Chips, agentic AI, and partners take center stage at AWS re:Invent 2024

When it comes to generative AI (GenAI), 2023 was about the ‘What’—sparked by the launch of ChatGPT, which opened enterprises’ eyes to the technology’s possibilities. In 2024, the focus shifted to the ‘Why’ as organizations prioritized investment areas such as predictions, personalization, and productivity. 2025 promises to be about the ‘How,’ with enterprises tackling challenges around data, technology, culture, skills, and processes. AI-led agentic services will take center stage in addressing the ‘How,’ augmenting human capabilities with smart AI agents to optimize processes and decision-making—if we could only orchestrate and govern them.

HFS Research anticipates that by 2030, software and services will become one blended, scalable solution that we call ‘Services-as-Software.’ Against this backdrop, AWS hosted re:Invent 2024 in Las Vegas—a milestone event drawing more than 60,000 in-person attendees, 3,500 speakers, and 400,000 online viewers and bringing together engineers, business executives, startups, partners, and more. re:Invent is a learning conference, and HFS Research identified the key themes that stood out at this year’s event. Amid the numerous tech announcements, we noted a little something extra—a keen focus and perhaps enhanced respect from AWS for its partners, especially tech and services. We call this the ‘Garman effect’—after the former intern and man who’s touched most functions at AWS is driving some overdue appreciation for the power of ecosystems.

NVIDIA needs company—AWS looks to retain its computing power and enrich its value prop with silicon

Today, there’s really only one choice on the GPU side, and it’s just NVIDIA. We think that customers would appreciate having multiple choices.

— AWS CEO Matt Garman

As companies continue to invest in GenAI, the amount of data being churned out will rise exponentially—and, with it, the demand for computing power and the chips that power them. This boom fueled the demand for chips to train AI models, resulting in NVIDIA’s rapid expansion. As a leader in cloud computing, AWS should continue to expand its services to keep up with the demand. At re:Invent, it unveiled significant investments in compute and building custom silicon.

• General availability of Trainium 2: AWS launched Trainium 2, delivering 30–40% better price performance than GPU-based alternatives. Designed to efficiently handle AI workloads such as training and running large language models (LLM), Trainium 2 instances power the new EC2 Trainium 2 UltraServers, which offer 4x the computing power of regular instances. These are built to manage the largest AI models with greater speed and efficiency, making them an ideal choice for businesses scaling their AI efforts.
• Trainium 3 set for late 2025 release: AWS announced that its next-gen AI training chip, Trainium 3, will be made using the 3nm process—improving performance, power efficiency, and density. This chip is expected to provide 4x greater performance than previous versions.
• Project Rainier: In partnership with Anthropic, AWS also announced ‘Project Rainier,’ a nod to the mighty Mt. Rainier. This project features the creation of a massive AI-focused cluster containing thousands of Trainium 2 chips to power cutting-edge AI model training.

At the event, Apple revealed that it is using AWS’ custom AI chips for services such as search and will evaluate whether the company’s latest AI chip can be used to pre-train its models such as Apple Intelligence. As a technology company at the forefront of innovation, this endorsement is a huge boost for AWS as it competes with NVIDIA and rival hyperscalers that are also investing in similar chips.

Agentic AI is on the move as AWS looks toward Services-as-Software

I feel generative AI has the potential to transform every single industry, every single company out there

— AWS CEO Matt Garman

While the focus until now was largely productivity-centered, the real promise of GenAI and agentic technologies is their potential to change how we execute work, not just reduce manual toil. HFS’ vision of technological innovation for the next few years shows the lines clearly blurring between software and services. As we rapidly embrace agentic technologies, rote manual processes are getting automated and businesses will increasingly consume Services-as-Software (see Exhibit 1).

Exhibit 1: HFS’ Services-as-Software showcases the lines blurring between software and services

Source: HFS Research, 2024

At re:Invent, AWS stated that Amazon Bedrock, a platform that enables users to build AI tools using LLMs, saw 5x growth last year. ISVs such as Salesforce, SAP, and Workday have integrated Bedrock into their own applications. The reason behind its growth is the availability of choice for its users in the models they use. Choice, much like the building blocks in AWS’ logo, remains a distinct hallmark of AWS’ value.

Amazon announced the launch of Amazon Bedrock support for multi-agent collaboration. With the fully managed multi-agent collaboration capability on Amazon Bedrock, specialized agents work within their domains of expertise, coordinated by a supervisor agent who breaks down requests, delegates tasks, and consolidates output into a final response. The service takes automation beyond routine tasks to more complex tasks, providing avenues to improve coordination capabilities, communication speed, and overall effectiveness. This orchestration has been sorely lacking with AWS and its competitors.

AWS also announced new AI capabilities to Amazon Q Developer, an AI assistant that enables developers to easily modernize and migrate legacy applications, especially those on Windows, VMware, and mainframes. While mainframe migrations typically take years, AWS claims Q’s new capabilities can reduce the time significantly. The advancements in code generation enable developers to use code to create codes and perform mundane but important tasks such as documentation, allowing them to focus more on building innovative products and increasing productivity. AI, along with the developer, can effectively create a super developer. This is a prime example of HFS’ Services-as-Software vision coming to life.

The Garman effect: Partnerships will be more important than ever to succeed

While working methods will evolve as GenAI continues to disrupt the way companies work, the importance of partnerships will remain paramount.

Security is a key building block to all products and services, and AWS announced four new certifications for partners focused on security and governance. These include AI Security, Digital Sovereignty, Amazon Security Lake Service Ready, and AWS Security Incidence Response.

AWS sees the marketplace as a key enabler for customer solutions, with 99% of the top 1,000 customers having at least one active marketplace subscription. To democratize access to the AWS marketplace, ‘Buy with AWS’ is now available across partner websites at the click of a button. More than 50 partners have already integrated it into their platforms.

AWS CEO Matt Garman believes that partners should offer Services-as-Software to be profitable in 2025 and beyond. Investments in proprietary IP that build bespoke solutions can result in greater margins and customer buy-ins for partners. AWS is also pushing for joint go-to-market opportunities, which partners should make the most of, as it provides greater visibility, quicker deal closures, and deal volumes.

In the future of Services-as-Software, much of the human-driven rote work—of which a large part is done by the service provider community—will become obsolete. However, AWS seems to realize that process domain and industry expertise are unique areas of expertise that the partner channel brings in spades. Word from the event floor was that AWS is uncharacteristically open and engaged with its system integrators (SI) and tech partners.

The Bottom Line: AWS re:Invent provides pointers toward a Services-as-Software future where innovation will be essential.

The AWS re:Invent 2024 highlighted how enterprises can leverage computation advancements, AI tools, and partnerships to drive transformation. Innovations such as Trainium chips and Bedrock simplify AI adoption, while partnerships enhance security and expand AWS Marketplace access. These announcements point to a future where services are delivered increasingly as software—scalable, efficient, and seamless to integrate—enabling enterprises to focus on innovation over infrastructure. As AI reshapes workflows, enterprises should embrace these tools and collaborations to accelerate their digital transformation and unlock new value. AWS is committed to continuing its role as the go-to provider of options across compute, storage, data, databases, and AI-enablement.