Exams / AIP-C01 / Study guide / Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones)
Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones): AIP-C01 study notes and practice questions
What to know about Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones) for the AIP-C01 exam, from the Implementation and Integration domain (26% of the exam guide), then 1 free question to test it. Part of the AIP-C01 study guide.
AIP-C01 scenarios name a constraint: data cannot leave the facility, a mobile feature needs replies in very little time, or a plant has unreliable connectivity. The correct answer satisfies that constraint with the least new infrastructure.
Data must stay on premises
AWS Outposts runs AWS compute and storage in your own data center. When prompts and documents cannot leave the building, host an open-weight model on GPU instances in Outposts, with the vector store on the same hardware. Bedrock is a Regional service, so a call to it still sends data to the Region. You also take on model serving, scaling and patching.
Private connectivity from on premises
When data may leave the building but must not cross the public internet, connect over AWS Direct Connect or Site-to-Site VPN into a VPC, then use interface VPC endpoints (AWS PrivateLink) for Bedrock. Traffic stays on the AWS network.
Latency at the edge
AWS Wavelength places compute inside telecom 5G networks, and Local Zones place it in a metro area. Both suit very low latency for nearby users. They offer a subset of Regional services, so a common pattern runs a small model or preprocessing at the edge and calls a full-size model in the Region when the latency budget allows.
Split pipelines
Run detection and redaction close to the data, on premises or in country. Use Amazon Comprehend PII detection, a small local model, or a sensitive information filter applied with the ApplyGuardrail API. Only sanitized text then goes to the cloud model. Check that logs and retrieval indexes follow the same rule.
Residency inside the cloud
Geographic cross-Region inference profiles, such as an EU profile, keep processing inside one geography. Global profiles can route to any supported commercial Region, so they break residency rules. Pair the profile choice with a Region restriction in your service control policies.
How to choose
Scenario wording
Choose
“Prompts and documents cannot leave our data center”
Open-weight model on GPU instances in Outposts
“Mobile app needs fast replies for users in one city”
Local Zone or Wavelength Zone for the front part
“On-premises calls must avoid the internet”
Direct Connect or VPN plus Bedrock interface endpoint
“EU customer data must be processed only in the EU”
EU geographic inference profile
“Account numbers must not reach the cloud”
Redact on premises, send only sanitized text
“Small fast model locally, large model when needed”
Edge model with fallback call to Bedrock
Common traps
Treating Bedrock as on premises because it is a managed service. Use Outposts-hosted models when data may not leave.
Global cross-Region inference for data bound to one geography.
Wavelength for a workload that needs a full set of Regional services, which Wavelength does not provide.
Sending raw records to the cloud and planning to redact afterward.
Public internet paths for on-premises calls when PrivateLink would keep traffic private.
A manufacturing company inspects finished parts with cameras on its factory floor. Photos of the parts are subject to a customer contract that forbids them from leaving the plant's network.
The company wants a generative model that writes a short inspection note from each photo, and each note must appear within two seconds. The plant has a small server room and a reliable link to AWS for dashboards and reporting. Plant managers want the notes to appear on the line terminals, so the design must also work when the internet link is down.
Which solution meets these requirements?
Choose one answer.
AA. Install an AWS Outposts rack in the plant, and run an open-weight vision model on GPU instances in the rack that the line terminals call locally.
BB. Upload each photo to an Amazon S3 bucket in the nearest AWS Region over the plant's VPN, and call Amazon Bedrock from a Lambda function in that Region.
CC. Deploy the model in an AWS Wavelength Zone in the nearest telecom network, so that the photos travel only over the carrier's private 5G network.
DD. Use Amazon Bedrock with a geographic cross-Region inference profile for the plant's country, so that processing stays within that country's Regions.
Show the answer and explanation
Answer: A
A. Right: Outposts runs AWS infrastructure on premises, so the photos and the model stay inside the plant network. GPU instances in the rack can meet the two-second note target while the reporting link carries only summaries. Learn more: AWS Outposts (opens in a new tab)
B. Wrong: Every photo would still leave the plant network and travel to a Region, which breaks the contract. A VPN protects the traffic in transit but does not keep the data inside the plant.
C. Wrong: Wavelength reduces latency for mobile devices, but the photos still leave the factory and cross a carrier network. That does not meet a rule that the photos never leave the plant network.
D. Wrong: A geographic profile keeps processing in one country, which meets a country residency rule. The contract, however, requires the photos to stay inside the plant network, and this design sends them to a Region.
Explanation
Outposts brings AWS infrastructure into the plant, so the photos and the model stay on the local network while the company still reports to AWS. Designs that send photos to a Region, a Wavelength Zone or a country profile all move the images outside the plant, which the contract forbids.
Want more on Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones)? The 5 full-length practice sets have 7 more Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones) questions, 375 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →
Study Hybrid and edge GenAI for data residency (Outposts, Wavelength, Local Zones)