Routes for machine learning startups
| Route | Max credits | Who qualifies | Status |
|---|---|---|---|
| AWS Activate Founders | $1,000 | Self-funded, live website, no prior Activate credits | Open |
| Activate via a Provider | Up to $100,000 | Backed by an accelerator, VC or incubator in the Activate network | Open |
| AWS Credits for AI Startups | $200k+ | AI | Invite-only, listed now |
| Generative AI Accelerator | Up to $1M | Generative AI, pre-Series A with an MVP | Page shows the 2024 window; current cohort unconfirmed |
| Our partner match | Same as AWS | Any machine learning startup; we suggest the best route | Open |
Amounts are programme ceilings, not what every company gets. Approval is never guaranteed.
What credits cover, and what they don't
- Compute: EC2, Lambda, ECS, Fargate
- Databases: RDS, Aurora, DynamoDB
- Storage: S3, EBS
- AI models: Amazon Bedrock on-demand usage
- Networking: CloudFront, data transfer
- Most first-party AWS services
- AWS Marketplace third-party software
- Route 53 domain name registration
- Upfront Reserved Instances and Savings Plans
- Business or Enterprise Support plans
- Taxes added to your bill
- Mining or other prohibited use
AWS programmes for machine learning
- Open
AWS Credits for AI Startups
$200k+
AI. Invite-only, listed now.
Source: AWS credits page
- Status unconfirmed
Generative AI Accelerator
Up to $1M
Generative AI, pre-Series A with an MVP. Page shows the 2024 window; current cohort unconfirmed.
Source: AWS
Does a machine learning company qualify?
- Product startups (software or hardware you sell) qualify for Activate. Agencies, consultancies and dev shops that bill by the hour usually don't, unless they're building their own product.
- Your company is usually under 10 years old, pre-Series B for the Provider tier, and hasn't had Activate credits before.
- Past the startup stage? Established machine learning companies can look at business and migration funding instead.
- Nonprofit or researcher? The AWS Imagine Grant and Cloud Credit for Research are separate from Activate.
Before you apply: readiness checklist
Most rejections come from fixable paperwork. Tick what you have ready: 0 of 7 done.
What credits buy for a machine learning product
Amazon Bedrock or GPU EC2 for models, S3 for data, SageMaker for training, a small API on Lambda or ECS
MVP
$800/mo
Bedrock$290
EC2 (GPU)$175
SageMaker$110
S3$80
Lambda$50
Data transfer, logs & support$95
Runway
$5,0006.3 mo
$25,00031 mo
$100,000125 mo
Seed
$5,000/mo
Bedrock$1,800
EC2 (GPU)$1,100
SageMaker$700
S3$500
Lambda$300
Data transfer, logs & support$600
Runway
$5,0001.0 mo
$25,0005.0 mo
$100,00020 mo
Growth
$25,000/mo
Bedrock$9,000
EC2 (GPU)$5,500
SageMaker$3,500
S3$2,500
Lambda$1,500
Data transfer, logs & support$3,000
Runway
$5,0000.2 mo
$25,0001.0 mo
$100,0004.0 mo
Planning estimates based on us-east-1 On-Demand list prices for the typical stack above, including storage, data transfer and support. Not an AWS quote; your bill depends on region, traffic and commitments. Credits usually expire after 12 to 24 months.
Estimate your machine learning bill
AI costs scale with usage, not servers. Enter your numbers to see the monthly bill and how long credits last.
Model tokens (300k req/mo)$840
GPU time$0
API, storage, logs$60
Monthly total$900
Credits last
$1,0001.1 mo
$25,00028 mo
$100,00099+ mo
$200,00099+ mo
Bedrock on-demand and EC2 Linux on-demand list prices, us-east-1, as listed on the AWS pricing page. Excludes prompt caching and batch discounts. Credits usually expire in 12 to 24 months, so long runways are capped by expiry.
Where machine learning startups burn credits fastest
Idle GPU instances
A single p4d.24xlarge left running over a weekend costs around $1,500. Experiment boxes are often started and forgotten.
Fix: Use SageMaker training jobs or Spot that stop on finish, and set a budget alarm per GPU instance.
Frontier model by default
Output tokens on top-tier Bedrock models cost 20 to 100 times more than small models.
Fix: Route simple tasks to Nova Micro/Lite or Haiku and reserve large models for hard prompts.
Full-table scans in Athena and Redshift
Athena charges $5 per TB scanned. Unpartitioned CSV data means every query reads everything.
Fix: Store data as partitioned Parquet and use workgroup scan limits.
Always-on inference endpoints
A SageMaker real-time endpoint on a GPU bills 24/7, even with no traffic at night.
Fix: Use serverless or asynchronous inference, or scale endpoints to zero.
