How Amazon Built a $10 Billion Business on Free Software

The Great PostgreSQL Heist | How Amazon Toppled The Oracle Empire - YouTube

—And Why Nobody Could Stop Them

Amazon Web Services transformed the enterprise database market by weaponizing open-source PostgreSQL into Aurora, a cloud service that matched Oracle's performance at a fraction of the cost. The strategy was entirely legal, enormously profitable, and profoundly unfair to the volunteer developers whose decades of work generated billions for one of the world's richest companies.

For three decades, Oracle Corporation held the Fortune 500 in a financial stranglehold. In climate-controlled server rooms across the world, Larry Ellison's database software was the price of doing business. If you wanted reliability—if you needed bank transactions to clear or airline tickets to print—you paid the Oracle tax. Licenses were complex and punitive. CIOs lived in fear of audits that could trigger tens of millions in unexpected fees.

"Oracle's licensing was incredibly complex and punishing," said Gartner analyst Donald Feinberg in 2014. "Companies would get audited and suddenly owe millions in unexpected fees. It created enormous anxiety."

By the early 2010s, Oracle controlled approximately 40% of the global relational database market, generating over $9 billion annually from database products alone. Enterprises accepted the costs because alternatives couldn't match Oracle's ability to handle mission-critical workloads—the transactions powering banking systems, airline reservations, and retail operations.

The irony was that Amazon itself was one of Oracle's largest customers.

The Internal Oracle Problem

Despite pioneering cloud infrastructure through Amazon Web Services, Amazon's retail and fulfillment operations ran extensively on Oracle databases well into the 2010s. Every Prime Day, every holiday rush, Amazon wrote checks to a company whose business model AWS was trying to disrupt.

"We were one of Oracle's biggest customers," Andy Jassy, then CEO of AWS, acknowledged in 2016. "Every dollar we spent with them was a dollar that went to a competitor. It was strategically untenable."

A 2019 AWS blog post later confirmed the scale: Amazon had migrated 75 petabytes of data from nearly 7,500 Oracle databases—a project that took years and became both a strategic necessity and a marketing opportunity. If Amazon could successfully migrate its extraordinarily complex operations away from Oracle, it could prove the viability of cloud-native alternatives to skeptical enterprise customers.

The challenge was that existing open-source databases couldn't match Oracle's performance and reliability at scale.

The PostgreSQL Foundation

PostgreSQL emerged from the University of California, Berkeley in the 1980s as an academic research project. Unlike MySQL, which Oracle acquired in 2010, PostgreSQL was governed by the PostgreSQL Global Development Group—a loosely organized community of volunteer developers who had spent decades building world-class database technology nights and weekends, driven by engineering excellence and community ideals.

The critical detail: PostgreSQL is licensed under the PostgreSQL License, a permissive open-source license similar to BSD or MIT licenses. Unlike the GNU General Public License (GPL) that governs Linux and requires sharing modifications, PostgreSQL's license explicitly permits creating proprietary derivatives without requiring source code disclosure.

"The PostgreSQL License was chosen deliberately to maximize adoption," said Bruce Momjian, a core PostgreSQL developer since 1996, in a 2015 interview. "We wanted companies to be able to build commercial products on top of PostgreSQL without legal concerns."

The implicit social contract was clear: We won't force you to contribute back legally, but if you build a successful business on our work, you'll contribute voluntarily out of enlightened self-interest and community responsibility.

That social contract would be systematically violated.

Aurora's Technical Breakthrough

AWS announced Aurora at its re:Invent conference in November 2014. When Andy Jassy took the stage in the Venetian ballroom in Las Vegas before 30,000 attendees, the industry expected another incremental cloud service update.

Instead, Jassy unveiled a database engine that would reshape the enterprise market.

"I am very excited to announce the launch of a brand new database engine called Amazon Aurora," Jassy said. "Aurora means new dawn. We've been working on this for three years."

He projected a graph showing Aurora outperforming standard MySQL by five times while providing "the performance and availability of high-end commercial databases at one-tenth the cost."

Aurora's innovation was architectural: separating the database compute layer from the storage layer. Traditional databases tightly couple these functions on a single server—if the server dies, you're in trouble. If you need more storage, you need a bigger server.

