- The Relentless Optimist
- Posts
- Meliora Propose FairGPT And A UK Sovereign AI Model
Meliora Propose FairGPT And A UK Sovereign AI Model
Plus, Microsoft Recorded US$10 Billion In Revenue From OpenAI I One Quarter and Mistral Releases the Weights of Its Trillion-Parameter AI Model
Subscribe | 2nd November, 2026

In this fast-moving Vertical AI economy, headlines are everywhere but optimism is rare.
Here’s our take on 3 stories that will help you be relentlessly optimistic about the future.
1. Meliora Propose FairGPT And A UK Sovereign AI Model
Fifty months ago, generative AI barely existed in its current form. Today, just under 2 billion people use it every week. At Manchester’s Beyond The Music festival, Meliora Managing Partner Clive Dickens argued that governance has not kept pace with that adoption, and that the UK still has time to lead rather than follow. His starting point is that AI is now a basic human right, in the same way water, shelter, and power are, which means the way it is built matters to all 8.2 billion people on the planet. His answer is to reimagine GPT as FairGPT, with models that are: Fair Governed: operating under clear regulation and law, with mandatory reach reporting across society, a treaty of non-proliferation around harmful AI, and pre-approval before release, just as medicine needs regulatory sign-off before it reaches the public. Fair Powered: run on renewable energy, with far more done to power data centres this way. Fair Trained: built without scraped or stolen work, with provenance, consent, and payment to sources made mandatory. Dickens then set the UK a challenge. Rather than wait to regulate models built elsewhere, the country should train its own, a hypothetical Hope V.1, drawing on the UK’s vast stock of non-digitised knowledge and created by November 2027, when Manchester hosts the G20. People say it is too late, he argued, but “we are in the first five minutes of the next industrial revolution.” |
Here’s why this matters: The music industry is often the first to feel a technology shift, and the debate over consent and payment is already live. Dickens turns that argument into a design brief. FairGPT gives regulators, creators, and technology companies a shared test: is a model governed fairly, powered responsibly, and trained with permission? It also gives sovereignty a constructive purpose. A UK model that rewards the people whose work feeds it would let the country decide how its creative culture is used and exported, and would make creators partners in the AI economy. That matters well beyond music. As governments across Europe build national models, the question of what a fair model looks like will shape procurement, licensing, and trust for every media, telecoms, and technology business that depends on AI. Dickens has set a date and a place for the answer.
2. Microsoft Recorded US$10 Billion in Revenue From OpenAI in One Quarter
Microsoft generated US$10 billion in revenue from commercial arrangements with OpenAI in the quarter ended 30 September, including revenue-sharing payments. That is nearly 11% of its total quarterly revenue of US$92.3 billion, and compares with US$24.1 billion for the whole of its previous fiscal year. At that pace, OpenAI-related arrangements would bring in around US$40 billion annually. OpenAI also owed Microsoft US$7.2 billion in receivables at the end of September, up from US$6 billion in June, and Microsoft’s ownership stake has slipped to 24% from about 25% in June. |
Here’s why this matters: Microsoft and OpenAI now compete for many of the same enterprise customers, yet each remains one of the other’s most important commercial relationships. For Microsoft, OpenAI-related revenue is more than a tenth of everything it earns, which makes OpenAI’s growth a direct driver of Microsoft’s results. The revenue-sharing arrangement runs through 2030 subject to a cap, even though OpenAI is now free to buy compute from rival clouds. For enterprises choosing between AI providers, the lines between partner and competitor have blurred, and those ties will shape pricing and product decisions for years.
3. Mistral Releases The Weights of Its Trillion-Parameter AI Model For Cyber Defence
Mistral has released the weights of Mistral Large 4, a trillion-parameter open model nicknamed Le Chonk, allowing businesses and governments to download and run it themselves. The company is positioning it for cyber defence, where organisations want control over their data and systems, and says it is comparable in performance to Chinese open-weight rival GLM-5.3 while using 47% fewer tokens per task. It also claims the top spot among open models on a coding agent benchmark. The weights are available on Hugging Face, and the model can be run on-premises. |
Here’s why this matters: Mistral is making a specific argument: the organisations defending critical systems should own the AI that protects them. Because the weights can be downloaded, no provider can withdraw access, a point that carries weight after the June export order showed how quickly a closed model can disappear. Running the model on-premises also keeps sensitive security data inside an organisation’s own environment, a priority for regulated European industries and governments. The efficiency claim matters because a trillion-parameter model still demands serious hardware. Whether defenders can use that capability faster than attackers is the open question, and Mistral’s release will test it.
Still Curious?
DoorDash to acquire Instacart for US$17 billion
Anthropic just launched tools to show companies what their AI spend is buying
Meta has a new AI pet app called Waddle World
Apple unveils its first touch-screen laptop computer
AI’s insatiable appetite for chips helps Samsung rake in a record US$80.5 billion
OpenAI debuts ‘Shepherd’ Model for AI research validation
Put your AI agents to work with these 5 killer tasks
Our Pursuit Of Better
Eugenia Kuyda, who founded the AI companion app Replika, has taken her new venture Wabi to a US$1.25 billion valuation with a US$55 million round led by IVP. Wabi began as a tool for building apps from prompts and has become a personal AI agent that builds mini apps for tasks like meal planning and fitness plans. |
Ask a career question on SuperMe and it convenes a group chat of AI profiles built from real experts’ knowledge, including well-known venture investors. The start-up is raising US$20 million led by Forerunner, with Greylock participating. Founded by former Pinterest executives Casey Winters and Ludo Antonov, it is free for now and also works inside Claude and ChatGPT. |
Troubled that fewer buildings are built in the US than decades ago, two former Palantir executives have launched Dream with US$17 million in seed funding led by Andreessen Horowitz. Its world model, trained on more than 100 million square feet of permit data, designs structurally sound buildings, and is deployed on 10 projects in Texas and California. |






