This guide is for tech leads and senior engineers who deploy an application into a VPC or virtual network that a platform team may or may not own. It sits under Netbase's cloud platform engineering work, and the checks belong in the pipelines of DevOps and platform engineering.
In this guide
- What does an application team actually decide in a cloud network?
- Which six mistakes cause most VPC and private endpoint problems?
- Gateway endpoint, private endpoint or NAT: which path fits each destination?
- The pre-go-live network review in order
- When should you not add private endpoints or central egress inspection?
- Where AI fits in network design reviews
- What delivery record exists, and what does not
- Alternatives: who owns the application network?
- What are the limits of this guide?
- Frequently asked questions
- Next step
What does an application team actually decide in a cloud network?
An application team usually decides four things inside the network it is given: subnet sizes, how private workloads reach provider services, how they reach the internet, and which names resolve to private addresses. The platform team decides the address plan and topology; the landing zone checklist covers that layer.
The three providers use different names for the same building blocks:
-
Private network
- AWS
- VPC
- Azure
- Virtual network
- Google Cloud
- VPC network
-
Private access to provider and partner services
- AWS
- Gateway endpoint or interface endpoint (PrivateLink)
- Azure
- Private Endpoint (Private Link)
- Google Cloud
- Private Service Connect endpoint
-
Managed outbound address translation
- AWS
- NAT gateway
- Azure
- Azure NAT Gateway
- Google Cloud
- Cloud NAT
All three managed NAT services pass outbound connections and their replies only: neither Cloud NAT nor Azure NAT Gateway accepts unsolicited inbound connections.
Which six mistakes cause most VPC and private endpoint problems?
Six mistakes recur because each works in a test environment and fails only under production load, a zone failure or a new connection.
-
Overlapping or undersized address ranges
Peering or a hybrid link cannot be created; subnets run out of addresses
- Check before go-live
- Compare every range with the central address plan
- Fix
- Take ranges from the plan, with room to grow
-
One NAT gateway for every zone
A single zone outage removes internet access from all zones
- Check before go-live
- List NAT gateways per zone and each subnet's route
- Fix
- One NAT gateway per zone, or a zone-redundant service
-
Provider traffic sent through NAT
High NAT data processing for storage and registry traffic
- Check before go-live
- Check route tables for storage and database endpoint routes
- Fix
- Gateway endpoints for storage and key-value services
-
Private endpoint without private DNS
The application still reaches the public address, or name lookups fail
- Check before go-live
- Resolve each service name from every calling network
- Fix
- Private DNS zones linked to every calling network
-
Incomplete endpoint set in a private network
Containers stay pending; logs never arrive
- Check before go-live
- Start a workload in a subnet with no internet route
- Fix
- Add every endpoint the platform depends on
-
Unplanned egress
Outbound calls fail, or data can leave anywhere
- Check before go-live
- List every outbound destination
- Fix
- An explicit egress path with an allow list
Mistake 1: address ranges that overlap or cannot grow
On AWS a VPC block must sit between a /16 and a /28 netmask, and the size of an existing block cannot be increased or decreased; you can only add secondary blocks, which must not overlap a peered network. AWS also warns that some of its services use 172.17.0.0/16. Ask the platform team for ranges instead of taking a console default.
Mistake 2: one NAT gateway for every zone
AWS creates each NAT gateway in one Availability Zone. If resources in several zones share one gateway and that zone goes down, the other zones lose internet access; AWS recommends a gateway per zone with routes that keep traffic in its own zone. On Azure the Standard NAT gateway is zonal, while StandardV2 operates across all zones of a region.
Mistake 3: provider traffic sent through NAT
A private subnet without an endpoint reaches Amazon S3 or DynamoDB through a NAT device. Gateway endpoints route that traffic privately and carry no additional charge, but only subnets whose route tables are associated with the endpoint use it. Traffic to the service in another region still takes the internet path.
Mistake 4: private endpoints without working private DNS
A private endpoint is only used if the service name resolves to its private address. Microsoft states that an existing public DNS configuration must be overridden to connect through a private endpoint, usually with a private DNS zone linked to the network. AWS notes that VPC endpoints support only the provider's own DNS; a custom resolver needs conditional forwarding.
Mistake 5: an incomplete endpoint set in a fully private network
A workload with no internet route needs every endpoint its platform calls, not only the obvious one. AWS documents that Amazon ECS tasks pulling from Amazon ECR need the ECR endpoints and an S3 gateway endpoint, because image layers are stored in S3, and that Fargate tasks sending logs need a CloudWatch Logs endpoint too.
Mistake 6: egress nobody planned
Microsoft states that since 31 March 2026 new Azure virtual networks default to private subnets without default outbound access, so a workload needs an explicit outbound method such as a NAT gateway. On AWS, IPv6 addresses are public by default, and an egress-only internet gateway allows outbound IPv6 while blocking connections started from the internet. A NAT gateway takes no security group, so outbound rules sit on the workloads.
Gateway endpoint, private endpoint or NAT: which path fits each destination?
Choose the path per destination, not per network. Google's Private Service Connect lets consumers reach Google APIs and published services with their own internal addresses, so the same decision applies on all three clouds.
| Criterion | Gateway endpoint (AWS) | Private or interface endpoint | NAT gateway |
|---|---|---|---|
| Destinations | Amazon S3 and DynamoDB in the same region | One provider, partner or own service each | Any internet address |
| How traffic reaches it | Route table entry with the service prefix list | Private address and private DNS name | Default route of the subnet |
| Cost model | No additional charge | Charged per endpoint and data, see provider pricing | Charged per gateway and data, see provider pricing |
| Choose it when | Private subnets read or write S3 or DynamoDB | Data must stay private, or the service is used from other networks | The destination has no endpoint, such as a third-party API or package mirror |
The pre-go-live network review in order
-
List the destinations
Write down every service, API and mirror the application calls.
