The next wave of AI adoption will be driven by deflation
Cheaper intelligence, not better benchmarks, is what reaches the next billion users.
The AI adoption gap
In the second quarter of 2026, 73.3% of the UAE’s working-age population used generative AI, the highest share anywhere. In the Global South the share was 16.2%, against 28.8% in the Global North. That gap widened from 12.1 to 12.6 percentage points in a single quarter. Access shows the same split: internet access is 65.7% in the Global South against 90.1% in the North, and electricity access is 88.9% against 98.1%.
Why AI costs are falling
The cost of a given level of intelligence is falling fast. Stanford’s 2025 AI Index found that querying a model at roughly GPT-3.5-level quality fell from $20 per million tokens in November 2022 to $0.07 in October 2024, a 280-fold drop. Epoch AI finds the price of reaching a fixed level of performance has fallen 9x to 900x a year, depending on the task. The trend has continued: Silicon Data’s index put average inference prices at roughly $1.16 to $1.18 per million tokens in early August 2026, a 2026 low.
What falling AI costs mean for adoption
What cost dollars per million tokens just a few years ago now costs fractions of a cent. This democratization is being aggressively accelerated by open-weight models, which now deliver near state-of-the-art performance for a fraction of the computing overhead, allowing for rapid localized adaptation.
When the cost of analyzing an image or data point ceases to be a constraint, industrial operations fundamentally change. Inspections that are currently sampled intermittently, simply because a human surveyor’s time is expensive, can instantly become continuous, exhaustive, and automated.
What cheaper AI does not fix
However, a cheaper API call does not lay fiber-optic cables or stabilize a rural power grid. Falling inference costs do not supply the connectivity required to stream heavy visual data from a remote construction site or an offshore shipyard back to a distant data center. For heavy industry, the cloud remains a fragile dependency.
Why BrahmAksh runs AI at the edge
This is exactly why we engineered the BrahmAksh platform to run autonomously at the absolute edge. We deploy our multi-modal inspection and safety models directly onto the physical site, executing localized, sub-65-millisecond reasoning even in the presence of a spotty constant connection to a centralized server.
If bandwidth and connectivity are the fundamental constraints of the physical world, the only solution is to move the intelligence directly to the device deployed at the site where the work is happening.