AI at Amazon: No Part of Company Unaffected, Says Head of AGI

Amazon's Extensive AI Integration

Vishal Sharma, Amazon’s Vice President of Artificial General Intelligence, stated on Monday at Mobile World Congress in Barcelona that artificial intelligence is permeating nearly every facet of the company. He countered suggestions that open-source models would lessen computational demands and refrained from commenting on potential shifts in European generative AI strategies due to geopolitical tensions with the United States.

AI Deployment Across Amazon

Sharma explained that Amazon is currently implementing AI technologies through its proprietary foundational models. These are being utilized across various divisions, including Amazon Web Services – the company’s cloud computing arm – the robotic systems within its fulfillment centers, and the Alexa consumer product.

The scale of this deployment is significant. Amazon currently operates approximately 750,000 robots, which are responsible for tasks ranging from item selection to autonomous navigation within warehouses. Alexa, he noted, represents the most widely adopted AI product for home use currently available.

The Nova Model Suite

In December, Amazon Web Services unveiled a new collection of four text-generating models, collectively known as Nova. These models are rigorously evaluated against established public benchmarks.

Sharma emphasized the diversity of applications requiring different AI capabilities. He stated that a “one-size-fits-all” approach is ineffective. Certain applications, such as video generation, demand different characteristics than others, like Alexa, where rapid and predictable responses are crucial. He specifically highlighted the need to avoid inaccurate or fabricated responses, stating, “You can’t hallucinate ‘unlock the back door’.”

Compute Requirements and Open Source Models

Regarding the potential for reducing computational needs through smaller, open-source models, Sharma expressed skepticism. He believes that as AI is implemented in more diverse scenarios, the demand for increased intelligence will continue to grow.

Bedrock and Model Flexibility

Amazon has also introduced “Bedrock,” a service within AWS designed to allow companies and startups to integrate and experiment with a variety of foundational models. This includes models from companies like China’s DeepSeek, enabling seamless switching between different AI engines.

Investment in AI Infrastructure

Amazon is actively constructing a substantial AI compute cluster utilizing its Trainium 2 chips, in collaboration with Anthropic, a company in which Amazon has invested $8 billion. This development occurs alongside initiatives like xAI’s release of Grok 3, powered by a large data center in Memphis equipped with approximately 200,000 GPUs.

Sharma asserted that computational resources will remain a critical consideration for the foreseeable future, stating, “My personal opinion is that compute will be a part of the conversation for a very long time to come.”

Response to Open Source and Geopolitical Factors

Sharma did not perceive Amazon as being significantly pressured by the recent emergence of open-source models from China. He indicated that Amazon is comfortable incorporating models like DeepSeek into its AWS offerings, emphasizing the company’s commitment to providing customer choice.

“We’re a company that believes in choice … We are open to adopting whatever trends and technologies are good from a customer perspective,” Sharma explained.

Amazon's Long-Standing AI Efforts

When questioned about whether Amazon was caught off guard by the introduction of ChatGPT by OpenAI in late 2022, Sharma refuted the suggestion. He stated that Amazon has been actively engaged in AI research and development for approximately 25 years.

He pointed to Alexa as an example, noting that it utilizes around 20 different AI models and benefits from billions of pre-existing language parameters. “We’ve been looking at this for quite some time,” he added.

European Strategy and Geopolitical Considerations

Addressing the potential impact of recent geopolitical events and strains in U.S.-European relations on European companies’ GenAI strategies, Sharma acknowledged the complexity of the situation. He admitted that predicting the consequences was “very hard.”

However, he diplomatically suggested that some companies might adjust their approaches, stating, “What I will say is that it is the case that technical innovation responds to incentives.”