Aurora kept the PostgreSQL SQL processing engine—the part developers knew and loved—but replaced the storage subsystem entirely with a purpose-built, distributed storage service.

According to a 2017 paper published in SIGMOD by Aurora's engineering team, the system replicates data across six storage nodes in three availability zones. Rather than writing complete data pages, Aurora transmits only redo log records—the minimal information needed to reconstruct changes—reducing network I/O by approximately 7.7x compared to traditional database replication.

"The key insight was that in the cloud, the network is the bottleneck, not the CPU," wrote Aurora principal engineer Anurag Gupta in the technical paper. "By pushing more intelligence into the storage layer and reducing what we send over the network, we could achieve much higher throughput."

This architecture enabled Aurora to automatically scale storage from 10GB to 128TB without downtime, maintain six copies of data across three availability zones, and recover from failures in under 30 seconds—capabilities that previously required expensive enterprise software and dedicated database administrators.

The strategic brilliance: Aurora looked exactly like PostgreSQL to developers and applications. Any code written for PostgreSQL ran on Aurora without modification. It was a Trojan horse—familiar open-source interface concealing a proprietary cloud engine underneath.

The Market Shift

For enterprises, Aurora solved the durability problem that had kept them locked to Oracle. Companies could lose an entire data center without losing a single byte of data, with recovery in seconds rather than hours. And it was self-healing—storage volumes grew automatically, eliminating the expensive database administrator armies Oracle's complexity required.

The migration numbers began slowly, then accelerated into a landslide. Capital One, Expedia, Samsung, Nasdaq, and Intuit publicly discussed moving mission-critical workloads to Aurora.

"We migrated our core banking systems from Oracle to Aurora," said Capital One Distinguished Engineer George Brady at AWS re:Invent 2017. "We reduced database costs by 70% while improving performance and reliability. That would have been unimaginable five years ago."

Nasdaq announced in 2017 it was migrating core trading systems to AWS. "The scalability and resilience of Aurora matches what we previously could only achieve with expensive commercial databases," Nasdaq CIO Bradley Peterson told The Wall Street Journal.

The financial implications were staggering. According to Gartner, AWS held approximately 14% of the global database management system market by 2019, up from near zero before Aurora's launch. Oracle's market share declined from 41% in 2013 to 28% in 2020.

Larry Ellison initially dismissed cloud databases. "Amazon's database doesn't even have basic features like read consistency," he said at Oracle OpenWorld 2014. "You can't run a bank on that."

But by 2018, facing undeniable market reality, Ellison acknowledged on an earnings call: "We didn't take the cloud seriously enough, soon enough. That was a mistake."

AWS database services (primarily Aurora, RDS, and DynamoDB) generated an estimated $8-10 billion in annual revenue by 2021, according to Cowen analyst Derrick Wood. Aurora became the fastest-growing service in AWS history.

Google Cloud and Microsoft Azure rapidly copied the strategy, launching AlloyDB and Azure Database for PostgreSQL Hyperscale respectively. All followed Aurora's playbook: maintain compatibility with popular open-source databases while implementing proprietary technology underneath.

The Open-Source Controversy

Aurora's relationship with PostgreSQL ignited fierce debate within the open-source community. While AWS contributed to PostgreSQL development—employing several core developers and funding infrastructure—critics argued the company captured disproportionate value from community work.

The numbers tell the story: According to a 2021 analysis, the PostgreSQL Global Development Group operates on roughly $1-2 million annually in donations and sponsorships. Meanwhile:

• AWS database services: ~$10 billion annual revenue
• Google Cloud PostgreSQL services: ~$1-2 billion annual revenue
• Microsoft Azure PostgreSQL services: ~$1-2 billion annual revenue

Cloud providers generate over $10,000 in revenue for every $1 that flows back to core PostgreSQL development.

Volunteers who spent decades building world-class database technology received thank-you notes and modest stipends. Amazon—owned by one of the world's richest individuals—extracted billions.

The controversy intensified in 2018 when MongoDB, Redis Labs, and other open-source database companies changed their licensing specifically to prevent cloud providers from offering managed versions without contributing back. MongoDB introduced the Server Side Public License (SSPL), explicitly designed to require cloud providers to open-source their entire infrastructure stack or negotiate commercial licenses.

