Cloud Services Cross-Reference: AI & Machine Learning

This document maps AI and machine learning services across AWS, Azure, Oracle Cloud Infrastructure (OCI), and Google Cloud Platform (GCP). All four providers offer managed ML platforms, generative AI APIs, speech and vision services, and AI-optimized compute infrastructure, but their ecosystems diverge significantly in model breadth, proprietary hardware, integration depth, and pricing. AWS leads in breadth of pre-trained APIs and multi-model foundation model access via Bedrock. Azure holds a structural advantage through its exclusive enterprise deployment of OpenAI models. GCP is unique in offering proprietary TPU accelerators unavailable from other providers. OCI differentiates on AI infrastructure scale and Oracle Database integration.

1. ML Platforms

Managed ML platforms provide end-to-end tooling for data preparation, model training, experiment tracking, model registry, deployment, and monitoring. They are the backbone of custom model development workflows.

AWS — Amazon SageMaker AI SageMaker AI is AWS's unified ML platform covering the full model lifecycle. It provides JupyterLab-based Studio notebooks, managed training clusters via HyperPod (which reduces training time by up to 40% through automated fault recovery), MLflow integration for experiment tracking, a Feature Store, and model monitoring. SageMaker JumpStart offers a model hub for deploying pre-trained and fine-tuned models. SageMaker Autopilot provides AutoML. Inference options span real-time, serverless, asynchronous, and batch patterns across more than 70 instance types. In 2025, SageMaker was rebranded as the center for both analytics and AI, with SageMaker Unified Studio integrating data engineering, analytics, and ML workloads in a single environment.

Azure — Azure Machine Learning Azure Machine Learning (Azure ML) is a fully managed platform for building, training, deploying, and managing ML models. It supports AutoML, Designer (visual drag-and-drop pipeline builder), managed compute clusters, and MLflow integration. Azure ML integrates with Azure Databricks for large-scale data processing, Microsoft Fabric for unified data and analytics, and Azure DevOps for CI/CD. The Responsible AI Dashboard provides model explainability, fairness assessment, and error analysis. Azure ML is the custom model development platform; it is distinct from Azure AI Foundry, which focuses on generative AI application development.

OCI — OCI Data Science OCI Data Science is a fully managed platform built around JupyterLab notebooks. It provides the Accelerated Data Science (ADS) SDK, managed ML pipelines, distributed training with NVIDIA GPUs, model versioning via a Model Catalog, and model monitoring for data and concept drift. AutoML capabilities are available through the ADS library and Oracle Machine Learning (OML), the latter of which runs in-database on Autonomous Database. OCI Data Science integrates with OCI Object Storage, OCI Vault, and OCI Functions for MLOps automation.

GCP — Vertex AI Vertex AI is Google's unified ML and generative AI platform, consolidating what were previously separate products (AI Platform, AutoML, AI Hub). It provides Vertex AI Workbench for notebook development, Vertex AI Pipelines (Kubeflow-compatible) for workflow orchestration, Model Registry, Feature Store, Vertex AI Evaluation for model quality assessment, and Vertex AI Experiments for tracking. Vertex AI Agent Engine (GA in 2025) provides hosted infrastructure for deploying AI agents. The platform supports PyTorch, TensorFlow, JAX, and scikit-learn natively.

Key differentiators:

• OCI Data Science is the only platform with a first-class in-database ML option (OML on Autonomous Database), enabling SQL-based model training without data movement.
• GCP Vertex AI Pipelines uses the open Kubeflow Pipelines SDK, providing the highest portability for pipeline definitions.
• Azure ML's Responsible AI Dashboard is the most comprehensive built-in model explainability and fairness toolset among the four providers.

2. Generative AI / LLM Services

Generative AI services provide managed access to large language models (LLMs) and foundation models via API, without requiring customers to host or operate the underlying model infrastructure.

AWS — Amazon Bedrock Amazon Bedrock is a fully managed service providing a unified API to nearly 100 foundation models from AI21 Labs, Anthropic (Claude), Cohere, DeepSeek, Meta (Llama), Mistral AI, OpenAI, Stability AI, Writer, and Amazon's own Nova model family. Bedrock Agents enables agentic workflows with tool use and retrieval-augmented generation (RAG). Bedrock Knowledge Bases provides managed RAG with automatic document chunking and vector storage. Bedrock Guardrails applies content filtering and topic restrictions. Amazon Q is a separate generative AI assistant for business and developer use cases. As of late 2025, Bedrock Marketplace offers 100+ additional models beyond the core catalog, making it the broadest model marketplace of the four providers.

