Review: AWS AI and Machine Learning stacks up, and up

Amazon Web Services provides an impressively broad and deep set of machine learning and AI services, rivaling Google Cloud and Microsoft Azure.

1 2 3 4 Page 4
Page 4 of 4

Amazon Transcribe Medical

Amazon Transcribe Medical is an automatic speech recognition service that lets you add medical speech-to-text capabilities to your voice-enabled applications. It provides accurate speech-to-text for use cases such as voice scribes for clinical documentation, call analytics in pharmacovigilance, subtitling, and accessibility during telehealth sessions. Transcribe Medical supports transcription for different specialties, which you can specify when you start a stream. There are different settings for clinician-patient dialogue and post medical encounter dictation.

AWS provides a broad and deep portfolio of machine learning infrastructure services with a choice of processors and accelerators to meet your unique performance and budget needs. Amazon EC2 P4d instances provide the highest performance for machine learning training in the cloud with the latest Nvidia A100 Tensor Core GPUs coupled with 400 Gbps instance networking. P4d instances are deployed in hyperscale clusters, called EC2 UltraClusters, offering supercomputer-class performance for the most complex machine learning training jobs. For inference, Amazon EC2 Inf1 instances, powered by AWS Inferentia chips, provide high-performance and low-cost inference.

You can choose from TensorFlow, PyTorch, Apache MXNet, and other popular frameworks to experiment with and customize machine learning algorithms. You can use the framework of your choice as a managed experience in Amazon SageMaker, or use the AWS Deep Learning AMIs (Amazon machine images), which are fully configured with the latest versions of the most popular deep learning frameworks and tools.

Choosing AWS for AI and machine learning

As we’ve seen, Amazon Web Services provides a broad and deep set of machine learning and AI services. AWS AI and Machine Learning is competitive with Google Cloud AI and Machine Learning and Microsoft Azure AI and Machine Learning in the areas of ready-to-use AI services, AI service customization, data science in Jupyter notebooks, and infrastructure. There are a few areas of differentiation, however.

For example, AWS recently announced some interesting new initiatives for industrial and medical AI services, although most of them are still in preview. AWS does have more data centers in place than either Google Cloud or Microsoft Azure, but for data scientists in most populated first-world locations that doesn’t really matter.

AWS has the new SageMaker Data Wrangler to help with the first steps in model building. Google uses a third-party product for this. Microsoft Azure has an easy drag-and-drop pipeline builder, and also has an integrated Databricks implementation.

Right now Microsoft Azure seems to be a little better than AWS or Google Cloud for creating new vision models with transfer learning because it needs fewer training images to get to the same accuracy and recall, and seems to be a little ahead in the area of Responsible AI. I have no idea whether those differences will last long. AWS is a little bit behind Google Cloud in the number of languages it can translate, and behind both Microsoft Azure and Google Cloud in the number of languages that it supports for text to speech. But what matters there is whether the languages you need are supported, not the overall count.

Microsoft Azure allows users to run some AI services in containers on-premises, which can be a cheap way to bring the services to the data and maintain data security. AWS offers AWS Outposts servers you can run in your own data centers, but those are currently big and expensive; the planned 1U and 2U form factors are promised for later in 2021.

Overall, however, these distinctions are minor and likely to change over time. Right now, you can’t go far wrong using machine learning and AI services from any of the big three cloud services.


Amazon SageMaker: Free tier, 2 months; notebooks, $0.05 to $6.45/hour depending on instance size; training clusters additional.

Amazon Kendra: Free tier, 750 hours for the first 30 days; Developer Edition, $2.50/hour; Enterprise Edition, $7/hour.

Amazon Personalize: Free tier, 2 months; ingestion ($0.05 per GB), training ($0.24 per training hour), and inference ($0.20 per TPS-hour) are charged separately.

AWS Contact Center Intelligence: Contact AWS sales or an AWS partner.

Amazon Comprehend: Free tier, 50K units of text (5M characters); $0.0001/unit/service with volume discounts. A unit is 100 characters, with a 3 unit minimum per request.

