For a small or mid-size team, the multi-cloud decision comes down to how much data must cross between the clouds each month, and whether a second provider buys anything the first cannot. At 2026 list prices, a 1 Gbps private link between AWS and Google Cloud in the US West costs USD 3,555.10 a month with no per-GB charge, and it undercuts AWS internet egress only above about 41 TB a month in one direction.12 Below that, and with no workload that needs the second provider's service, the second cloud adds a bill without adding a saving.
Most firms run several clouds, and few designed it
Flexera's 2026 State of the Cloud report, released 18 March 2026 from a survey of more than 750 cloud decision-makers and users, says multi-cloud adoption keeps rising, often through mergers, SaaS sprawl and decentralized teams rather than deliberate strategy.3 It puts AWS in 83% and Azure in 79% of active enterprise workloads.3 If both figures describe the same respondents, at least 62% run both (83 + 79 − 100).
Flexera sells cost-management software, so read the survey as a count of what firms run, not as evidence that two clouds pay.
A second provider that arrived by acquisition is a different decision from one a team chose.
What crossing the cloud boundary costs at list price
Data leaving one provider's network is billed per gigabyte unless a private link replaces the internet path. AWS announced general availability of Interconnect - multicloud on 14 April 2026 with Google Cloud as the first partner, and named Azure and Oracle Cloud Infrastructure for later in 2026.4 AWS's pricing page, updated 5 November 2026, says the link is billed hourly by bandwidth and tier, with no per-gigabyte charge, from creation to deletion in one-hour increments. It allows one free 500 Mbps interconnect per provider per Region, and it says the other cloud sets its own price for its side.2
| Cost item | Unit | Price (date, region) | Source |
|---|---|---|---|
| AWS internet egress | per GB | USD 0.09 for the first 10 TB a month after 100 GB free, USD 0.085 for the next 40 TB; us-west-2, read 17 Apr 2026; excludes NAT, cross-AZ, CloudFront | Classmethod1 |
| AWS Interconnect, Tier 1, 1 Gbps | per hour | USD 1.37 (USD 1,000.10 per 730-hour month); us-west-2 to Google us-west1, 17 Apr 2026; excludes Google's side | Classmethod1 |
| Google Partner Cross-Cloud Interconnect for AWS, 1 Gbps | per hour | USD 3.50 (USD 2,555.00 per month); North America, 17 Apr 2026; no data transfer fee | Classmethod1 |
| AWS Interconnect, Tier 1, 10 Gbps | per hour | USD 12.33 (USD 9,000.90 per month); us-east-1 to Google us-east4, page updated 5 Nov 2026; excludes Google's side | AWS pricing page2 |
| AWS Interconnect, Tier 4, 10 Gbps | per hour | USD 51.78 (USD 37,799.40 per month); us-east-1 with a Singapore Cloud WAN edge, 5 Nov 2026 | AWS pricing page2 |
Tier follows topology. AWS assigns each interconnect one tier based on the highest-tier path it uses, so attaching a Singapore Cloud WAN edge to a us-east-1 interconnect moved AWS's own 10 Gbps example from USD 12.33 to USD 51.78 an hour (51.78 × 730 = USD 37,799.40 a month).2 A diagram of two clouds does not show that.
In the EU, Article 34(2) of the Data Act lets providers charge egress for parallel use of two services only to pass on costs they incur.5 Article 29(1) bars switching charges for leaving a provider from 12 January 2027.6 In September 2025 Google said qualifying parallel traffic in the EU and UK would bill at zero, Microsoft introduced at-cost fees, and AWS said EU customers could request reduced rates. Those offers are over six months old, so confirm current terms.7
One workload at 50 TB a month
Take an application on AWS in us-west-2 (Oregon) and an analytics service on Google Cloud in us-west1 (Oregon). The team moves 50 TB a month, one direction, from AWS to Google. The volume is illustrative. The prices are the list prices above, read on 17 April 2026.1
- Internet path: first 100 GB free; 9,900 GB × USD 0.09 = USD 891.00; 40,000 GB × USD 0.085 = USD 3,400.00; total USD 4,291.00.
- Private 1 Gbps link: AWS 1.37 × 730 h = USD 1,000.10; Google 3.50 × 730 h = USD 2,555.00; total USD 3,555.10 at any volume.
- Difference at 50 TB: 4,291.00 − 3,555.10 = USD 735.90 a month in the link's favor. At 10 TB the link costs USD 2,664.10 more (3,555.10 − 891.00).
- Break-even: (3,555.10 − 891.00) ÷ 0.085 = 31,342 GB, plus the first 10,000 GB, gives about 41,342 GB, or about 41 TB a month.