Azure — Azure OpenAI Service (Foundry Models) Azure OpenAI Service provides exclusive enterprise deployment of OpenAI's models — GPT-4o, GPT-4o mini, o1, o3, DALL-E 3, Whisper, and text embedding models — under Azure's compliance, data residency, and private networking guarantees. The service is part of Azure AI Foundry (formerly Azure AI Studio), which provides a unified development environment for generative AI applications. Azure AI Foundry Models (formerly Azure Model Catalog) extends access beyond OpenAI to include models from Mistral AI, Meta (Llama), DeepSeek, xAI (Grok), Black Forest Labs (FLUX), and others. Azure AI Agent Service provides infrastructure for building multi-agent systems.

OCI — OCI Generative AI Service OCI Generative AI is a fully managed service providing access to models from Cohere (Command R+, Command A), Meta (Llama 3.x), and Mistral on shared or dedicated AI clusters. The service supports text generation, summarization, and embeddings. Fine-tuning on dedicated clusters using customer data is available (T-few fine-tuning method). OCI AI Agent Platform (Agent Hub, in beta as of late 2025) provides RAG-based agent creation connected to enterprise data sources. The service integrates directly with Oracle Fusion Applications and Oracle Integration Cloud for enterprise use cases.

GCP — Gemini on Vertex AI / Model Garden Vertex AI provides access to Google's first-party Gemini model family (Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite) as well as third-party models including Anthropic's Claude and open models (Gemma, Llama 3.x). Gemini 2.5 Pro features a 1-million-token context window and adaptive thinking for complex reasoning tasks. Gemini 2.5 Flash delivers low-latency responses with a configurable thinking budget. The Gemini Live API (GA in 2025) supports real-time multimodal voice and video conversation. Vertex AI Model Garden provides a curated catalog of 200+ models. Imagen handles image generation; Veo handles video generation; Chirp handles speech.

Key differentiators:

• Azure OpenAI Service is the only way to deploy OpenAI's GPT and o-series models with enterprise SLAs, data residency guarantees, and private networking.
• AWS Bedrock offers the widest model selection from the largest number of third-party providers, making it suitable for multi-model experimentation and vendor diversification.
• GCP is the only provider offering proprietary multimodal generative models for image (Imagen), video (Veo), and audio/speech (Chirp) in a unified platform.
• OCI Generative AI's dedicated AI cluster option gives the highest isolation for regulated industries requiring physical separation of model serving infrastructure.

3. Speech-to-Text and Text-to-Speech

Speech services provide automatic speech recognition (ASR) for audio-to-text transcription and neural text-to-speech (TTS) synthesis for generating natural-sounding audio from text.

AWS — Amazon Transcribe (ASR) / Amazon Polly (TTS) Amazon Transcribe supports real-time and batch transcription, custom vocabulary, custom language models, speaker diarization, channel identification, and content redaction for PII. Medical-specific transcription is available via Amazon Transcribe Medical. Amazon Polly provides 100+ voices across 40+ languages and language variants, including Neural Text-to-Speech (NTTS) voices for higher naturalness and Brand Voice for custom voice creation.

Azure — Azure Speech (Foundry Tools) Azure Speech Service unifies speech-to-text, text-to-speech, speech translation, and speaker recognition in a single API. The speech-to-text engine supports real-time and batch transcription with custom speech model training. Text-to-speech provides Neural voices in 400+ voices across 140+ languages, including personal voice cloning and custom neural voice creation. Speaker recognition includes verification and identification. Azure Speech integrates with Azure AI Language for combined transcription-plus-NLU workflows.

OCI — OCI Speech OCI Speech provides both ASR and TTS capabilities. ASR supports English and several other languages with support for custom vocabulary. TTS offers neural voices with adjustable speed, pitch, and SSML support. The service handles real-time and asynchronous processing modes and integrates with OCI Object Storage for batch audio file workflows.

