This Week In AI: Intelligence will not live in a server, it will live in your house

One of the big tech debates of our time is around the distribution of AI, and hence intelligence. Will it be served through huge server farms, controlled by a few companies? Or will it be available to people directly, on demand and in their control? This week in AI the debate became sharper thanks to Qwen AI.Artificial Intelligence will live in your house.IThere is a gulf between what the tech world, particularly those in Silicon Valley and Bengaluru, talk about nowadays. And what the rest of humanity is doing. Last week, the geeks of the world were busy installing Qwen AI 3.8, the one with 27 billion parameters, on their computers. The rest of humanity was, well, going about its life the way it has always done. That is in a gloriously mundane fashion. Or Albert Camus once described it, writing that “they’ll be able to sum up modern man in a single sentence: he fornicated and read the papers.” Though in 2026, we can safely replace “read the papers” with “scrolled the reels.”But then Camus lived 70 years ago, which in 2026 feels like an eternity away. We are living in the time of wonders, such as the Qwen AI 3.8 27B. This is a condensed and compressed version of the full Qwen AI 3.8 model. Created by Alibaba Cloud, this is a model that shows China has caught up to America on AI. More or less. The most powerful American models, including the Claude Fable, are barely ahead of Qwen AI 3.8 and GLM 5.3.But the intelligence of the Qwen 3.8 did not set the hearts of geeks and AI nerds fluttering. Instead it was Qwen AI 3.8 27B, a model small enough to be run on a mid-range MacBook Pro or a computer with a high-end Nvidia graphics card. It requires just 24GB of video or unified RAM. A bigger surprise is what happens when you install it on your computer. Like most other Chinese AI models, Gwen 3.8 too is an open-weight AI model. Anyone can download and install it, unlike ChatGPT and Claude that must be used through OpenAI and Anthropic servers. A lot of people did download Qwen AI 3.8 27B and ran it on their local computers. It ran like not a small AI model, but like a giant in terms of intelligence.The most astounding aspect of the Qwen AI 3.8 27B, and if you are paying attention you already know it by now, is that it provides intelligence at the level of something like ChatGPT Luna and Claude Opus 4.6 directly on your computer without connecting to any servers. It is local intelligence, in your control and completely free. This has sparked a debate: what is the future of intelligence? I believe the debate will intensify. At one end you have Anthropic CEO Dario Amodei and others, arguing that AI is best served through huge server farms, metered and controlled. At the other end, there are Chinese companies which are releasing ever powerful open-weight AI models, which anyone can run.In this equation Qwen AI 3.8 27B shows there is possibility of extremely high-intelligent AI systems that can run locally on regular computers. The technology can be made efficient enough to become completely personal.In a way, it pushes the envelope and frees AI intelligence. It weakens the hold of tech giants over AI. If Qwen AI 3.8 27B is so good locally, in another year or so we can expect small models that will match the best of what we currently have, such as Claude Fable and ChatGPT Sol 5.6, while running on a local computer. The big models and huge data server farms will still offer the absolute best intelligence, but for most people Qwen AI 3.8 27B class AI would be more than enough.Think of AI as a drill or a hammer: You probably have one in your home, although when you need to remodel your entire house you still seek professional services and heavy-duty tools. The AI scene may look somewhat similar in a couple of years.IIIf Chinese are now beginning to match AI intelligence that Americans offer, with something as small as Qwen AI 3.8 27B, how should Americans respond? The AI fight between China and the US is one of the most interesting geo-political tussle of this decade, and probably the next. It is our moon race, or who can make the biggest bomb. It is the new cold war, with espionage and subterfuge deployed overtly and covertly.I believe the American tech companies have started feeling the heat. Until the beginning of this year, except for the DeepSeek moment in early 2025, the US companies were clearly ahead. They were releasing frontier AI models quickly and had a big-enough lead over their Chinese counterparts. And because they had the lead they could charge a premium for their AI, could dictate how it would be used, and could keep it in their tight control. The Chinese, meanwhile, kept hammering with their AI models, releasing them for free as open-weight models, trying to sabotage the business case for ChatGPT and Claude.Now, Americans are getting the message. They are responding. OpenAI is trying to reduce the cost of its AI. It has announced that ChatGPT Luna 5.6, its smallest but