"Private link" is not one price. The same 50 TB over the older route costs more:
| Route | Fixed monthly fee | Per-GB charge | Total at 50 TB | Beats internet above |
|---|---|---|---|---|
| Internet egress | none | USD 0.09, then 0.085 | USD 4,291.00 | n/a |
| Interconnect + Partner CCI, 1 Gbps | USD 3,555.10 | none | USD 3,555.10 | about 41 TB |
| Direct Connect 10 Gbps + Cross-Cloud Interconnect 10 Gbps + VLAN, non-redundant | USD 5,803.50 | USD 0.02 on the AWS side | USD 6,803.50 | about 100 TB |
| Same, four-way redundant | USD 23,214.00 | USD 0.02 on the AWS side | USD 24,214.00 | about 650 TB |
The Direct Connect rows use Classmethod's prices and my one-direction arithmetic: 5,803.50 + 50,000 GB × 0.02 = 6,803.50, and 23,214.00 + 1,000.00 = 24,214.00.1 The break-evens are mine. For the non-redundant route, 5,803.50 + 0.02V equals 4,291 + 0.07 × (V − 50,000) at V ≈ 100,250 GB. Classmethod says the managed link carries four-way redundancy on Google's side, while the non-redundant route carries a 95% AWS SLA and no Google SLA cover.1
Two mistakes distort this comparison. The first treats the link as a monthly commitment. AWS bills by the hour, so a one-off 50 TB move at 5 Gbps takes about 22 hours (50,000 GB × 8 ÷ 5 Gbps ÷ 3,600 s). On the 10 Gbps pair that costs about USD 689 (12.33 × 22 + 19.00 × 22), against USD 4,291.00 by internet.12 The second sizes the link to peak load. A 1 Gbps link carries at most 328,500 GB a month (1 Gbps × 2,628,000 s ÷ 8), so 50 TB uses about 15% of it.
What the documents don't say
Price pages list on-demand rates. Negotiated discounts and commitment terms sit in private contracts, so neither table nor break-even includes them. The Google rate and AWS's 1 Gbps rate were read on 17 April 2026, over six months before this draft. AWS's page confirms the structure and the 10 Gbps Tier 1 rate, but not the 1 Gbps rate.2 If the free 500 Mbps AWS tier fits and Google's side still bills at 1 Gbps, the link costs USD 2,555.00 and breaks even near 30 TB ((2,555.00 − 891.00) ÷ 0.085 + 10,000 = 29,576 GB). The sources do not say whether the two sides' bandwidths must match, so treat that figure as untested.2
What a second cloud does for availability
AWS's region-level EC2 SLA, last updated 25 May 2022, commits to 99.99% monthly uptime.8 A 31-day month holds 44,640 minutes, so 99.99% allows about 4.5 minutes of unavailability. AWS's builder guide says the commitment applies to instances spread across two or more Availability Zones, with credits of 10% from 99.0% up to 99.99%, 30% from 95.0% up to 99.0%, and 100% below 95.0%. The same guide lists 25%, not 30%, for the middle band on S3, Lambda and RDS, so EC2's figures do not carry over to other services.9
AWS's post-event summary puts its us-east-1 DynamoDB disruption between 11:48 PM PDT on 19 October 2025 and 2:20 PM PDT on 20 October, in three periods of customer impact.10 That is 14 hours 32 minutes, or 872 minutes. Suppose, as an illustration, that every minute counted as unavailable for a customer's EC2, though the SLA defines unavailability by service. October uptime would be 1 − 872 ÷ 44,640 = 98.05%, inside the 30% band.9 The remedy is a bill credit capped at 30% in that band, not payment for lost revenue.
Neither document says what a second provider would have changed that day.
Reported StatusGator, a monitoring company, reports that its web application was down about 3.5 hours across two periods after the 20 October AWS outage, that it responded by adding a second region, and that during the 18 November Cloudflare outage it could not disable its web application firewall because Cloudflare's control panel was also down.11 Its stated first priority afterward was a second CDN and firewall provider, ahead of reducing reliance on hyperscalers, which it expected to take a year or more.11 The account is self-reported, but it puts shared dependencies at the edge layer.
Who gains, who pays and who is quietly worse off
At 50 TB the link moves money between providers. AWS collects USD 1,000.10 in hourly fees where internet egress would have paid it USD 4,291.00, which is USD 3,290.90 less. Google collects USD 2,555.00 it would not otherwise have earned. The team saves USD 735.90. These are my arithmetic on list prices and say nothing about either provider's intent.
The quietly worse-off are teams below the break-even, who pay up to USD 2,664.10 more a month at 10 TB, and teams whose topology lands in a higher tier.
Decision rule by purpose and volume
A second provider earns its cost when a workload needs a service the first does not sell, when a contract or regulator requires one, or when a measured failover test shows the second environment works. Without one of those, choose a single provider: the table shows a new fee and no offsetting saving.
If one does hold, choose by monthly volume, one direction, at the US West list prices above:
- Under about 41 TB: internet egress is cheaper (about 30 TB if the free AWS tier applies and Google's side stays at 1 Gbps).
- From about 41 TB to the 1 Gbps ceiling of 328.5 TB: the managed 1 Gbps link.
- Above 328.5 TB: the 1 Gbps link is full. The 10 Gbps pair costs USD 22,870.90 (9,000.90 + 13,870.00) and beats internet only above about 382 TB, so internet is cheaper in between.12
- One-off migration: the link, billed by the hour.
In the EU, check whether parallel-use traffic qualifies for a reduced or zero rate before pricing it at USD 0.09 per GB.7 Test failover before counting on it. StatusGator reported that the mitigations it added after the AWS outage worked during the Azure incident but still needed manual activation.11


Comments