GCP — Cloud Speech-to-Text / Cloud Text-to-Speech Cloud Speech-to-Text supports 125+ languages and dialects, real-time streaming recognition, speaker diarization, word-level timestamps, and medical speech recognition (MedLM). Cloud Text-to-Speech provides 380+ voices across 75+ languages using WaveNet and Neural2 voice technologies. The Chirp model (available in Vertex AI) provides a universal ASR model supporting 100+ languages with improved accuracy on short utterances.

Key differentiators:

• Azure Speech provides the most comprehensive speaker management and voice cloning capabilities, including personal voice cloning in preview.
• GCP's Chirp model (Vertex AI) delivers state-of-the-art accuracy on short utterances across 100+ languages using a single universal model.
• AWS Transcribe Medical is a purpose-built, HIPAA-eligible ASR service with medical vocabulary support unavailable as a distinct offering from GCP or OCI.

4. Computer Vision / Image Analysis

Computer vision services provide pre-trained APIs for detecting objects, faces, scenes, text in images, and analyzing video streams without requiring custom model training.

AWS — Amazon Rekognition Amazon Rekognition provides image and video analysis including object and scene detection, facial detection and analysis, facial search and comparison against a stored collection, celebrity recognition, content moderation, text-in-image detection (OCR), and personal protective equipment (PPE) detection. Custom Labels allows training custom object and scene detectors using AutoML with as few as 10 labeled images. Rekognition Video analyzes stored or streaming video and integrates with Amazon Kinesis Video Streams.

Azure — Azure AI Vision (Foundry Tools) Azure AI Vision provides image analysis (object detection, scene understanding, captions, dense captions using GPT-4V), OCR, face detection and attribute analysis (via Azure AI Face), and image segmentation. The Florence foundation model underlies many Vision capabilities, providing high-accuracy image understanding. Custom Vision allows training custom classifiers and object detectors. Video Indexer provides video analysis including transcription, translation, named entities, and speaker detection.

OCI — OCI Vision OCI Vision provides pre-trained models for image classification, object detection, and text recognition (OCR). Custom model training with labeled image datasets is supported. The service integrates with OCI Document Understanding for document-specific vision tasks. Use cases include quality inspection, defect detection, and retail shelf analysis via custom models.

GCP — Cloud Vision API / Video Intelligence API Cloud Vision API supports label detection, object localization, face detection (no biometric identification — facial recognition limited by policy), logo detection, landmark detection, OCR (both standard and document OCR), image safe search, and web entity detection. Video Intelligence API extends analysis to video with shot detection, object tracking, text detection, and content moderation. Vertex AI Vision provides a managed computer vision platform for building and deploying custom vision models at scale.

Key differentiators:

• AWS Rekognition provides the most complete facial analysis and searchable face collection capability; it remains the only major provider offering facial recognition as a core product (subject to AWS policy controls).
• Azure AI Vision's Florence-based image captioning (dense captions) produces natural-language descriptions of images, a unique differentiator for accessibility and content indexing.
• GCP restricts general facial recognition by policy; Google Vision API does not return facial identity, only detection and attribute estimation.

5. Natural Language Processing / Text Analytics

NLP services extract structured information and insights from unstructured text, including sentiment, entities, key phrases, classification, and language detection.

AWS — Amazon Comprehend Amazon Comprehend is AWS's NLP service. It provides sentiment analysis (positive/negative/neutral/mixed), named entity recognition (people, places, organizations, dates, quantities), key phrase extraction, language detection, PII detection and redaction, and topic modeling. Comprehend Medical is a purpose-built service for extracting medical conditions, medications, dosages, and procedures from clinical text. Custom classification and custom entity recognition are trained via a managed AutoML pipeline.

Azure — Azure AI Language (Foundry Tools) Azure AI Language consolidates NLP capabilities previously split across Text Analytics, LUIS, and QnA Maker. It provides sentiment analysis and opinion mining, named entity recognition, key phrase extraction, language detection, PII and PHI detection, document and conversation summarization, text classification (custom), custom named entity recognition, and conversational language understanding (CLU, replacing LUIS). Cognitive Search integration enables NLP-enriched search. LUIS was retired in October 2025; CLU is the replacement.

