Microsoft Ignite 2023: the age of Copilots — and the year Microsoft built its own silicon

17. November 2023

  • Home
  • Blog
  • Microsoft Ignite 2023: the age of Copilots — and the year Microsoft built its own silicon
Microsoft Ignite 2023: the age of Copilots — and the year Microsoft built its own silicon

14–17 November 2023, Seattle Convention Center (hybrid). Roughly five thousand of us on-site and somewhere north of a hundred thousand watching online. This was the Ignite where Microsoft stopped describing AI as a feature and started describing it as the company. Satya Nadella walked out, said the line everyone quoted for the next twelve months — "We are the Copilot company" — and then spent a keynote renaming, rewiring, and rebranding most of the portfolio around that one word.

But the announcement that has aged the best wasn't a Copilot at all. It was two pieces of metal: Azure Maia 100 and Azure Cobalt 100, the first chips Microsoft ever designed itself for its own datacentres. We'll come back to why that's the throughline that connects this conference to everything Microsoft has shipped since.

The Seattle Convention Center, a Microsoft Ignite venue

"We are the Copilot company"

Nadella's framing was deliberate and total. The keynote pitch wasn't "here are some AI tools you might like." It was a worldview: that the assistant — the Copilot — becomes the way you reach knowledge, and the apps and infrastructure rearrange themselves behind it.

"We are the Copilot company. We believe in a future where there will be a Copilot for everyone and everything you do." — Satya Nadella, Ignite 2023 keynote

He went further, suggesting Copilot would become "the new UI" for getting at the world's knowledge and your organisation's knowledge, and that Microsoft was shipping roughly a hundred updates across the Copilot stack to make that real. That number is worth holding onto, because it tells you what kind of conference this was. Ignite 2023 wasn't one flagship reveal; it was a hundred coordinated moves all pointing the same direction.

The most visible of those moves was the simplest: Bing Chat became, simply, Copilot. The consumer assistant, the Windows assistant, the web assistant — all collapsed into one name. For anyone who'd spent 2023 explaining to customers the difference between Bing Chat, Bing Chat Enterprise, the Windows thing, and the Microsoft 365 thing, the unification was a relief even if the branding gymnastics were a little dizzying. Copilot for Microsoft 365 also became generally available for enterprise customers that same month, which mattered more than the rename: it was the moment the assistant stopped being a preview you demoed and became a line item a CIO could actually buy.

Copilot, but for builders: Studio, Azure, and Service

Underneath the branding, the genuinely useful releases for a shop like ours were the ones aimed at people who build with this stuff rather than just chat with it.

  • Copilot Studio arrived in public preview — a low-code environment for customising Copilot for Microsoft 365 and standing up your own custom copilots, wiring in your business data, plugins, and the workflows that make a generic assistant actually useful inside one specific company. This was the first time Microsoft gave a clear, supported answer to the question every customer was already asking us: "Can we make it talk to our data?"
  • Copilot for Azure (in preview) put a chat-driven companion into the Azure portal itself — an assistant that understands your role and your resources and helps you design, operate, and troubleshoot infrastructure. For the people who actually run the cloud rather than demo it, this was quietly one of the more practical things on stage.
  • Copilot for Service and Copilot in Dynamics 365 Guides extended the pattern into customer service and frontline work, grounding the assistant in a company's own knowledge bases.

The connective tissue here is grounding. Across all of these, Microsoft's pitch was the same: a copilot is only as good as the data and the guardrails behind it. That's a message we'd been preaching to Mittelstand customers for a year, so seeing it become the official party line was validating — and it made our consulting conversations a great deal easier.

A laptop on a desk

Microsoft Fabric finally went GA

If the Copilot news was the headline, Microsoft Fabric reaching general availability was the announcement that immediately changed sentences we could say to clients.

Fabric is the unified analytics platform built on OneLake — the idea being that Power BI, data engineering, data warehousing, real-time analytics, and data science all sit on one lake instead of four disconnected products bolted together with pipelines and prayer. We'd spent since 2022 telling Mittelstand customers a version of this story as a future: that they didn't need a Synapse-plus-Power-BI-plus-Purview-plus-a-data-lake mess. With GA, the future became a thing we could put on a statement of work. "Yes, we'll standardise your analytics on Fabric" went from aspiration to deliverable overnight.

