Five Predictions for the AI Market: Why the Inference Era Rewards Distribution, Not Concentration

The AI Market Map, Redrawn: Five Predictions

Amon offered a historical frame at Davos that deserves attention. He compared the current AI moment to the year 2000 dot-com crash: “Go back to the year 2000 and you think about what we thought back then, what the internet would be. Today, 25 years later, it’s actually way bigger than what most people imagined. AI, in the long run, is going to be bigger than people think. It’s probably under-hyped for the long run.”

Five predictions for how the AI market map evolves, grounded in the Qualcomm thesis:

Prediction 1: Inference Becomes the Majority Revenue Pool by 2028

The training market won’t shrink, but inference will grow faster. NVIDIA’s response — acquiring Groq, building inference-optimized products — confirms the thesis. But Nvidia is optimizing GPUs for inference; Qualcomm is building inference-native architectures. The distinction matters at scale.

Prediction 2: Personal AI Devices Exceed Smartphones in Unit Volume by 2030

Not because smartphones decline, but because glasses, earbuds-with-cameras, rings, ambient devices, and industrial endpoints collectively outnumber phones. Qualcomm’s platform, powering the most diverse set of AI devices for the most brands, captures volume share.

Prediction 3: On-Device Inference Restructures SaaS Economics

As local AI compute becomes powerful enough for agent interactions, the cloud inference tax that currently inflates SaaS — as explored in the shift from SaaS to agentic service models — costs will erode. Software companies will shift from “AI means higher margins through cloud inference fees” to “AI means competing on local compute capability.”

Prediction 4: The Barbelled AI Silicon Market Consolidates

Nvidia owns the training/high-end inference pole through ecosystem dominance. Qualcomm builds the efficiency/edge-to-cloud pole through breadth and power optimization. Intel continues to decline. AMD remains viable but undifferentiated. The middle empties.

Prediction 5: Horizontal Platform Wins the Device War; Vertical Model Wins the Ecosystem War

Both can coexist. Apple will build the most integrated personal AI experience for its users. Qualcomm will power the most diverse set of AI devices across every form factor. The markets don’t fully overlap — there’s going to be a lot of diversity.

The Bottom Line

Qualcomm’s business model evolution tells you three things about the AI market map’s future.

First, the map is inverting. Training dominated the first era of AI infrastructure — as explored in the economics of AI compute infrastructure — . Inference — distributed, efficient, edge-aware — will dominate the next. The $180B training market isn’t going away, but the $50B+ inference market is growing faster and will be larger.

Second, the edge is not the periphery — it is becoming the center. When 85% of enterprise workloads are inference-based, and inference increasingly runs on-device or at the edge, the company with 2B+ deployed NPUs has more structural relevance than the company with the most powerful GPU cluster.

Third, the winner of the next era is not the biggest chip — it is the broadest fabric. Nvidia wins the training era by owning the most powerful GPU. The inference era winner will be the company that provides the most seamless compute continuum from pocket to data center.

The AI market map you know — with Nvidia at the top and everything else underneath — is a snapshot of 2024. The map that is emerging in 2026 has inference at its center, the edge as its largest surface area, and the connective tissue between cloud and device as the true chokepoint.

The training era rewarded concentration. The inference era will reward distribution. Qualcomm’s transformation is the map.

This is part of a comprehensive analysis. Read the full analysis on The Business Engineer.