OCI — OCI Language OCI Language provides text classification, named entity recognition, sentiment analysis, key phrase extraction, language detection, personal information detection, and text translation. All capabilities operate at scale via batch APIs against documents stored in OCI Object Storage. Custom model training is available for classification and NER through a managed training pipeline.

GCP — Cloud Natural Language API Cloud Natural Language API provides entity analysis, sentiment analysis, syntax analysis, content classification, and entity sentiment analysis. The API classifies documents into 700+ predefined categories. Vertex AI supports custom text classification and NER model training. Healthcare Natural Language AI provides PHI extraction from clinical notes.

Key differentiators:

• Azure AI Language's opinion mining extracts aspect-level sentiment (e.g., "the battery life is great but the screen is dim"), providing finer granularity than simple document-level sentiment.
• Azure AI Language now unifies conversational NLU (CLU) with analytics NLP in a single service; LUIS retirement in October 2025 makes migration to CLU the only supported path.
• AWS Comprehend Medical and Azure Healthcare Language are the only purpose-built, HIPAA-eligible clinical NLP services among the four providers; GCP Healthcare Natural Language AI is a comparable option.

6. Translation Services

Cloud translation services convert text between languages, with options for neural machine translation (NMT), document-level translation, and custom terminology enforcement.

AWS — Amazon Translate Amazon Translate provides neural machine translation for 75+ languages with real-time and batch translation modes. Custom Terminology allows enforcing domain-specific term translations (e.g., brand names, technical terms). Active Custom Translation enables fine-tuning translation quality on customer-provided parallel corpora. Document translation handles HTML, Word, Excel, and PowerPoint files. Integration with Amazon Comprehend allows detect-then-translate workflows.

Azure — Azure Translator (Foundry Tools) Azure Translator supports 100+ languages with text translation, document translation, and transliteration (script conversion without semantic translation). Custom Translator provides a training interface for fine-tuning NMT models on domain-specific bilingual corpora. The synchronous API supports up to 10,000 characters per request; the Document Translation API handles batch document translation preserving original formatting. Language detection is included. Azure Translator is available as a standalone API or via Azure AI Speech for speech-level translation.

OCI — OCI Language Translation OCI Language includes text translation capabilities supporting major world languages. It integrates with other OCI Language NLP features (sentiment, NER, classification) as part of a unified text processing API. Dedicated custom translation model training is not available as a standalone feature in the same way as AWS Active Custom Translation or Azure Custom Translator; translation is one capability of the OCI Language service rather than a dedicated product.

GCP — Cloud Translation API Cloud Translation API Advanced supports 100+ languages, neural machine translation, document translation (preserving source format for Google Workspace documents, Word, PDF, and PowerPoint files), custom glossaries for domain-specific terminology, and batch translation. AutoML Translation enables training custom NMT models from bilingual corpora. The Translation Hub provides a managed human-in-the-loop translation workflow for enterprise document translation at scale.

Key differentiators:

• Azure Translator and GCP Translation API both support 100+ languages; OCI Language translation lags significantly in language breadth.
• GCP Translation Hub uniquely provides a managed enterprise document translation workflow with human review integration, suited for high-volume regulated content translation.
• Azure Custom Translator and GCP AutoML Translation both offer model fine-tuning on domain-specific parallel corpora, providing higher accuracy than generic NMT in specialized fields.

7. Document AI / Intelligent Document Processing

Intelligent document processing services extract structured data from documents, forms, invoices, receipts, and ID documents, combining OCR with layout analysis and ML-based field extraction.

AWS — Amazon Textract Amazon Textract automatically extracts text, forms (key-value pairs), tables, and query-based data from scanned documents and images. It supports synchronous (single page) and asynchronous (multi-page PDF, TIFF) processing. Queries allow extracting specific fields from documents without training a custom model. Lending AI and Expense Analysis are pre-built document types for mortgage and expense processing. Textract integrates with Amazon Comprehend for NLP enrichment and with Amazon A2I for human review of low-confidence extractions.