"Cloud providers have built multi-billion dollar businesses on the back of open source without adequately contributing to the communities that created the software," MongoDB CEO Dev Ittycheria said in a 2018 blog post announcing the license change.

AWS responded by creating DocumentDB (MongoDB-compatible) and ElastiCache (Redis-compatible) services that reimplemented the APIs without using the original code—exactly what Aurora had done with PostgreSQL.

The difference: PostgreSQL's permissive license explicitly allowed proprietary derivatives. The new SSPL licenses attempted to prevent it. But by then, for PostgreSQL, it was too late.

The Legal Reality vs. The Moral Question

AWS's use of PostgreSQL was entirely legal. The PostgreSQL community had deliberately chosen a permissive license to maximize adoption, explicitly allowing companies to build commercial products without sharing modifications.

"The PostgreSQL License gives companies like AWS explicit permission to do exactly what they did," explained Bradley M. Kuhn, policy fellow at the Software Freedom Conservancy, in a 2019 blog post. "The community can dislike it, but they can't claim a license violation. That's why MongoDB and others changed to more restrictive licenses—but by then, it was too late for PostgreSQL."

If PostgreSQL had used the Affero GPL (AGPL)—a strong copyleft license requiring code sharing even for network services—AWS would have faced three options: use PostgreSQL and open-source all of Aurora's storage layer, reimplement PostgreSQL from scratch (prohibitively expensive), or negotiate a commercial license.

Instead, PostgreSQL's permissive license allowed AWS a fourth option: use PostgreSQL legally while sharing nothing proprietary.

Several PostgreSQL developers have expressed regret. "If we'd known cloud providers would build $10 billion businesses on PostgreSQL while contributing back $1 million, we might have made different licensing choices," core contributor Joe Conway said in a 2019 interview. "But you can't put that genie back in the bottle."

The legal framework had failed to protect community interests. The social contract had been broken. Volunteers who created enormous value were systematically shortchanged.

But expecting corporations to voluntarily pay for things they can legally get free is naive. It asks them to act against their fiduciary duty to shareholders.

Oracle's Decline and Adaptation

Oracle's database revenue remained substantial—approximately $8-9 billion annually—but showed minimal growth as customers shifted to cloud alternatives. The company launched its Autonomous Database service in 2017 and aggressively pursued cloud infrastructure through Oracle Cloud Infrastructure, repositioning itself as a cloud provider competing directly with AWS.

Amazon's internal migration off Oracle databases became a high-profile demonstration of AWS capabilities. According to AWS's 2019 blog series, Amazon moved 75 petabytes from 7,500 Oracle databases, shut down 1,700 Oracle instances, reduced database costs by over 60%, and reduced latency by 40%.

"This was our most complex technical challenge since building AWS itself," wrote AWS's Jeff Barr in the announcement. "If we could do this, any company could."

The migration served dual purposes: eliminating costs to a competitor and creating marketing credibility for enterprise customers considering cloud adoption.

The Broader Pattern

The Aurora story isn't unique. It represents a systemic problem across open-source software:

Linux: Multi-trillion dollar economic impact, primarily captured by corporations, while kernel developers struggle for funding.

OpenSSL: Secured trillions in internet commerce, operated on donations until the Heartbleed vulnerability exposed they had approximately one full-time developer.

curl: Used by billions of devices worldwide, maintained largely by one person who until recently struggled to fund his work.

Tech corporations have built empires on volunteer labor, then claimed "we followed the license" when questioned about giving back.

The permissive licenses were created with a specific philosophy: remove all barriers to adoption so technology spreads as widely as possible. Corporations took the "no obligations" part literally while ignoring the reciprocity expectation.

Current State and Implications

As of 2024, cloud database services account for over 50% of database management system spending, with AWS, Microsoft Azure, and Google Cloud collectively holding approximately 75% of the cloud database market.

PostgreSQL has become the fastest-growing database technology according to DB-Engines rankings, driven largely by cloud provider adoption. Stack Overflow's 2023 Developer Survey found PostgreSQL was the most admired database technology, preferred by 45% of professional developers.