Cross-AZ data transfer
Moving datasets and checkpoints between Availability Zones costs $0.01 per GB each way.
Fix: Keep training data, compute and checkpoints in one AZ.
Unbounded CloudWatch logs
Debug-level logs from every request are ingested at about $0.50 per GB and kept forever by default.
Fix: Set log retention to 14 or 30 days and lower log levels in production.
How a machine learning request flows through AWS
- 0Your users
- 1Amazon Bedrock
- 2SageMaker training jobs
- 3Amazon S3
- 4Amazon Athena or Redshift Serverless
- 5Amazon Kinesis Data Streams or MSK Serverless
- 6OpenSearch Serverless or Aurora pgvector
Reference AWS setup for machine learning
- ModelsAmazon Bedrock · Claude, Llama, Mistral and Nova through one API, paid per tokenModel access is requested per model
- TrainingSageMaker training jobs (g5, g6, p5) · Pay only while the job runs; Spot for up to ~70% off
- LakeAmazon S3 + AWS Glue Data Catalog · Cheap storage with a schema layer on top
- QueryAmazon Athena or Redshift Serverless · SQL over the lake without running a clusterRedshift Serverless bills a minimum of base RPUs while active
- StreamingAmazon Kinesis Data Streams or MSK Serverless · Event and clickstream ingestion
- Vector searchOpenSearch Serverless or Aurora pgvector · Retrieval for RAG and similarity searchOpenSearch Serverless has a minimum monthly charge
- APILambda + API Gateway · Thin, scale-to-zero API layer
Accelerators and investors for machine learning
The bigger credit tiers usually come through a programme or investor. These are active in machine learning; check each one's current AWS offer directly.
- Accelerator
Large share of recent batches are AI and data companies
- Startup programme
Free programme for AI startups; GPU discounts and partner cloud offers
- Grant / accelerator
Pre-seed AI product companies; compute credits from partner clouds
- Accelerator
AI-focused programmes in several cities; Activate network member
- Seed VC
Data infrastructure, ML tooling and developer platforms
Startups like you
Companies in this space named in AWS programme cohorts. Programme maximums are not what each company received.
- LlamaIndex
- Cartesia
- Hedra
- Pathway
- Archetype AI
- Reality Defender
- Relevance AI
- Marqo
- Smallest AI
- Lettria
AWS vs Google Cloud vs Azure for machine learning
AWS Activate
$1,000 to $100,000+
Widest service range, largest partner and provider network
Google for Startups Cloud
Up to $200,000 ($350,000 for AI)
BigQuery, Vertex AI and Gemini
Microsoft for Startups
Up to $150,000 Azure
Azure OpenAI, Microsoft 365 and enterprise sales channels
All three now fund AI heavily. Google often offers the largest AI pot; AWS gives Bedrock access to many model families; Azure is the home of OpenAI models.
Ceilings are programme maximums and change often. Check each provider before applying.
Rules and gotchas for machine learning
- Credits for AI Startups is invite-only
- The larger AI credit pot is offered through Activate Providers and AWS account teams, not a public form.
- GPU capacity isn't guaranteed
- Credits don't reserve p5 or p4d capacity. Big runs may need EC2 Capacity Blocks bought in advance.
- Marketplace items aren't covered
- Third-party models and data sets bought through AWS Marketplace are usually not paid by promotional credits.
Related niches
Questions
Is there a special AWS credit programme for machine learning startups?
Yes: AWS Credits for AI Startups ($200k+; Invite-only, listed now); Generative AI Accelerator (Up to $1M; Page shows the 2024 window; current cohort unconfirmed). Standard AWS Activate also applies.
How much in AWS credits can a machine learning startup get?
The largest pot is AWS Credits for AI Startups: $200k+. Otherwise Activate gives up to $200,000 through a provider.
Do AWS credits cover what a machine learning product runs on?
Usually yes. A typical setup (Amazon Bedrock or GPU EC2 for models, S3 for data, SageMaker for training, a small API on Lambda or ECS) is covered by promotional credits. Marketplace purchases, support plans and taxes are not.
How long will credits last for a machine learning startup?
At about $800/month (MVP), $25,000 lasts around 31 months, though credits usually expire after 12 to 24 months.
What if AWS rejects my machine learning application?
Try a different route: an Activate Provider, another niche programme, Google for Startups Cloud Program (AI track) or Microsoft for Startups Founders Hub (Azure OpenAI credits), or our partner match.
Find your best route for machine learning
Answer four questions and we'll match you with a partner who knows machine learning. Free, and the credits are the same as going direct.
Cloud Credits is independent and not affiliated with, endorsed by or sponsored by Amazon Web Services. AWS and AWS Activate are trademarks of Amazon.com, Inc. or its affiliates. Data last checked 3 October 2026; check AWS for current terms.