Guest: Frank Nagle
Company: The Linux Foundation
Show Name: An Eye on AI
Topics: AI Infrastructure

“If we use open models, then our data becomes open as well.” Frank Nagle, Chief Economist at The Linux Foundation, has heard this statement from multiple companies, governments, and organizations. It’s also completely wrong. This single misconception is costing enterprises thousands while open source AI models offer not just cost savings, but better data security and customization capabilities they desperately need.

The Data Security Paradox

The irony runs deep. When you run open models on your own servers, your data never leaves your firewall. It stays within your controlled environment. Closed proprietary models running in external clouds? Your data actively leaves your system every time you make a request. The fear driving companies toward closed models is actually pushing them toward the exact outcome they’re trying to avoid.

Nagle acknowledges one legitimate concern: support and liability. When something breaks at 3 AM, enterprises want someone to call. Closed model providers offer that safety net. Open source ecosystems, at least currently, lack the same immediate support infrastructure. For established organizations with compliance requirements and risk-averse cultures, this matters.

Where Startups Lead, Enterprises Follow

Startups tell a different story. Cost-conscious and comfortable with technical risk, they’re racing toward open models. They care less about liability and more about burning through runway. The contrast reveals an important truth: model selection isn’t about technical superiority—it’s about organizational priorities and risk tolerance.

The Three-Step Framework

Nagle offers a practical path forward. First, educate your team on the actual pros and cons of open versus closed models. Strip away misconceptions. Second, evaluate what you genuinely need. Do you prioritize vendor support or customization? Cost savings or liability protection? Third, experiment. Don’t bet the company, but test open models in low-risk environments.

The low-hanging fruit exists in use cases where customization matters. African governments are building on open models to improve local language support—something closed models haven’t prioritized. Companies with specialized domains can fine-tune open models for industry-specific terminology and use cases. You can’t do this with ChatGPT or Claude.

Beyond the Binary Choice

The either-or framing misses the point. AI strategy isn’t about choosing Team Open or Team Closed. Different tools serve different purposes. Some workloads demand the reliability and support of closed models. Others benefit from the cost efficiency and customization of open alternatives. The mistake is defaulting to closed models without evaluating whether open models might serve you better.

Platforms like Hugging Face host thousands of fine-tuned open models. Organizations are building custom solutions daily. The ecosystem is maturing. Support options are emerging. The gap between “enterprise ready” closed models and “experimental” open models is closing faster than most decision-makers realize.

The question isn’t whether to use open or closed AI models. It’s whether you understand your use cases well enough to choose the right tool for each job.