Tech industry experts are raising alarms about the surge of free or heavily subsidized AI tokens offered by major vendors. While these deals look attractive on the surface, analysts warn they can trap organizations into specific ecosystems, leading to significant vendor lock-in and long-term costs.
The hidden cost of ‘free’ AI access
AI vendors are currently engaging in a land grab for customers by offering cheap tokens, often subsidized by venture capital. They are also deploying forward-deployed engineers (FDEs) to push their specific models into enterprise environments. Max Leaming, head of data science and AI solutions at ManpowerGroup, notes that these incentives encourage companies to build workflows around proprietary Large Language Models (LLMs) and agents.
Once business processes are built around a specific model, switching becomes difficult and expensive. Jack Gold, principal analyst at J. Gold Associates, observes that while some adopt hybrid strategies to cut costs, the risk of being tied to a single vendor’s ecosystem remains high as the landscape evolves.
Adopting a multi-vendor strategy
To mitigate these risks, Max Goss, senior director analyst at Gartner, advises IT decision-makers not to fear adopting a multi-vendor approach. He argues that it is unlikely any single AI vendor or model will meet all of an organization’s requirements. By using different tools for different tasks, companies can extract value from various strengths without becoming dependent on one provider.
Logan Wolfe, partner of global AI strategy at Kyndryl, adds that the future likely involves a multi-model landscape. However, he emphasizes that strategies should be grounded in specific use cases rather than vendor preferences. For highly regulated sectors like finance or healthcare, safety and privacy concerns may limit rapid model switching based solely on cost.
Practical implementation for everyday users
For lower-stakes use cases, a flexible approach is prudent. Wolfe suggests that during heavy load times, such as in a customer support data center, you might switch to a more capable model. During quieter periods like evenings and weekends, you can optimize by using less expensive options. This dynamic switching prevents breaking the bank while maintaining performance when needed.
Kellie Romack, Chief Digital Information Officer at ServiceNow, advocates for understanding how AI is built within your organization. She resists ripping out one vendor’s platform to replace it with another, preferring to look at existing technology and find the ‘best of breed’ solutions. Her team runs multiple models in-house, such as Anthropic’s Claude and Microsoft’s Copilot, through a single LLM gateway.
Resilience against outages
Avoiding vendor lock-in is critical for service continuity. Recent months have seen outages hit major AI services from OpenAI and Claude. Gartner’s Goss points out that relying on a single provider with a single model introduces significant risk. A multi-model approach provides necessary fallback options, ensuring that operations can continue even if one service goes down.
Romack also highlights the importance of transparency in AI spending. Her team monitors token spend daily, comparing costs between engineers to identify inefficiencies. For example, they might find Claude better for reading long Word documents, while Copilot excels at quick summaries. This granular control helps prevent unnecessary expenses and ensures that human beings understand how the AI is built, debugged, and maintained.
What this means for you: If you are using AI tools in your Windows workflow, be cautious of ‘free’ tiers that lock you into a specific ecosystem. Consider whether your current setup allows you to switch models if needed. Diversifying your AI toolset can save money and protect your productivity against service outages.
Over to you: Are you currently using a mix of AI tools like Copilot and Claude, or have you settled on a single provider?
