Loading prices…
STKR NewsSTKR News0 of 3 free this month
AI

5 Interesting Learnings from the Databricks AI Index: Open-Weight Jumps to 31% of Enterprise Tokens, 72% of Companies Now Use Multiple Model Providers, and a 281x Gap Between Median and Power Users

New data from Databricks shows enterprise AI is shifting away from single-provider loyalty toward a messy, multi-model reality dominated by open-weight efficiency.

Originally on SaaStr →
AB

Adrian Boysel

Contributor

Nov 18, 2026

5 min read

Photo illustration / STKR News

The End of the Monolith

For the last two years, the narrative around enterprise AI has been dominated by a single name. We all know who it is. If you were building a tool for a Fortune 500 company, the assumption was you were plugging into a specific proprietary API and praying the rate limits didn't kill your margins. But the latest data from the Databricks AI Index tells a different story. This isn't survey data where CIOs tell you what they plan to do; this is actual token traffic from over 10,000 companies routing through the Unity Gateway. The reality on the ground is far more fragmented than the marketing departments in San Francisco would have you believe.

What we are seeing is the death of the one-model-fits-all approach. For builders, this is the most important shift since the launch of GPT-4. It means the "moat" of being an early partner with a major provider is evaporating. Enterprise customers are no longer interested in being locked into a single ecosystem. They are becoming pragmatists, and that pragmatism is driving a massive surge in open-weight adoption and multi-provider architectures.

The Rise of the Open-Weight Workhorse

The most striking metric in the report is that open-weight models now account for 31% of enterprise tokens. That is a massive jump. To be clear, we aren't talking about hobbyists running Llama on a laptop. We are talking about massive corporations moving production workloads away from closed systems. Why is this happening? It isn't just about cost, though that’s a factor. It’s about control.

When you use a proprietary model, you are at the mercy of the provider’s weights and their specific alignment tweaks. If they update the model and your carefully crafted prompts start hallucinating, you’re stuck. With open-weight models, enterprises can freeze a version, fine-tune it on their proprietary data, and deploy it with a level of predictability that closed APIs can’t match. For founders, this means your tech stack needs to be model-agnostic from day one. If your product is built solely on the capabilities of one specific closed model, you are competing with the eventual commoditization of those capabilities by the open-source community.

The Multi-Model Reality

The report notes that 72% of companies are now using multiple model providers. This is the death knell for the "exclusive partnership" era. Large organizations are realization that different models have different strengths. You might use a heavy-hitter like Claude or GPT-4 for complex reasoning or customer-facing logic, but you’re likely routing simpler classification or extraction tasks to a smaller, faster, cheaper model.

This creates a massive opportunity for builders. The infrastructure required to manage this complexity—routing, monitoring, and cost-attribution across multiple providers—is still in its infancy. If you are building tools that help developers orchestrate this mess, you are in a good spot. Enterprises don't want to manage five different billing dashboards and five different security postures. They want a unified layer that abstracts the model away from the application logic.

The Gap Between Winners and Losers

Perhaps the most jarring stat in the index is the 281x gap between the median AI user and the power users. We talk about AI as a horizontal technology that lifts all boats, but the data suggests it is actually widening the gap between the technically literate and the laggards. The power users aren't just using AI more; they are using it fundamentally differently. They aren't just putting a chatbot on their website; they are embedding AI into core operational workflows.

As a founder, this gap represents your target market. Selling to the median user is a slog because they are still trying to figure out what a prompt is. Selling to the power users is where the real scale happens. These are the companies that have already crossed the chasm of experimentation and are now looking for ways to optimize their token spend and improve their latency. They are the ones feeling the pain of the 281x usage volume, and they are willing to pay for tools that solve that scale problem.

Why Most AI Implementation Still Fails

Despite the massive growth in token volume, there is a skeptical side to this story. A lot of this traffic is still experimental. We see companies throwing tokens at problems without a clear understanding of the return on investment. The transition from "token growth" to "business value" is where the next wave of failures will happen. Many enterprises are over-engineering solutions with massive models when a simple heuristic or a small fine-tuned model would suffice.

The skepticism comes from the fact that while token usage is up, the number of truly transformative AI applications in the wild is still relatively small. We are in the "infrastructure build-out" phase. Companies are laying the pipes and testing the flow, but the actual water—the value—is still trickling. Builders who focus on the plumbing will do well in the short term, but the long-term winners will be those who can prove that those tokens are actually moving the needle on the bottom line.

What This Means for Builders

  • Don't get married to an API: If 72% of your potential customers are multi-model, your software must be too. Hard-coding for a single provider is a technical debt trap.
  • Focus on the orchestration layer: The complexity of managing open-weight and proprietary models simultaneously is a major pain point. Solve the management overhead, and you solve a CEO-level problem.
  • Target the power users: The median enterprise is moving too slowly. Look for the outliers who are consuming 200x the average volume; they are the ones with the budgets and the acute pain points.
  • Optimize for the 31%: Open-weight models are not a fad. Ensure your tools support local deployment and fine-tuning workflows, as that is where the enterprise is heading for their production loads.

The Takeaway

The enterprise is moving faster than expected toward a decentralized AI stack. The dominance of the big proprietary players is being challenged not by a single competitor, but by the collective utility of open-weight models and the pragmatic need for multi-provider redundancy. If you’re building for the enterprise, stop thinking about AI as a feature you plug in. Start thinking about it as a complex, multi-vendor resource that needs to be managed, optimized, and controlled. The winners of this cycle won't be the ones with the best prompts; they'll be the ones who build the most resilient and flexible systems for an increasingly fragmented world.


Read the original at SaaStr →

The Brief

Stay Updated on Cutting-Edge Tech

A six-minute morning dispatch on the markets and the technology shaping them.

Free. No spam. Unsubscribe anytime.

Write for STKR

Become a Contributor

Earn $STKR for published stories on markets, protocols, and culture.

  • Earn $STKR for every published piece
  • Editorial support from the STKR desk
  • Byline visibility across the network
  • First look at the upcoming creator program
Apply to Write

Keep reading

All stories

Comments

24 reader responses