Microsoft's Own AI Chip: The Maia 300 and the Plan to Deploy a Million Chips
Microsoft has announced the Maia 300, the third generation of its own custom AI chip. The company plans to deploy potentially more than a million of them next year to power its AI services.
Every time you ask an AI chatbot a question, somewhere in a huge building full of computers, a special chip wakes up and does the thinking. For years, most of that thinking has happened on chips made by one company: Nvidia. Now Microsoft, one of the biggest technology companies in the world, has announced the newest version of its own answer — a chip it designed itself, called the Maia 300. And Microsoft is not planning to make a few of them as an experiment. It says it plans to deploy potentially more than a million of these chips next year. That is not a small test. That is a statement: Microsoft wants to power its AI future with its own hands.
The short version
On Monday, October 12, 2026, Microsoft announced the Maia 300, the third generation of its custom AI chips. The company says it plans to deploy potentially more than one million of them next year to power its internal AI services. The chip is built from smaller pieces called chiplets joined together, using very advanced 2-nanometre technology from Taiwan’s TSMC for its main computing parts. Microsoft says the Maia 300 gets about 40% more useful work done per megawatt of electricity than leading GPU-based systems — but note carefully, that is Microsoft’s own claim, and the company did not name which systems it compared against. The bigger story is that Microsoft and its rivals are all racing to build their own chips so they depend less on Nvidia.
First, what is an AI chip, in simple words?
Think of a computer chip as the brain of a computer. A normal brain in your phone or laptop is a good all-rounder — it can do a little bit of everything. An AI chip is more like a brain that went to school for one subject: maths, repeated billions of times, incredibly fast. AI learns and answers by doing enormous amounts of simple calculations, one after another, without stopping. The chip is where all those calculations physically happen.
These chips live in giant buildings called data centres, packed in tall metal racks. When millions of people use AI services at the same time, the companies behind them need mountains of chips — and right now, most of those mountains are built with Nvidia’s chips. Nvidia became the main shop in town, and everyone else has been standing in its queue. That is the situation Microsoft is trying to change.
Meet the Maia 300
The Maia 300 is the third generation of Microsoft’s own custom AI chips. The word “generation” here is like phone models — the third version, built after learning lessons from the first two. This is not a sudden idea. Microsoft has been quietly working on its own chips for years, and each version gets better than the one before.
The plan for this chip is seriously big. Microsoft says it plans to deploy potentially more than a million Maia 300 chips next year, to back its internal AI services — the AI features and tools that Microsoft itself runs for its customers around the world. Each chip is a powerful computer brain, and a million of them working together is like building a small city of brains, all thinking at once.
Built like building blocks
Here is the clever part. Most big chips are made as one huge single piece, like carving a statue from one block of stone. Microsoft took a different road with the Maia 300. Its central processor is made of “chiplets” — smaller, specialised chips bonded together, like building with blocks instead of carving from one stone.
Why build it this way? Because different parts of the chip do different jobs, and not every job needs the most expensive material. The key computing parts — the parts that do the heavy thinking — use 2-nanometre technology from TSMC in Taiwan. The simple way to understand it: the smaller that number, the tinier and more advanced the pathways inside the chip, and generally the more work it can do with less electricity. Other chiplets use cheaper 3-nanometre technology for simpler jobs, like moving data on and off the processor.
It is a bit like a delivery company. You buy your fastest trucks for the main highways, but not racing trucks just to move boxes inside the warehouse. Microsoft used the expensive technology where speed matters most, and the cheaper one where the job is simple. Smart spending, inside a chip.
The electricity question
AI chips are hungry. They use a lot of electricity, and in the AI world, power is one of the biggest costs and biggest limits. This is why the numbers Microsoft shared about its server racks matter.
The Maia 300 chips will go into racks that hold 96 chips each and use about 167 kilowatts of power. For comparison, Nvidia’s most recent GB300 AI server racks hold 72 graphics chips and use about 142 kilowatts. Kilowatts are hard to picture, so think of it this way: these single racks use electricity on the scale of a street full of houses, not a single room. When a company plans to run a million chips, even small savings in electricity per chip add up to enormous savings overall — in money, and in how much power the buildings need.
This is also why Microsoft designed the chip the way it did. Rani Borkar, the president of Azure hardware systems and infrastructure at Microsoft, explained the idea behind the design in simple terms: the goal is to deliver more intelligence for every watt of electricity, every dollar spent, and every byte of memory, by making the whole system work together better instead of improving just one part. In everyday language: do not just make the brain faster — make the whole body smarter.
Microsoft’s big claim — read it carefully
Now comes the headline number, and it deserves honest handling. Microsoft says the Maia 300 chips will get about 40% more useful computing done per megawatt than “leading GPU-based systems.”
That sounds impressive, and it might be true — but here is what you should know. This is Microsoft’s own claim about its own product, and the company did not name the systems it compared against, so nobody outside Microsoft can check the comparison directly. That does not mean the claim is false. Treat it the way you would treat a shopkeeper saying his mangoes are 40% sweeter than the market’s — maybe they are, but you would still like to taste and compare yourself. The real test comes when the chips run at full scale next year.
Microsoft is not starting from zero
One more thing makes this plan believable: the Maia 300 is not Microsoft’s first step into the AI world. The previous generation of Microsoft’s custom chips is already working today. Those chips already run Microsoft’s in-house AI models — the AI systems Microsoft builds itself. They also run models from OpenAI, the company behind ChatGPT, which are delivered to customers through Microsoft’s Azure cloud computing service.
So the Maia 300 is not a dream on paper. It is the next step of a machine that is already running. Microsoft is upgrading the brains it already uses, and planning to install them by the million.
Why is everyone building their own chips now?
Microsoft is not alone in this race. According to Reuters, Microsoft’s cloud rivals and AI companies such as OpenAI are also designing their own custom chips. All of them have the same goal: to loosen Nvidia’s hold on the AI computing market.
Think of it like a town where only one shop sells flour. Every bakery in town must buy from that one shop, at that shop’s prices, whenever the shop has stock. Sooner or later, the bigger bakeries start thinking: maybe we should build our own flour mill. That is exactly what is happening in the AI world. The biggest AI companies have grown tired of depending on one supplier, so they are becoming chip makers themselves.
For the rest of us, this race is quiet good news. More competition usually means AI services get cheaper and faster. The Maia 300 may never sit inside your phone — it is built for giant data centres. But the answers you get from AI tools, and how fast they arrive, depend on chips exactly like this one.
Frequently Asked Questions
What is the Maia 300?
The Maia 300 is Microsoft’s newest custom AI chip, announced on October 12, 2026. It is the third generation of AI chips Microsoft has designed itself.
How many Maia 300 chips will Microsoft deploy?
Microsoft says it plans to deploy potentially more than one million of them next year, to power its internal AI services.
What are chiplets?
Chiplets are smaller, specialised chips bonded together to make one bigger processor. In the Maia 300, the main computing parts use advanced 2-nanometre technology, while simpler parts use cheaper 3-nanometre technology.
Is the Maia 300 really 40% better than other systems?
That is Microsoft’s own claim — the company says the chip does about 40% more useful computing per megawatt than leading GPU-based systems, but it did not name the systems it compared against.
Why is Microsoft making its own chips?
Like its rivals, including OpenAI, Microsoft wants to depend less on Nvidia, which currently holds a powerful position in the AI chip market.
Sources: Reuters (October 12, 2026).