The future of generative AI: A comparison of AWS, Google, and Microsoft

The field of generative AI relies heavily on extensive computational resources and vast datasets, making the public cloud an optimal platform of choice. Among the leading hyper scalers in this domain, namely Amazon Web Services, Google, and Microsoft, a new phase of intense competition has emerged.

Given the resource requirements of generative AI, the public cloud offers an advantageous environment. Public cloud providers are actively vying to capture the attention of developers and enterprises by providing a range of services, including foundational models as a service, training capabilities, and fine-tuning options for generative AI models.

This article delves into the evolving strategies of Amazon, Google, and Microsoft in the field of generative AI. The following table provides a summary of the current state of GenAI services offered by these key public cloud providers:

Through an analysis of these market players, we aim to shed light on the ongoing competition and advancements in the realm of generative AI.

AWS Generative AI

Amazon Web Services (AWS) is making significant investments in the field of generative AI, despite joining the party a bit later than some of its main competitors. AWS is swiftly catching up by focusing on three key services: Amazon Sagemaker JumpStart, Amazon Bedrock, and Amazon Titan.

Amazon SageMaker JumpStart provides users with an environment to access, customize, and deploy machine learning (ML) models. Recently, AWS introduced support for foundation models, allowing customers to utilize and fine-tune popular open-source models. Through a partnership with Hugging Face, AWS has simplified the process of performing inference or refining existing models from a curated catalog of open-source models. This approach offers a quick way to leverage generative AI capabilities within SageMaker.

In a private preview, AWS introduced Amazon Bedrock, which serves as a serverless environment or platform for consuming foundation models through an API. Although AWS has not shared extensive details, it appears to be a compelling alternative to Azure OpenAI. Customers will have access to secure endpoints that are exposed through the private subnet of the Virtual Private Cloud (VPC). Furthermore, Amazon has collaborated with GenAI startups like AI21Labs, Anthropic, and Stability.ai to provide text and image-based foundation models through the Amazon Bedrock API.

Another noteworthy offering is Amazon Titan, a collection of proprietary foundation models developed by AWS’s own researchers and internal teams. Titan is expected to incorporate models that power popular services such as Alexa, CodeWhisperer, Polly, Rekognition, and other AI services.

Anticipating further developments, I believe Amazon will soon launch commercial foundation models for various applications, including code completion, word completion, chat completion, embeddings, translation, and image generation. These models will be made accessible through Amazon Bedrock for consumption and fine-tuning.

Additionally, there are indications that Amazon might introduce a dedicated vector database as a service, potentially as part of the Amazon RDS or Aurora product families. Currently, AWS supports pgvector, a PostgreSQL extension that enables similarity searches on word embeddings, which are available through Amazon RDS.

Google Cloud Generative AI

The recent Google I/O 2023 event saw a wave of announcements related to GenAI, which took center stage. For Google, generative AI holds significant importance not only for its cloud services but also for its search and enterprise offerings under Google Workspace.

Google’s commitment to GenAI is evident through its investment in four foundational models: Codey, Chirp, PaLM, and Imagen. These models are readily accessible to Google Cloud customers via Vertex AI, enabling them to utilize and refine the models with their own custom datasets. The model garden within Vertex AI comprises a collection of open-source and third-party foundational models. To facilitate the development of GenAI-based applications, Google has introduced GenAI Studio, a dedicated playground, and Gen App Builder, a set of user-friendly, no-code tools.

In a bid to extend the capabilities of LLM models to DevOps, Google has seamlessly integrated the PaLM 2 API with its Google Cloud Console, Google Cloud Shell, and Google Cloud Workstations. This integration introduces an assistant-like functionality that accelerates operational tasks. This powerful feature is now available through Duet AI, specifically designed for Google Cloud.

However, it’s worth noting that Google’s GenAI portfolio currently lacks a native vector database. It would be beneficial for Google to incorporate the ability to store and search vectors within BigQuery and BigQuery Omni. As of now, customers have the option to leverage the pgvector extension added to Cloud SQL or opt for a third-party vector database like Pinecone to fulfill their vector storage and search needs.

Microsoft Azure Generative AI

Microsoft has positioned itself as a frontrunner through its exclusive partnership with OpenAI. Among the diverse offerings in the public cloud, Azure OpenAI stands out as a mature and established GenAI platform.

Azure OpenAI seamlessly brings the majority of OpenAI’s foundational models to the cloud, with the exception of Whisper. Through a unified API and client libraries, users can easily leverage engines such as text-DaVinci-003 and gpt-35-turbo on Azure. What’s more, these engines can be accessed within existing subscriptions and, if desired, through a private virtual network, ensuring enhanced data security and privacy for customers.

Recognizing the significance of these foundational models, Microsoft has seamlessly integrated them with Azure ML, a managed platform-as-a-service for machine learning. This integration empowers customers to employ familiar tools and libraries to consume and fine-tune these foundational models according to their specific needs.

Another notable investment made by Microsoft is the Semantic Kernel, an open-source project catering to C# and Python developers. The Semantic Kernel focuses on facilitating LLM (Language Model) orchestration, including prompt engineering and augmentation. Similar to the widely popular LangChain library, it enables developers to interact with LLMs effectively.

In addition to these advancements, Microsoft has expanded the capabilities of Azure Cosmos DB and Azure Cache for Redis Enterprise to support semantic search within vector databases. This enhancement enables users to perform more meaningful and contextually aware searches.

By leveraging Microsoft Azure’s OpenAI investment, users gain access to a powerful suite of AI tools and capabilities that drive innovation and efficiency. Microsoft’s continuous commitment to advancing these technologies underscores its dedication to empowering businesses and developers in the AI landscape.

AWS, Google, or Azure generative AI FAQs