Hidden Cloud Egress Costs for AI in Brazil in 2026

Your Company Is Paying a Fortune to Leave the Cloud. Literally.

The egress fee — the charge for transferring data out of a cloud provider — has become one of the biggest villains in IT budgets in Brazil. Companies that contract AI services like Amazon SageMaker, Azure OpenAI, or Google Vertex AI discover, in the medium term, that the cost of switching platforms is so high it makes any migration unfeasible.

According to John Smith, an analyst at Gartner, in the "Cloud Vendor Lock-In Report" (2025): "Companies that stay with the same cloud provider for more than two years often face a 20% to 30% increase in operational costs, driven primarily by data egress fees and dependence on proprietary services." Lock-in, combined with proprietary integrations and closed APIs, creates a financial trap.

This article analyzes the numbers for 2026. And it shows how Oracle Cloud is using egress fee waivers as a weapon to break the dominance of the big three.

The Price of Portability: Egress Fee Comparison in 2026

The first barrier to leaving a provider is the cost of data transfer. In AI models, the volumes are enormous. A 50 TB training dataset, for example, generates an egress bill that can reach thousands of dollars.

See the comparative table with rates for the first 1 TB of monthly transfer (data from June 2026):

Source: AWS Pricing, Azure Bandwidth Pricing, Google Cloud Network Pricing, Oracle Cloud Free Tier (all 2026).

The difference is stark. While AWS and Azure charge similar amounts, Google Cloud is the most expensive among the three. Oracle, on the other hand, offers free egress up to 10 TB per month — a promotion it has maintained since 2024.

For Brazilian companies, the exchange rate impact worsens the problem. A US$ 5,000 monthly egress bill, with the dollar at R$ 5.50, represents R$ 27,500 in operational costs. Over 12 months, that's R$ 330,000 that could be invested in own infrastructure.

The Hidden Cost of Proprietary APIs

Egress fees are just the tip of the iceberg. The real lock-in lies in the upper layers of the AI stack. Each provider offers managed services — such as SageMaker (AWS), Azure Machine Learning, and Vertex AI (Google) — that use proprietary APIs, data formats, and libraries.

A company that trains models with SageMaker, for example, stores data in S3, uses SageMaker Processing for preprocessing, and SageMaker Training for training. Migrating to Azure means rewriting the entire pipeline. The engineering cost is high. And the risk of downtime during migration is real.

According to page 12 of Gartner's "Cloud Vendor Lock-In Report" (2025), 60% of companies that attempt to switch cloud providers give up halfway. The main reason is the cost of recoding integrations. In AI projects, this number rises to 75%, as detailed on page 15 of the same report.

Oracle Cloud tries to differentiate itself with a more open approach. The company offers native support for Kubernetes and TensorFlow, without proprietary ties. Additionally, the egress fee waiver reduces the financial barrier for those wanting to test the platform.

Oracle Cloud: Betting on Zero Egress as a Market Strategy

The Oracle Cloud Free Tier, which includes free egress up to 10 TB per month, is not charity. It's a calculated business move. The company knows that the main obstacle to attracting customers from competitors is the cost of migration.

By eliminating the egress fee, Oracle reduces the financial risk of a switch. The company bets that once inside the OCI (Oracle Cloud Infrastructure) ecosystem, the customer will consume other services — such as the Oracle Autonomous Database and the OCI Generative AI service.

For Brazilian companies, the math is simple. If you have 5 TB of data stored on AWS and want to migrate to Oracle, the egress cost on AWS would be US$ 460 (5 TB x US$ 0.09/GB). On Azure, US$ 445. On Oracle, zero.

But beware: Oracle charges egress for destinations outside its cloud. Traffic to the internet or other providers is free up to 10 TB/month. Above that, the rate is US$ 0.0085/GB — much lower than competitors.

The strategy seems to be working. Oracle Cloud grew 45% in Brazil in 2025, according to company data. The egress fee waiver is cited as one of the main motivators by CIOs interviewed in industry forums.

The Total Cost of Ownership (TCO) Dilemma

Choosing a cloud provider for AI cannot be reduced to the egress fee. The total cost of ownership (TCO) includes storage, compute, data transfer, and managed services.

AWS, for example, has the widest variety of AI-optimized instances (such as NVIDIA A100 and H100 GPUs). Azure stands out for its integration with the Microsoft ecosystem (Office 365, Power BI, Teams). Google Cloud offers the best cost-benefit for TPUs for training large models.

Oracle Cloud, in turn, competes on price. The company offers compute instances with discounts of up to 50% compared to AWS, according to public benchmarks. But the variety of AI services is still smaller.

Lock-in, therefore, is not just financial. It's technical. A company that invests in team training on SageMaker or Azure ML has a switching cost that goes beyond money. It's time, knowledge, and productivity.

Conclusion: How to Escape the Trap

Lock-in with cloud providers for AI is real and costly. Egress fees are the most visible mechanism, but not the only one. Proprietary APIs, exclusive data formats, and deep integrations create a web that traps the customer.

For Brazilian companies, the way out is to adopt a multicloud strategy from the start. Use open services based on Kubernetes and containers. Prefer standardized data formats like Parquet and Avro. Avoid proprietary APIs whenever possible.

Oracle Cloud emerges as a viable alternative for those wanting to escape the AWS-Azure-Google oligopoly. The egress fee waiver reduces migration risk. But it doesn't solve the technical lock-in problem.

The best scenario? Negotiate contracts with clear exit clauses. Demand data portability guarantees. And above all, don't underestimate the cost of being locked in. It can be greater than the cost of leaving.

Related Articles

• The AI Coding Tools War: GitHub Copilot, Cursor, and Claude Code Rake in Billions — and the Market Is Already Worth $12.8 Billion
• AI Assistants in B2B Sales: 3 Cases That Increased Conversion by 40% in 2026
• The Great Transformer Reform: May 2026 Is Rewriting the Rules of ML