The explosion in PostgreSQL usage represents a technical victory for the community—their database conquered the market. But it's a Pyrrhic economic victory. Adoption came at the cost of the community's ability to capture fair value from their work.

"In retrospect, the database wars weren't about who had the best code," said Martin Casado, general partner at Andreessen Horowitz and former VMware executive, in a 2022 podcast. "They were about who controlled the operational platform. AWS won by making databases a managed service rather than software you install."

The lessons are harsh:

First, permissive open-source licensing enables commercial adoption but provides limited mechanisms for communities to capture value from their work. It's a gift to corporations and a trap for communities that create valuable infrastructure.

Second, cloud computing fundamentally changed software economics by shifting from upfront licensing to consumption-based pricing. This made expensive enterprise software more accessible while creating vendor lock-in through data gravity—once massive datasets reside in a cloud provider's infrastructure, migration becomes prohibitively expensive.

Third, API compatibility became more valuable than code ownership. By maintaining PostgreSQL compatibility while controlling the infrastructure, AWS captured the benefits of PostgreSQL's ecosystem—tools, extensions, developer expertise—while controlling the underlying platform.

Fourth, expecting corporations to voluntarily compensate open-source communities fairly is naive. Capitalism optimizes for profit extraction. Open-source permissive licenses remove barriers to profit extraction. Therefore, corporations rationally extract maximum profit while minimizing contributions.

What Should Happen (But Won't)

In a just world, AWS would voluntarily contribute $100-200 million annually to PostgreSQL development—roughly 1-2% of their database revenue. Other cloud providers would contribute proportionally. This would fund hundreds of full-time developers, dramatically accelerating PostgreSQL development, and everyone would benefit from better software while the moral debt was paid.

But this won't happen because there's no legal requirement, shareholders would revolt at "unnecessary" spending, competitors would free-ride on others' contributions, and corporate incentives oppose fair compensation.

Instead, we're seeing license proliferation as new projects avoid permissive licenses, defensive mid-stream license changes that break trust, corporate capture through hiring key maintainers, a sustainability crisis as critical infrastructure projects struggle for funding, and developer burnout as maintainers watch corporations profit from their unpaid work.

Conclusion

Amazon Web Services didn't violate PostgreSQL's license—it exploited the permissive licensing to build a proprietary implementation behind an open interface. This strategy, combined with superior cloud-native architecture and aggressive pricing, accelerated the decline of traditional database licensing and Oracle's market dominance.

The shift from Oracle to Aurora represents less a legal controversy than a business model revolution. Oracle sold expensive software licenses for on-premises deployment. AWS sold managed database services with pay-as-you-go pricing and cloud-native capabilities. Enterprises chose convenience, scalability, and economics over software ownership.

The database market no longer centers on selling software—it's about renting trust. Amazon, through calculated engineering and relentless pricing pressure, became the landlord of the internet's memory.

Whether this outcome serves the long-term interests of open-source communities remains bitterly debated. PostgreSQL usage has exploded globally. The technology won. But questions persist about whether the community receives adequate support from companies generating billions in revenue from their decades of volunteer work.

The tragedy is that basic fairness suggests those who profit enormously from others' work should compensate them fairly. The legal reality is that corporations optimizing for shareholder returns will extract maximum value while minimizing contributions—exactly what the license permits.

Amazon didn't break the rules. It used them with ruthless efficiency.

It's legal. It's profitable. It's rational corporate behavior.

And it's deeply, fundamentally unfair.

The volunteers who created enormous value were systematically shortchanged. The social contract was broken. And we're left with a system where brilliant engineers donate their work, corporations extract billions, and everyone shrugs because "that's what the license allows."

The database wars are over. The cloud won. But the cost was paid by people who never saw the profits.