Azure — Azure Document Intelligence (Foundry Tools) Azure Document Intelligence (formerly Form Recognizer) provides pre-built models for invoices, receipts, ID documents, W-2 tax forms, business cards, health insurance cards, and contracts. Layout analysis extracts text, tables, and selection marks with structural metadata. Custom models support labeled training for domain-specific document types. The Composed Model allows routing multiple document types to a single endpoint. Document Intelligence Studio provides a visual labeling and testing interface.

OCI — OCI Document Understanding OCI Document Understanding extracts text, tables, key-value pairs, and form data from documents using prebuilt models for invoices, receipts, and passports. Custom model training is supported for domain-specific document layouts. The service processes files stored in OCI Object Storage and returns JSON-structured results. It integrates with OCI Vision for image-level processing and with Oracle Process Automation for document workflow integration.

GCP — Document AI Document AI is Google's managed intelligent document processing service. Pre-built parsers cover invoices, receipts, identity documents, utility statements, W-2 tax forms, payslips, and lending documents (mortgage statements, homeowner insurance). Document AI Workbench enables training custom document parsers. Document AI Warehouse provides a managed document storage, search, and workflow orchestration layer on top of extracted data. Enterprise Document OCR is available as a standalone high-accuracy OCR processor.

Key differentiators:

• AWS Textract's query-based extraction allows targeting specific fields without labeled training data, reducing onboarding time for simple extraction tasks.
• GCP Document AI Warehouse uniquely provides a managed document repository with extraction, storage, search, and workflow capabilities as an integrated offering.
• Azure Document Intelligence covers the widest set of pre-built document types, including specialized financial and healthcare documents, without requiring custom model training.

8. Chatbot / Conversational AI Services

Conversational AI services provide platforms for building, deploying, and managing dialog-driven bots and virtual assistants with intent recognition, entity extraction, and multi-channel deployment.

AWS — Amazon Lex Amazon Lex is a managed conversational AI service using the same deep learning technologies as Alexa. It provides automatic speech recognition (ASR), natural language understanding (NLU), intent classification, slot (entity) extraction, and multi-turn dialog management. Lex deploys to web, mobile, Amazon Connect (contact center), Facebook Messenger, Slack, and Twilio. Visual conversation flow builder is available in the Lex V2 console. Amazon Kendra integration enables RAG-based bot answers from document repositories.

Azure — Azure AI Bot Service / Microsoft Copilot Studio Azure AI Bot Service is the managed deployment platform for bots built with the Bot Framework SDK. It handles channel connectors for Teams, Slack, Facebook, Twilio, and Direct Line. Microsoft Copilot Studio (formerly Power Virtual Agents) is the low-code/no-code authoring environment for conversational AI, deeply integrated with Microsoft 365 and Power Platform. Conversational Language Understanding (CLU, replacing LUIS as of October 2025) provides the NLU layer. Azure OpenAI integration enables generative AI-augmented dialog flows.

OCI — Oracle Digital Assistant Oracle Digital Assistant (ODA) is OCI's conversational AI platform. It provides intent classification, entity extraction, dialog flow design, and multi-channel deployment to web, mobile, voice, Microsoft Teams, Slack, and WhatsApp. ODA's skill-routing architecture allows building multi-skill bots where specialized sub-bots handle individual domains (HR, finance, supply chain). Native integration with Oracle Fusion Applications makes ODA the preferred option for Oracle ERP/HCM/SCM chatbot use cases. OCI Generative AI can be used within ODA for generative response fallback.

GCP — Dialogflow CX / Agent Builder Dialogflow CX is Google's enterprise conversational AI platform, providing flow-based dialog design, state machine-driven conversation management, and built-in testing tools. It is distinct from Dialogflow ES (the older essentials version). Contact Center AI (CCAI) combines Dialogflow CX with Agent Assist (real-time support for human agents) and CCAI Insights (conversation analytics). Vertex AI Agent Builder provides a higher-level RAG and LLM-augmented agent development framework. Gemini 2.5 Flash is available as the NLU backbone in Dialogflow CX.

Key differentiators:

• Oracle Digital Assistant's native skill-routing architecture and Oracle Fusion application integration make it the preferred conversational AI platform for Oracle ERP/HCM customers.
• Microsoft Copilot Studio's deep Power Platform and Microsoft 365 integration gives it a structural advantage in Microsoft-centric enterprises, particularly for Teams-based assistants.
• GCP CCAI provides the most complete contact center AI stack, combining virtual agent (Dialogflow CX), agent assist (real-time coaching), and conversation analytics (CCAI Insights) in a unified product.

