Independent author, specialized in decentralized artificial intelligence and crypto ecosystems. I analyze projects by what they actually do, not by the noise around them. Openly a supporter of the Qubic project: my articles about Qubic and its direct competitors carry a disclosure.
The net flows of U.S. spot bitcoin ETFs reach 2.73 billion dollars from September 1 to 28, 2026, according to data from Farside Investors. August ended at 3.52 billion. But September is not like August: 2.42 billion, or 88% of the total, arrived in six sessions from September 21 to 28. Cointribune reconstructed the two months session by session to show it.
Under the “AI crypto” label, search engines lump together networks that do not do the same job. Some sell an intelligence market, others graphics rendering, software agents, or model training.
As of September 6, 2026, the "Next French presidential election" market on Polymarket shows $128,026,232 in traded volume, according to the platform's public data. It is one of the most active political markets in the world, and it concerns a vote that will not take place until April 2027. At the same time, this market is legally inaccessible from France: the National Gaming Authority ordered its blocking on July 16, 2026. A financial instrument built on the blockchain, a real French audience, and a regulator who qualifies it as illegal gambling: the issue is not the race; it is the mechanics.
OpenAI launched GPT-6 Astra on September 3, 2026, its first model classified as "Critical" in cybersecurity: according to the company, it discovers unknown vulnerabilities and writes the corresponding exploit without human guidance at each step. Its president Greg Brockman sees a possible milestone towards general artificial intelligence. For crypto, the stake is not theoretical, and it is already quantified. As early as February 2026, the EVMbench benchmark, published by OpenAI with Paradigm and OtterSec, measured this capability on smart contracts: the best agent exploited 72.2% of the tested vulnerabilities.
The GPU shortage succeeded where no manifesto had. It turned "decentralized AI", until recently just a slogan, into a real market. High-end accelerators have been out of stock for months, with lead times stretching past a year, and that scarcity pushed developers toward networks that pool idle hardware from around the world. Except the label "decentralized AI platform" now covers very different realities, and confusing them is the surest way to pick the wrong one. Renting a GPU is not the same as training a model. Scoring machine-learning outputs is not hosting an unrestricted LLM. This guide sorts the main players of 2026 by what they actually do, weighs the strengths and blind spots of each, and sets an outsider, Qubic, against the rest: the only network on this list where mining itself trains neural networks. No investment advice follows, only a map of the landscape.