-
Check the address plan
Confirm each range came from the central plan and leaves room for more subnets.
-
Map each destination to a path
Use the table above: gateway endpoint, private endpoint or NAT.
-
Test name resolution
Resolve every private service name from each calling network.
-
Start a workload with no internet route
A container or function that starts and logs proves the endpoint set is complete.
-
Fail a zone on paper
Follow each subnet's route to its NAT gateway and confirm a zone outage leaves the others working.
-
Write the egress rule
Record allowed destinations and alert on anything else.
When should you not add private endpoints or central egress inspection?
Do not create a private endpoint for every provider service on day one. Each one adds an address, DNS records and a charge; for a low-volume service with no data sensitivity, a NAT path with outbound rules is often enough.
Do not route one small application through a central inspection firewall nobody operates. Central egress inspection pays off when several workloads share it and someone reviews its logs.
Where AI fits in network design reviews
AI helps with reading configuration and documentation, not with deciding the design. An assistant can compare route tables and DNS zones against the destination list, explain a failed connection from flow logs and draft endpoint modules from a reviewed table. Netbase works with the major commercial and open-source AI tools and models, chosen per project.
Engineers still own the address plan, approve every route and DNS change and apply it through the pipeline; no model gets rights to change networks, routes or firewall rules.
- In delivered work
MLOps pipeline
An MLOps pipeline Netbase delivered for a client that is not named; an AI delivery record, not a network design.
- Available capability
AI agents in operations
AI-assisted network configuration review is a way Netbase approaches the work, not yet tied to a published case.
What delivery record exists, and what does not
Netbase names AWS, Google Cloud, DigitalOcean and Cloudflare as the cloud providers it uses. It also works as offshore development partner on a multi-tenant cloud ERP delivered as SaaS to small and mid-sized businesses for a US client, not named.
Delivery runs in a six-step lifecycle with weekly reviews and KPI dashboards, and most Netbase projects are delivered on fixed-price contracts. No published Netbase record describes a network design in detail, no Azure project is claimed, and no duration, endpoint count or result figure is given here.
Alternatives: who owns the application network?
| Ownership model | Fits when | Selection criterion |
|---|---|---|
| Platform team owns a shared network; application teams use subnets | Several teams, one platform team | The platform team can turn endpoint and egress requests around fast |
| Each application owns its network inside a landing zone | Few teams, strong engineers per team | Address ranges and DNS still come from one central plan |
| Single network for one application | One team, one product | You accept a move later if a second team arrives |
When the network is part of a move, the switch itself follows a cutover plan with rollback points, within Netbase's cloud migration work.
What are the limits of this guide?
- It covers application-level design, not firewall products, hybrid routing or compliance controls.
- Provider names, defaults and charges change; check the current documentation and pricing before you build.
Plan the next step with a Netbase consultant
Frequently asked questions
Taking an address range without checking the central plan is the most common one. It works until the network must be peered or linked to an office, and an existing range cannot be resized, so the fix means new subnets or a rebuilt network.
Usually yes. Private endpoints cover provider and partner services that offer them; third-party APIs, package mirrors and other internet destinations still need an outbound path such as a NAT gateway with outbound rules.
Name resolution is the usual cause. The service name must resolve to the endpoint's private address in every network that calls it, which needs a linked private DNS zone or conditional forwarding from your own resolver.
Next step
If you want a review of your application's network design before go-live, book a solution review with Netbase.
Related services and solutions
AI-ready cloud platform engineering for SaaS and commerce workloads Netbase provides cloud platform engineering for SaaS and commerce teams that need a secure, repeatable home for their software and its AI features, on AWS, Google Cloud, DigitalOcean or Cloudflare. We design the architecture, write the environments as code, connect managed and AI services and hand over a platform your team can run, built the way we run ours.
Learn More
AI-assisted cloud migration with a cut-over plan and a way back Netbase provides cloud migration services for companies running business applications on ageing servers or hosting to move them to AWS, Google Cloud, DigitalOcean or Cloudflare without losing data or control. AI speeds up the code and dependency assessment; every migration still has an inventory, a rehearsed cut-over, cost controls from day one and a tested rollback point.
Learn More
AI-assisted DevOps and platform engineering for routine releases Netbase provides DevOps and platform engineering for SaaS and commerce teams to make releases routine instead of risky. We set up CI/CD pipelines, repeatable environments and observability on AWS, Google Cloud, DigitalOcean or Cloudflare, with AI assistance for tests, reviews and incident triage, so every change ships the same way and problems surface before customers notice.
Learn More
SaaS product accelerator: launch an AI-ready SaaS on proven Netbase modules A SaaS product accelerator is a set of reusable Netbase modules for accounts, billing, roles and integrations that helps founders and product teams launch subscription software faster, with room for in-product AI from the first release. Reusing these modules can shorten development time, and the approach is proven on Printcart, Netbase's own web-to-print SaaS.
Learn More
Work with Netbase remotely
Every Netbase engagement is delivered remotely, so clients in any country and any industry can work with us; Netbase has worked with clients in 80 countries. Contracts and invoices are issued by Netbase JSC, and payment is made by bank transfer to Netbase JSC's company account or PayPal Business. Talk to the CEO, Huy Nguyen (David), on WhatsApp or by email.
Discuss a project
Netbase JSC helps organizations design, build, modernize, and operate digital products and AI-enabled business systems.project@netbasejsc.com
+84 937 869 689
91 Nguyen Chi Thanh, Dong Da, Hanoi, Vietnam
Get in touch
Tell us what you want to build, modernize, or operate.