9. AutoML

AutoML services automate the selection of model type, feature engineering, hyperparameter tuning, and training, enabling non-expert users to build production-ready ML models.

AWS — SageMaker Autopilot SageMaker Autopilot automatically explores data, selects algorithms, performs feature engineering, and tunes hyperparameters. It supports tabular data classification and regression tasks with Ensembling mode (multiple algorithms, highest accuracy) and HPO mode (single algorithm, hyperparameter search). Autopilot generates explainability reports and human-readable notebooks explaining the model-building decisions, providing transparency into the AutoML process. Models deploy to SageMaker endpoints.

Azure — Azure Machine Learning AutoML Azure ML's AutoML supports classification, regression, time series forecasting, computer vision (image classification, object detection, segmentation), and NLP (text classification, NER). It runs multiple algorithms in parallel, selects the best model, and provides model explanations via the Responsible AI Dashboard. AutoML integrates with Azure ML Pipelines for operationalization. Compute resources must be configured before running AutoML jobs (managed compute clusters or serverless compute).

OCI — OCI Data Science AutoML / Oracle Machine Learning AutoML OCI provides AutoML capabilities through two paths. OCI Data Science's Accelerated Data Science (ADS) SDK includes an AutoML operator for tabular classification and regression with automated algorithm selection and hyperparameter tuning. Oracle Machine Learning (OML) on Autonomous Database provides a no-code AutoML UI that operates entirely in-database, selecting algorithms, handling feature sampling, and tuning without data movement. OML AutoML supports 12+ algorithms for classification, regression, and anomaly detection.

GCP — Vertex AI AutoML Vertex AI AutoML supports tabular data (classification, regression, forecasting), image (classification, object detection, segmentation), video (classification, object tracking, action recognition), and text (classification, extraction, sentiment). It is fully managed, with Google handling infrastructure, training, and hyperparameter search. AutoML models deploy directly to Vertex AI Endpoints or can be exported for edge deployment. No ML expertise is required; the service is designed for data analysts and domain experts.

Key differentiators:

• GCP Vertex AI AutoML covers the widest range of data modalities (tabular, image, video, text) in a single unified AutoML service.
• OCI OML AutoML is the only provider with a production-ready in-database AutoML that operates entirely within the database engine, eliminating data movement.
• Azure ML AutoML's integrated model explainability and Responsible AI Dashboard make it the strongest choice for regulated industries requiring model transparency documentation.

10. AI Infrastructure

AI infrastructure encompasses the specialized compute instances, custom silicon, and clustered networking required for large-scale model training and inference.

AWS — AI Compute (Trainium, Inferentia, GPU Instances) AWS offers three classes of AI compute. NVIDIA GPU instances include P5 (H100 SXM5), P4d (A100), G5 (A10G), and G6 (L40S) families. AWS Trainium (Trn1/Trn2) is a custom AWS-designed ML training chip. Trainium2 delivers 2.52 petaFLOPS FP8 per chip with 144 GB HBM3e; Project Rainier (activated 2025) deployed nearly 500,000 Trainium2 chips for training Anthropic's Claude models. AWS Inferentia (Inf1/Inf2) is a custom inference chip providing up to 4x higher throughput and 10x lower latency than Inferentia1. SageMaker HyperPod provides managed distributed training clusters with fault tolerance and automatic job recovery. UltraCluster networking provides non-blocking 3200 Gbps cluster interconnects.

Azure — AI Compute (GPU VMs, NDv5 H100, NVv5) Azure GPU instances span the ND-series (compute-optimized for training) and NV-series (visualization and lighter inference). NDv5 (Standard_ND96isr_H100_v5) provides 8x H100 SXM5 GPUs with 3.2 Tbps InfiniBand HDR interconnects, scaling near-linearly to 1,024 H100 GPUs. NDv4 (A100) and ND H200 v5 instances are available for training workloads. Azure does not produce proprietary ML silicon; all accelerators are NVIDIA-based. Azure Managed Prometheus on VM/VMSS (preview 2025) provides native GPU metrics collection. Azure AI infrastructure is deeply integrated with Microsoft's partnership with NVIDIA, enabling DGX Cloud deployments.

