AI is no longer just a technology sector. It is becoming an economic force. But the bigger question is: will AI-driven growth spread across economies and businesses, or concentrate wealth among a handful of companies and countries? That may define the next decade.
ECONOMY
-

GLM-5.3 Local FP8 Inference Can Require Multiple H100 GPUs
By
–
you just need 10–12× H100 for FP8, or $400.000–$650.000 to run GLM-5.3 locally.
-

CommerceAgentBench Tests Agents on 107 Open E-Commerce Tasks
By
–

Most AI benchmarks test whether a model can give the right answer. CommerceAgentBench asks whether an agent can actually finish the work. Accio has open-sourced 107 e-commerce tasks across procurement, product listings, operations, fulfillment and after-sales. Agents work across
-
AI Lowers Some Entry Barriers While Big Tech Retains Frontier Power
By
–
Is AI Really Leveling The Playing Field, Or Making Big Tech Even More Powerful? #ArtificialIntelligence is making powerful #technology cheaper and more accessible than ever, giving small businesses new ways to compete with much larger rivals. But with the most advanced #AI
-

Commentator Questions Local Mac AI Economics Given Rapid Hardware Obsolescence
By
–
My position is clear: none of the new Mac systems is suitable for running local AI at the necessary level. Even investing $50,000 would mean maintaining a local system that will be nearly obsolete in six months, forcing you to reinstall everything several [Traduit de FR vers EN]
-
Monitoring AI News on X Carries Data Collection and Token Costs
By
–
Not much. The hard part is collecting the posts. To grab 30,000 posts (about a day's worth from the AI community here on X) it costs $100-$150. Then you gotta pay for the tokens to process such. @blevlabs is paying that. Brayden, how much is my account running in terms of
-

Nvidia Earnings Leave Questions About Its Financial Influence on AI
By
–
Nothing on Nvidia’s call Wednesday addressed the persistent concerns about how it is using its financial power to shape the AI industry. As long as results are phenomenal, which they are, it’s easy to kick that can down the road… Nvidia’s stunning forecast Nvidia’s report
-
Replit Automates Model Selection Across Coding Agent Tasks
By
–
With Intelligent Model Routing, Replit picks the model for every task you send, so you can operate with quality, speed, and efficiency, at no extra cost.
— Replit ⠕ (@Replit) 27 août 2026
Here’s how it works, plus a breakdown on the difference between Free, Power and Max ↓ pic.twitter.com/INI9iYumlHWith Intelligent Model Routing, Replit picks the model for every task you send, so you can operate with quality, speed, and efficiency, at no extra cost. Here’s how it works, plus a breakdown on the difference between Free, Power and Max ↓
-
Grid Connections Constrain UK Compute Expansion More Than U.S. Expansion
By
–
The key difference between the US and UK (and much of Europe more generally) is that UK compute is far more “grid-gated” than American compute. In the UK, power comes before compute. In the US, power and compute are increasingly being built together. UK data centres are facing
-
Databricks Reports Higher OfficeQA Quality and Lower Cost With GLM Flash
By
–
.@Zai_org's latest open-weight model, GLM 5.3 Flash, is now available on Databricks!
— Databricks (@databricks) 27 août 2026
GLM 5.3 Flash joins 30+ open-source and frontier models on Databricks. On Databricks’ OfficeQA Pro v2 benchmark, it delivers 10% higher quality than GLM-5.2 at just one-tenth the cost, pushing… pic.twitter.com/j5fvuvxs5q.
@Zai_org
's latest open-weight model, GLM 5.3 Flash, is now available on Databricks! GLM 5.3 Flash joins 30+ open-source and frontier models on Databricks. On Databricks’ OfficeQA Pro v2 benchmark, it delivers 10% higher quality than GLM-5.2 at just one-tenth the cost, pushing