---

Verified Sources and Citations

1. Gartner Market Share Analysis: Gartner, Inc. "Market Share Analysis: Database Management Systems, Worldwide, 2020." Published June 2021. https://www.gartner.com/en/documents/4002489

2. AWS Aurora Announcement: Barr, Jeff. "Amazon Aurora – New Cost-Effective MySQL-Compatible Database Engine for Amazon RDS." AWS News Blog, November 12, 2014. https://aws.amazon.com/blogs/aws/highly-scalable-mysql-compat-rds-db-engine/

3. Aurora Technical Paper: Verbitski, Alexandre, et al. "Amazon Aurora: Design Considerations for High Throughput Cloud-Native Relational Databases." Proceedings of the 2017 ACM International Conference on Management of Data (SIGMOD '17), pp. 1041-1052. DOI: 10.1145/3035918.3056101

4. Amazon's Oracle Migration: Barr, Jeff. "Migration Complete – Amazon's Consumer Business Just Turned off its Final Oracle Database." AWS News Blog, October 30, 2019. https://aws.amazon.com/blogs/aws/migration-complete-amazons-consumer-business-just-turned-off-its-final-oracle-database/

5. PostgreSQL License: PostgreSQL Global Development Group. "PostgreSQL License." https://www.postgresql.org/about/licence/

6. MongoDB SSPL Announcement: Ittycheria, Dev. "MongoDB now released under the Server Side Public License." MongoDB Blog, October 16, 2018. https://www.mongodb.com/blog/post/mongodb-now-released-under-the-server-side-public-license

7. Oracle Database Revenue: Oracle Corporation. Annual Report (Form 10-K) for Fiscal Year 2020. Filed June 26, 2020. https://www.oracle.com/investor/

8. Capital One Aurora Migration: Brady, George. "Capital One's Migration to Amazon Aurora." AWS re:Invent 2017 presentation. https://www.youtube.com/watch?v=eyjGW_CTaF8

9. Nasdaq Cloud Migration: Nash, Kim S. "Nasdaq Moves Markets to Amazon's Cloud." The Wall Street Journal, November 29, 2017. https://www.wsj.com/articles/nasdaq-moves-markets-to-amazons-cloud-1511967001

10. IDC Database Market Data: IDC. "Worldwide Database Management Systems Software Forecast, 2020–2024." Document #US46894320, October 2020.

11. DB-Engines Rankings: DB-Engines. "DB-Engines Ranking." Updated monthly. https://db-engines.com/en/ranking

12. Stack Overflow Developer Survey: Stack Overflow. "2023 Developer Survey Results." https://survey.stackoverflow.co/2023/

13. Software Freedom Conservancy Commentary: Kuhn, Bradley M. "How To Sustain FOSS Projects With Funding." Software Freedom Conservancy Blog, April 2019. https://sfconservancy.org/blog/2019/apr/17/sustainability/

14. PostgreSQL Development History: Momjian, Bruce. "The History of PostgreSQL Development." Presentation at PostgreSQL Conference East, 2015.

15. PostgreSQL Funding Analysis: Asay, Matt. "The Elephant in the Cloud: PostgreSQL's Open Source Paradox." InfoWorld, March 2021.

16. Andy Jassy Interviews: Jassy, Andy. Various interviews including CNBC Mad Money, November 2016; AWS re:Invent keynotes 2014-2019.

17. Larry Ellison Statements: Ellison, Larry. Oracle OpenWorld keynote addresses 2014-2018; Oracle earnings calls (transcripts available via Oracle Investor Relations).

18. Cowen Research: Wood, Derrick. "Amazon Web Services Deep Dive." Cowen Equity Research, March 2021.

19. Google AlloyDB Announcement: Google Cloud. "Introducing AlloyDB for PostgreSQL." Google Cloud Blog, May 11, 2022. https://cloud.google.com/blog/products/databases/alloydb-for-postgresql-intelligent-scalable-storage

20. Azure PostgreSQL Hyperscale: Microsoft. "Azure Database for PostgreSQL - Hyperscale (Citus)." Azure Updates, 2019. https://azure.microsoft.com/en-us/updates/

21. Gartner CIO Analysis: Feinberg, Donald. "Predicts 2014: Database Management Systems." Gartner Research, December 2013.

22. Martin Casado Commentary: Casado, Martin. Interview on "Software Engineering Daily" podcast, September 2022.

23. Joe Conway Interview: Conway, Joe. "PostgreSQL in the Cloud Era." PostgreSQL Conference Europe, 2019.