The News
Speculation around potential IPOs from OpenAI and Anthropic has reignited debate about who will ultimately capture the most value in the AI industry. Bindesh Vijayan, Co-founder and CTO of Myndlab and a former Microsoft engineer, argues that framing this as a race between frontier model providers misses a more consequential shift: value in AI may be migrating away from the model layer entirely, toward the companies building applications, workflows, and integrations on top of it. His central contention is that public markets will force a reckoning, demanding evidence of adoption, revenue durability, and customer retention rather than benchmark performance.
Analyst Take
The IPO is a stress test, not just a milestone
Going public does something private funding rounds never require: it forces a company to explain, repeatedly and under scrutiny, exactly how it makes money and why customers stay. For OpenAI and Anthropic, that test will be severe. Private investors have been willing to fund capability development as a long-term bet. Public investors will want to see gross margin, churn, and net revenue retention. The AI industry has spent several years winning on technical merit. The public markets will ask a harder question: is any of this sticky?
Vijayan’s argument lands here. Most enterprise buyers are not choosing between GPT-4o and Claude 3.5 Sonnet the way they once chose between Oracle and SAP. They’re choosing between solutions to specific operational problems, and the model underneath is increasingly an implementation detail. That dynamic is not hypothetical. ECI Research’s 2026 Application Development: DevSecOps + AppSec survey found that AI code governance is the #1 priority investment area for enterprise security teams heading into 2026. Security teams are not buying frontier models. They’re buying governed, integrated tooling that happens to run on top of them.
The commoditization question
There is a reasonable counterargument that frontier model providers maintain durable moats through scale, proprietary training data, and inference infrastructure. That case is harder to make than it was two years ago. Open-weight models from Meta, Mistral, and others have closed the capability gap on many benchmarks. Inference costs have dropped sharply. The structural dynamic Vijayan is describing, where the application layer captures more value than the infrastructure layer, has precedent across multiple technology cycles. Cloud infrastructure commoditized compute. Kubernetes commoditized container orchestration. In both cases, the next wave of high-margin businesses were built on top of the newly commoditized layer, not inside it.
What makes AI different, and potentially more complex, is the speed at which the capability floor keeps rising. A model that was frontier-class eighteen months ago is now available for near-zero marginal cost. That creates real pressure on any company whose primary differentiation is model quality rather than distribution, workflow depth, or domain-specific fine-tuning.
What enterprise adoption data actually shows
The picture from enterprise buying behavior supports Vijayan’s framing more than it challenges it. According to ECI Research, half of enterprise AI leaders say their organizations still rely primarily on public AI tools like ChatGPT or Copilot. That figure captures the real adoption story: most enterprises are not deep in the model stack. They’re using general-purpose interfaces and increasingly demanding that AI capabilities be embedded in the tools they already operate. The implication for OpenAI and Anthropic is significant. Consumer and developer mindshare does not automatically translate into enterprise revenue. Enterprises buy through procurement, require security reviews, need contractual SLAs, and want integration with existing identity, compliance, and data governance systems. Those are application-layer problems, not model-layer problems.
ECI Research data reinforces the governance angle specifically. The same 2026 DevSecOps + AppSec survey found that 83.8% of respondents use code scan tools during CI/CD processes, a figure that illustrates just how deeply tooling is already embedded in development workflows. The companies that win enterprise AI spend will be those that embed AI natively into those workflows, not those that ask enterprises to route work through a standalone model interface.
Looking Ahead
The IPO of either OpenAI or Anthropic, whenever it comes, will be watched as a market signal. But the more durable story will be written by the companies building vertical applications, orchestration layers, and AI-native developer tooling that sits between the model and the end user. Those companies will compound on enterprise distribution advantages that frontier model providers will find expensive and slow to replicate.
The “model wars” framing Vijayan challenges will likely persist in the trade press through an IPO cycle, because it maps cleanly onto a competitive horse race narrative. Investors and analysts who look past it will find the more interesting question: which application-layer companies are accumulating the data flywheels, workflow integrations, and switching costs that define durable enterprise businesses? That is where the next generation of AI value will be concentrated, and it is a question the public markets will eventually force the industry to answer clearly.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
