the areas most critical to customers' future.

This round of layoffs did not disclose the specific number of people affected and cannot be directly interpreted as Amazon abandoning AGI. Instead, it brings the contradictions within Amazon's AI narrative to the forefront: Big Tech continues to invest tens of billions of dollars in AI infrastructure, yet the teams closest to long-term AI ambitions are beginning to face organizational contraction.

For investors, the question isn't how many people Amazon laid off, but that the valuation anchor for AWS's AI is being reassessed. In the past, the market was willing to pay a premium for Big Tech's AI investments, assuming that stronger models would lead to greater future revenue. Now, the more pressing question is when these investments will translate into paying customers, cloud revenue, and margin improvement.

AGI (Artificial General Intelligence) can be simply understood as a long-term, yet unrealized goal, enabling AI to learn and solve problems across different fields like a human. It represents long-term imagination but doesn't necessarily translate into immediate revenue. What AWS needs is to package AI capabilities into services that enterprises can buy, use, and customize today.

Greater AI Investment Leads to Tougher Organizational Trade-offs

The key to these layoffs isn't whether Amazon will continue building models. The official statement has already set the boundaries: large models remain a focus, but resources must be allocated to the projects with the highest priority for customers.

Organizational actions show that AI investment hasn't stopped, but the tolerance for error is decreasing. In December 2025, Amazon adjusted its AI-related leadership. Andy Jassy announced that Peter DeSantis would lead new organizations for AI models, chips, and quantum computing. Rohit Prasad left in late 2025, and Pieter Abbeel took over frontier model research within AGI. A report cited by Reuters also mentioned that AGI Lab head David Luan departed in February 2026.

Viewed together, these changes suggest Amazon isn't exiting the AI arms race but is re-prioritizing its internal investment portfolio. Long-term research retains its narrative value, but projects closer to customers, revenue, and productization are gaining higher priority.

This is also the common backdrop for all Big Tech AI trades. Over the past two years, the market primarily traded on who dared to spend, who had computing power, and who had the models. Now, capital expenditure alone is no longer scarce enough; investors are starting to question the return on investment: can model teams, chips, data centers, and talent ultimately solidify into revenue?

Nova Forge Provides a Path to Commercialization

To understand this adjustment, one must look at Nova Forge, which AWS launched during its re:Invent conference in December 2025. It's not an ordinary chatbot but a service designed to help enterprises train custom models.

In the traditional path, if an enterprise wants a frontier model suited for its industry, it either trains from scratch, which is extremely costly, or fine-tunes an existing model, yielding limited capability and controllability. Nova Forge's approach is to let customers start from checkpoints (intermediate training archives) within the Amazon Nova model training process, blending their own data with Amazon's curated datasets at different training stages.

Amazon calls this open training. In simple terms, enterprises don't need to build a large model from zero. Instead, they start from a model base that Amazon has already trained to a certain stage, injecting their industry knowledge early on. This allows them to inherit foundational capabilities while more easily achieving domain specialization.

This path is crucial for AWS because it attempts to turn model capabilities into cloud service products. Customers aren't just calling a model API; they are training, hosting, deploying, and optimizing their own models on AWS. If this product works, it could lead to significant compute consumption, platform stickiness, and ongoing operational revenue.

However, current information doesn't prove that the resources from the cut AGI positions have been redirected to Nova Forge. A safer conclusion is that the AGI organizational adjustment coincides with the emergence of customer-oriented products like Nova Forge, indicating Amazon's tendency to increase the weight of commercial projects.

AWS's Competitive Focus Shifts to Customer Customization

Amazon's position in the foundational model race has always been unique. It develops its own Nova models, invests in Anthropic, and must also maintain AWS's neutrality and model ecosystem as a cloud platform.

This dictates that AWS may not necessarily win by having the world's most powerful model alone. For enterprise customers, model leaderboards are important but not the only criterion. More practical questions are: Can it access internal corporate data? Can it meet security and compliance requirements? Can it reduce training costs? Can it run alongside existing cloud services?

Nova Forge aligns precisely with this competitive logic. It shifts the battlefield from a ranking of general-purpose model capabilities to enabling enterprises to train their own models at lower costs. If this path succeeds, AWS can embed AI revenue into its core cloud business, rather than betting solely on a consumer-grade AI product.

This also explains why Amazon is retaining its AGI narrative while cutting some positions. The former preserves long-term technological imagination, while the latter forces teams to direct resources toward directions more readily validated by customer demand.

For AMZN, the market will ultimately not just look at whether Amazon has an AGI team. More importantly, AWS needs to prove that its AI services increase customer spending, enhance stickiness, and do not significantly drag down profit margins.

Orders and Profit Margins Will Provide the Answers

These layoffs could easily be framed as one of two extremes: either it's a sign of Amazon's AI failure, or it's an insignificant routine optimization. Current information doesn't support either conclusion.

The more reasonable assessment is that Amazon is still in the AI arms race, but its internal budget and talent allocation are shifting towards areas that can be sold to customers. This change is meaningful for investors because AMZN's AI valuation premium will increasingly depend on AWS's commercialization results, rather than purely relying on model narrative.

The verification points will hinge on a few specific things. Can Nova Forge secure real enterprise customers? Are customers willing to pay continuously? Do the trained models offer better value than standard fine-tuning? These factors will determine whether it becomes an effective product.

Another variable is talent attrition. If the AGI organizational adjustment merely optimizes less critical roles, the impact is limited. However, if core research and engineering talent leaves, Amazon's long-term competitiveness in self-developed models will be weakened. The tension between official statements and organizational reality will ultimately be resolved through product adoption rates, AWS AI revenue, and the return on capital expenditure revealed in future earnings reports.