Windows AI Studio and the local-AI thread

Slightly off to the side of the keynote, but interesting to us, was Windows AI Studio: a toolkit that brought Azure AI Studio's tooling down onto the local machine, with a catalogue of models — including Meta's Llama 2 and Stability AI's Stable Diffusion XL — that developers could fine-tune, customise, and deploy to run locally and offline in Windows apps, with a Visual Studio Code extension on the way. It was an early, concrete signal that "AI" wasn't going to mean only "a call to a datacentre." Some of it would run on the device in front of you. That thread — small models, on-device, offline — has only grown since.

Microsoft also pushed Azure AI Studio (the precursor to today's Azure AI Foundry) further into preview as the place to build, evaluate, and safety-test generative apps, and extended its Security Copilot work into Defender XDR and Sentinel for GenAI-assisted investigation. The shape of the modern Azure AI stack was already visible here, even if half of it was still in preview.

The quiet headline: Microsoft's first custom silicon

Now the part we keep coming back to.

For all the Copilot noise, the announcement with the longest shadow was Azure Maia 100 and Azure Cobalt 100 — the first chips Microsoft has ever designed in-house for hyperscale Azure datacentres.

  • Maia 100 is an AI accelerator built on TSMC's 5-nanometre process, with roughly 105 billion transistors on a reticle-sized die. Microsoft showed it racked up with a custom "Sidekick" closed-loop liquid-cooling sidecar designed to retrofit into existing datacentre rows. It's the chip aimed at training and running large models — the OpenAI models, Copilot, the lot.
  • Cobalt 100 is a 128-core, Arm-based CPU — Microsoft described it as the first CPU it designed specifically for the Microsoft Cloud — built for general-purpose workloads, and already powering parts of Teams, Azure Communication Services, and Azure SQL.

Nadella was candid about the rollout: Microsoft would put Maia under its own workloads first and scale to third-party workloads afterwards, with both chips beginning to reach Azure datacentres in early 2024. This wasn't a product you could rent the next morning. It was a statement of intent.

And that intent is the whole point. Up to now, the economics of every AI workload anyone built on Azure were ultimately set by one external supplier's roadmap and pricing. By designing its own accelerator and its own CPU, Microsoft was buying itself control over the cost, the supply, and the efficiency of the cloud our customers run on. For us, sitting in Germany advising companies on where their data and their compute should live, that's not a spec-sheet curiosity — it's a structural change in the platform's economics, and it went straight onto every customer-architecture deck we own.

If you want a single reason this conference still matters in hindsight, it's this: the custom-silicon story that started as a single rack on the Ignite 2023 stage is a direct line to the Maia and Cobalt generations Microsoft has shipped since. Everything Microsoft has said about model optionality and controlling its own AI economics traces back to the moment it decided it would rather build the chip than only buy it.

Rows of servers in a modern datacentre

Why it mattered for us

Strip away the rebrand theatre and Ignite 2023 left us with three things that genuinely changed how we work and what we tell customers.

Fabric GA closed a story we'd been telling on credit. For two years we'd argued that Mittelstand analytics didn't have to be a four-product Frankenstein. GA let us stop saying "soon" and start writing it into contracts. One product, one lake, one bill.

The Copilot unification gave our customers permission to budget. As soon as Bing Chat became Copilot and Copilot for Microsoft 365 went GA with a real price, the conversation shifted from "what is this?" to "what's it cost and what does it touch?" That's the question you want customers asking, because it means they've decided it's real. If 2023 was the year Microsoft renamed everything to Copilot, 2024 was the year customers started budgeting for it — and we spent a good chunk of 2024 helping them do exactly that.

The silicon was the one to watch, and we said so at the time. Maia and Cobalt didn't sell a single customer anything in November 2023. But they told us where the platform was heading: toward a Microsoft that controls more of its own AI cost structure, which over a multi-year horizon is exactly the kind of thing that decides whether building on Azure stays affordable for a German SME. We were glad to be in the room for the Copilot keynote. We're gladder, in hindsight, that we paid attention to the metal.

Sources & further reading

Detailed Maia 100 specifications (transistor count, process node, cooling) are as disclosed by Microsoft and subsequently elaborated at Hot Chips; the exact attendance figures are approximate and drawn from contemporaneous reporting.

Image credits

All photos are used under their respective Creative Commons licences; we are grateful to the photographers.

Microsoft Ignite Azure Maia Azure Cobalt Copilot Copilot Studio Microsoft Fabric Custom Silicon Azure AI
BACK TO BLOG