Amazon Textract: Free tier, 3 months, limited to 1K pages per month using the Detect Document Text API and 100 pages per month using the Analyze Document API; after that $1.50 to $65/1K pages with quantity discount.

Amazon Translate: Free tier, 2 million characters per month for 12 months; standard $15/M characters; active custom $60/M characters.

Amazon Lookout for Metrics: Currently in free preview. After that, Free tier will allow tracking 100 metrics for one month. Production costs will be $0.75/metric/month with volume discounts.

Amazon Forecast: Free tier, two months, limit 10K forecasts per month; after that $0.60/1K forecasts, $0.088/GB storage, and $0.24 per training hour.

Amazon Fraud Detector: Free tier, 2 months, with some usage limits; model training $0.39/hour, model hosting $0.06/hour, predictions $0.03 each with volume discounts.

Amazon Lookout for Vision: Free tier, 3 months, limited to 10 training hours and 4 inference hours per month; after that $2.00 per training hour and $4.00 per inference hour with volume discounts.

AWS Panorama: AWS Panorama Appliance Developer Kit, $2,499 per device; AWS Panorama Appliance, $4,000 per device, $8.33 per month per active camera stream.

Amazon Rekognition: Free tier, 12 months, 5K images a month; image analysis, $1/1K API calls with volume discounts; face metadata storage, $0.00001/face per month; video analysis, $0.10/API/minute.

Amazon Polly: Free tier, 12 months, 5M characters a month; standard voices, $4/M characters; neural voices, $16/M characters.

Amazon Transcribe: Free tier, 12 months, 60 audio minutes per month; after that, $0.024/minute with volume discounts; Transcribe Medical $0.075 per minute.

Amazon Lex: Free tier, 12 months, 10K text requests and 5K speech requests or speech intervals per month; after that, $4/1K speech requests and $0.75/1K text requests.

Amazon DevOps Guru: $0.40 per 10K API Calls, plus $0.0028 to $0.0042 per resource per hour analyzed, plus any SNS or System manager notifications used.

Amazon CodeGuru: 90 days free; Reviewer, $0.50 per 100 lines of repository code analyzed with a volume discount, plus $0.75 per 100 lines of pull request code analyzed; Profiler, $0.005 per sampling hour for the first 36,000 sampling hours per profiling group per month, free after that.

Amazon Monitron: Starter Kit (5 Sensors, 1 Gateway), $715; Service, $50 per sensor per year.

Amazon Lookout for Equipment: 1 month free; Data Ingestion, $0.20 per GB; Model Training, $0.24 per training hour; Scheduled Inference, $0.25 per inference hour.

Amazon Healthlake: $0.27 per Data Store hour, and your first 10 GB of storage and 3,500 FHIR queries per hour are included for free. Additional charges: data storage, $0.25/GB/month; queries, $0.015/per 10K queries, per hour; medical NLP, $0.001 per 100 characters; FHIR export, $0.19/GB.

Amazon Comprehend Medical: Free tier, 3 months, 25K units of text (2.5M characters); Medical Named Entity and Relationship Extraction (NERe) API, $0.01/unit; Medical Protected Health Information Data Extraction and Identification (PHId) API, $0.0014/unit; Medical ICD-10-CM Ontology Linking API, $0.0005/unit; Medical RxNorm Ontology Linking API, $0.00025/unit.


Hosted on Amazon Web Services; dedicated on-premises hardware options available for sale or lease.

At a Glance
  • Amazon Web Services has a broad and deep set of machine learning and AI services. AWS is competitive with Google Cloud AI and Microsoft Azure AI and Machine Learning in the areas of ready-to-use AI services, AI service customization, data science in Jupyter notebooks, and infrastructure.


    • Wide and deep offerings of AI services
    • Expanded ecosystem for data scientists in SageMaker Studio
    • Wide range of infrastructure for deep learning and inference
    • Can compile your custom models for use in edge devices


    • To run AI services on-premise, you need to install expensive AWS Outposts hardware
    • Fewer language translation and text-to-speech choices than competitors

Copyright © 2021 IDG Communications, Inc.

1 2 3 4 Page 4
Page 4 of 4
How to choose a low-code development platform