OCI — AI Infrastructure (GPU Bare Metal, OCI Supercluster) OCI differentiates on GPU infrastructure scale and bare metal access. Available GPU options include NVIDIA B200 (Blackwell), H200, H100, A100, L40S, and A10 on bare metal and VM shapes. OCI Supercluster scales to 65,536 H200 GPUs (260 ExaFLOPS) or 131,072 GPUs in the zettascale configuration. All inter-GPU communication uses RDMA-based cluster networking at 3.2 Tbps per node, providing microsecond-level latency. B200 bare metal is generally available in 2025, offering up to 30x faster inference and 4x faster training than H100. OCI does not have proprietary ML silicon but leads on raw cluster scale and bare metal GPU access with RDMA networking.

GCP — AI Infrastructure (TPUs, A100/H100/H200 GPUs, Hypercomputer) Google Cloud uniquely offers proprietary TPU (Tensor Processing Unit) hardware unavailable from any other cloud provider. TPU v5e and v5p are broadly available for cost-efficient training and inference. Ironwood (TPU v7), introduced at Cloud Next 2025, is designed for inference at scale: 192 GB HBM3E per chip, >4x performance per chip versus Trillium (v6), 9,216-chip pods delivering 42.5 ExaFLOPS FP8. GPU options include A3 (H100), A4 (H200), and A4X (B200) VM families. The AI Hypercomputer integrates TPUs, GPUs, and Axion (custom Arm CPU) with high-bandwidth cluster networking for distributed training and serving. Axion processors provide 50% better performance and 60% better energy efficiency versus x86 equivalents.

Key differentiators:

• GCP is the only cloud provider with proprietary TPU hardware. The Ironwood (TPU v7) is purpose-designed for inference workloads with 4x performance per chip over Trillium, with no equivalent from AWS, Azure, or OCI.
• OCI leads in maximum raw cluster GPU scale, offering up to 131,072 GPUs in a single interconnected cluster with bare metal access and RDMA networking — the highest published cluster ceiling.
• AWS Trainium2 (via SageMaker HyperPod) provides a vertically integrated training solution with managed fault tolerance and automatic job recovery, reducing operational overhead for large distributed training runs.
• Azure NDv5 offers the deepest NVIDIA partnership with DGX Cloud deployments and InfiniBand HDR interconnects, but tops out at smaller documented cluster sizes than OCI or GCP TPU pods.

11. Pre-trained Model APIs

Pre-trained model APIs provide immediate access to hosted AI capabilities without any model training, suitable for rapid integration into applications.

AWS offers the broadest API catalog through individual services (Rekognition, Comprehend, Textract, Translate, Polly, Transcribe) plus Bedrock-hosted models for generative tasks. Azure consolidates most APIs under the Azure AI Services umbrella (Vision, Language, Speech, Document Intelligence, Translator) with Azure OpenAI providing GPT and DALL-E endpoints. OCI covers core categories (Vision, Speech, Language, Document Understanding, Anomaly Detection) but has gaps in content moderation, image generation, and video analysis. GCP provides APIs across Vision, Natural Language, Speech, Translation, Video Intelligence, and Document AI, with Vertex AI hosting Gemini for generative tasks.

Summary: Provider Positioning

References

• Amazon SageMaker AI
• Amazon Bedrock
• Amazon Rekognition
• Amazon Transcribe
• Amazon Polly
• Amazon Comprehend
• Amazon Translate
• Amazon Textract
• Amazon Lex
• SageMaker Autopilot
• AWS Trainium
• AWS Inferentia
• Azure AI Foundry Tools
• Azure OpenAI in Foundry Models
• Azure Machine Learning
• Azure AI Speech
• Azure Translator
• Azure AI Bot Service
• OCI AI Services
• OCI Data Science
• OCI Generative AI Service
• OCI GPU and AI Infrastructure
• OCI AI Infrastructure
• Vertex AI Platform
• Cloud Speech-to-Text
• Cloud Text-to-Speech
• Cloud Translation API
• Cloud TPUs
• Dialogflow CX
• Document AI