As generative AI gains prominence, major hyperscale players, including Amazon Web Services, Google, and Microsoft, are entering into another phase of intense competitive rivalry. The computational demands and extensive datasets required by generative AI make public cloud platforms an ideal choice. From providing foundational models as a service to facilitating the training and refinement of generative AI models, public cloud providers are actively vying to attract both developers and businesses. This article delves into the evolving strategies of Amazon, Google, and Microsoft within the generative AI sector.
Amazon Web Services: Making Significant Investments in Amazon Bedrock and Amazon Titan
While AWS entered the generative AI landscape later than its primary competitors, it is rapidly narrowing the gap. AWS is currently focusing on three key services within the realm of generative AI—Amazon Sagemaker JumpStart, Amazon Bedrock, and Amazon Titan.
Amazon SageMaker JumpStart provides an environment for accessing, customizing, and deploying ML models. Recently, AWS introduced support for foundation models, enabling users to utilize and refine popular open-source models. The collaboration with Hugging Face simplifies tasks such as inference or fine-tuning existing models from a curated catalog of open source models, offering a swift way to integrate generative AI capabilities into SageMaker.
In private preview, AWS unveiled Amazon Bedrock, positioned as a serverless environment or platform for consuming foundation models through an API. While specific details are limited, it appears to be a competitive offering comparable to Azure OpenAI. Users will likely have access to secure endpoints exposed through the private subnet of the VPC. Collaborating with GenAI startups like AI21Labs, Anthropic, and Stability.ai, Amazon aims to provide text and image-based foundation models through the Amazon Bedrock API.
Amazon Titan, on the other hand, comprises in-house developed foundation models by Amazon’s researchers and internal teams. Expected to incorporate models powering services such as Alexa, CodeWhisperer, Polly, Rekognition, and other AI services, Titan is poised to play a significant role.
Anticipating Amazon’s move, I foresee the launch of commercial foundation models for tasks like code completion, word completion, chat completion, embeddings, translation, and image generation. These models would be made accessible through Amazon Bedrock for consumption and fine-tuning.
There is also the possibility of Amazon introducing a dedicated vector database as a service, likely under the Amazon RDS or Aurora family of products. Presently, it supports pgvector, a PostgreSQL extension facilitating similarity searches on word embeddings available through Amazon RDS.
Google Cloud: Built on the Foundations of PaLM
A multitude of announcements related to Generative AI took center stage at Google I/O 2023. Generative AI holds significance for Google not only in its cloud business but also in its search and enterprise ventures based on Google Workspace.
Google’s investments have been directed towards four foundation models: Codey, Chirp, PaLM, and Imagen. These models are accessible through Vertex AI, allowing Google Cloud customers to utilize and fine-tune them with custom datasets. The model garden within Vertex AI comprises both open source and third-party foundation models. Additionally, Google has introduced a playground (GenAI Studio) and no-code tools (Gen App Builder) to facilitate the development of apps based on Generative AI.
Expanding the influence of LLM models into DevOps, Google has integrated the PaLM 2 API with Google Cloud Console, Google Cloud Shell, and Google Cloud Workstations to provide an assistant for expediting operations. This capability is accessible through Duet AI for Google Cloud.
However, a native vector database is currently absent from Google’s GenAI portfolio. Integrating the ability to store and search vectors in BigQuery and BigQuery Omni should be considered. Presently, customers must rely on the pgvector extension incorporated into Cloud SQL or resort to using a third-party vector database like Pinecone.
Microsoft Azure: Leveraging Its Investment in OpenAI to the Fullest
Through an exclusive collaboration with OpenAI, Microsoft has positioned itself as a frontrunner in the generative AI arena, outpacing its competitors. Azure OpenAI stands out as a mature and proven GenAI platform within the public cloud domain.
Azure OpenAI seamlessly migrates most foundation models from OpenAI to the cloud, excluding Whisper. Accessible through the same API and client libraries, engines like text-davinci-003 and gpt-35-turbo can be readily utilized on Azure. Launched within existing subscriptions and optionally within a private virtual network, this setup ensures customers enjoy the benefits of enhanced security and privacy for their data.
To enhance user experience, Microsoft has integrated foundation models with Azure ML, a managed ML platform as a service. This integration allows customers to employ familiar tools and libraries for consuming and refining the foundation models.
In a commitment to further innovation, Microsoft has invested in an open source project called the Semantic Kernel. This initiative aims to introduce LLM orchestration, including prompt engineering and augmentation, to C# and Python developers. It bears resemblance to LangChain, a popular open source library for interacting with LLMs.
Regarding vector databases, Microsoft has expanded the capabilities of Azure Cosmos DB and Azure Cache for Redis Enterprise to accommodate semantic search functionality.