still very potent AI model, would be free to use with unlimited chatting even if you don’t subscribe to any plan. The price of Sol, the top OpenAI model, has been cut by 20 per cent for developers. Mark Zuckerberg, when he recently wrote his missive on AI, spoke of easily accessible AI intelligence and open source AI models. Grok and Gemini continue to be relatively cheap to use. And now Nvidia is committing around $6 billion to develop an open-source AI model that could match the best from the Chinese companies.Will this be enough to ward off the Chinese AI threat? We will see. But I do feel that American tech giants, Anthropic and OpenAI in particular, have painted themselves in a corner with their approach to AI. They have invested 100s of billions of dollars and all this investment makes business sense only if they are leaps and bounds ahead of open-source AI models. Currently, they are not and Chinese, who run on a different sort of business model, are fast catching up to them.IIIThe mysterious AlphaLast week a new AI model cropped up on OpenRouter, a website that lets people use multiple AI models. Called Ox Alpha, it is free at the moment. It also, the way it is delivering AI intelligence in different tasks, seems like a frontier model, matching or surpassing the best that are out there.Here is how OpenRouter introduces it to users: “Ox Alpha is a stealth model. It is developed and operated by a third-party provider who has chosen to remain anonymous during this preview. It is a reasoning model designed for coding, sustained agentic work, and production workloads. It is suited for long-horizon software engineering, complex reasoning, and workflows that combine text with visual context.”Alpha has created a lot of buzz and people are speculating who could be behind it. There are various names popping up. Maybe it is Elon Musk and his xAI team. Or may be it is Google, which has been somewhat silent since the release of Gemini 3.1 Pro months ago. Google also has a history of doing a mystery release as we saw with Nano Banana earlier. Or maybe it is some Chinese company, probably the GLM 5.3 maker Zhipu AI.No one knows for sure except the OpenRouter folks. But now with the model out in public, and creating buzz, its owner should be revealed in a week or two.IVRecord set by Usain Bolt is broken by a robotThere is an Olympics going on in the world right now. An Olympics for robots. General intelligence is already somewhat solved in terms of practical application in mundane tasks, although not for creativity and abstract reasoning. The next step for the tech industry is going to merge all their top AI models with the physical world. They will largely do this through robots.It is one of the most exciting spaces to watch, as exciting as the games taking place in humanoid Robots Olympics in China. In one game, a robot broke the 100-meter record set by Usain Bolt, by a fairly sizable difference of 19 seconds. It's another matter that after breaking the record, the robot slammed into a soft break-wall. No, this is not the one which had gone viral earlier this week when it slammed into the wall and broke its spine. This one survived after the crash.Other robots too are proving fairly entertaining. Some are fighting boxing matches. Some are doing high jumps, and so on and so forth.These are early days — I know, even I am tired of using these words but such is the world right now — in robotics. Robots are now improving fast in how they handle balance, speed, and in how they use their mechanical actuators. They are now getting the sensory intelligence through AI models we have.Think of 2026 as the year 1905. Cars at the time were slow and clunky. The first airplane, built by a few bicycle mechanics, had its first flight just two years ago in which it flew for 12 seconds. New York was still mostly lit by oil lamps. In India, there were likely less than 10 elevators.That world changed dramatically and quickly in the next 20 years. Given the pace of change nowadays we are looking at a much faster timeline. Extremely smart and dexterous robots are coming and I have a feeling that in the next 5 years we will begin to see many of them around us.AI vs cancerLast week Moderna and Merck made a joint announcement: they have created a mRNA vaccine to check melanoma. The vaccine has successfully passed the Phase 3 human trial and now the companies will be working to make it available generally. In this news there was also a hint that AI played some — likely minor — role in the development of the vaccine. But coming just days after Anthropic CEO said AI would defeat most diseases in 5 to 10 years, this was significant. We still don’t know how much AI can really push the envelope on medical sciences. But the potential is there, particularly around drug formulations, research and data analysis, finding right protein mixes, and simulations. It could be game changing even if Dario Amodei proves wrong in his overly optimistic prediction.- EndsPublished On: Aug 23, 2026 14:39 IST