Microsoft buys OpenAI

well.. 49% of it. And they're not done yet.

11 Mar 2023 | in ai

10 Jan 2023

If they close it, I think it's a steal for Microsoft (eg compare with Linkedin's full acquisition at a mature stage for $26B vs potential of OpenAI) - and OpenAI's best bet at AI "domination" (or as close as it gets), at the same time.

My mind has been racing for a while about what this association can mean - for the future of AI, business and the economy (GDP) at large...

It was already clear that Microsoft's plan was to leverage the technology to improve their products and services.

They had a strategic relationship with OpenAI since 2019 and were on the cap table already, albeit as a minority shareholder. I saw this as a "Kickstarter investment".. investing, but primarly to get prime access to the technology.

Github Copilot: a successful Proof Of Concept

And they have already started leveraging OpenAI's technologies in their products, such as Github Copilot - an "AI pair programmer", or "AI code writing assistant", ie auto-complete for code.
(Microsoft bought Github - a code repository, where developers store their code - in 2018 for $7.5B. Copilot is a paid-for add-on to Github)

I'm writing this note in VS Code (a Microsoft product) - the IDE of choice for many (most) developers these days it seems.

IDE, or Integrated Development Environment = the app that programmers use to write code


And I have the Github Copilot (GC) extension installed.

I have been using GC every day since November !ai/github-copilot - and I love it. It saves me time and keeps impressing me. Is it perfect? No. But it's better than without it, and it will keep getting better.

Before that, I had seen only praises for Github Copilot (in my algorithmic and/or mental bubble perhaps?).

It was announced in 2021 and came out of beta last June.
Actually it was Andrej Karpathy (former Head of AI at Tesla) on Lex Fridman's podcast, saying he "wouldn't code without it anymore" that made me try it.
This is a strong statement from someone like him on a new product.
And like him, after a couple months of use, I couldn't code without it anymore.

A perfect example of Product-Market Fit !startups/product-market-fit.

Side note: the quasi-dominance of Microsoft (with VS Code and Github) in the developer space is mind-blowing to me.
Especially when considering what Microsoft was 20 years ago (the antithesis of open-source).

Satya Nadella will go down as one of the greatest Tech CEOs of all time.
His turnaround of Microsoft, and 3D chess moves of acquisitions, are incredible.
I recommend his autobiography "Hit Refresh":

So Microsoft + OpenAI's POC (Proof Of Concept) with developers is a success.

Now onto the business world.

Business world

Microsoft has ~85% of the market share in the desktop Operating System market.

Outside of the Tech industry, most people use Windows & the Office suite for productivity.

In other words, the vast majority of business computers, ie what powers knowledge workers across the world and thus the global economy (at a low level).

Now let's take what Microsoft has done with Github Copilot in the Developers' world in a short amount of time and apply it to the Business world - and think long-term...

There are synergies across the entire MSFT portfolio of products, with OpenChat-like and DALL-E like capabilities anywhere & everywhere.

  • in the Office suite (Word, Excel, PowerPoint, Outlook, etc.)
  • in Linkedin
  • in Windows (desktops and tablets), ie embedded in everything you do on your computer
  • in a reborn Bing (currently a "ghost" search engine), which will be able to take the fight up with Google, with a shot at becoming the main (AI) search engine.

So the foundation of productivity for most computer workers will be boosted - just like what Github Copilot has started doing with developers.

Let's not forget also that:

  • OpenAI has started to cover the array of generative AI (text and images - video soon I imagine) but also other areas of useful AIs like speech-to-text (with Whisper which is really good - see !ai/whisper, tested myself with !projects/transcribee).
  • all of these are still in their infancy (ie just released out of Beta last year or open to the public as "Research release")
  • other projects & products are brewing under the hood
  • the pace of progress is accelerating
  • OpenAI provides its AIs as APIs so an entire ecosystem of companies will license its technology to build new features or entire companies on top of it.

Now where it will become even more interesting is that I expect at some point in the future, OpenAI will be able to have its AIs trained on the full scope of Microsoft-generated content.

Or in other words, each company will be able to train its own AI on its own data.

So the "AI-Assistants everywhere" will be able to generate content and answers that are specific to the company, its products, its customers, its industry, its brand guidelines, etc.

This means that all knowledge workers will be empowered to do more, faster and better.
Increasing productivity and efficiency of all workers, and thus the economy at large (in terms of GDP).

As this early stage of Github Copilot (it will improve as it keeps learning) I see time savings of say 20% - meaning 1) I can do more and 2) this number will increase.

Translate this now from low-level, at scale, to high-level - what will the impact be on global GDP ($100 Trillion as it stands)?

What will be the impact on Microsoft's revenues, valuation and stock price?

The potential is huge.

Will some jobs get lost in process? Sure.

But mostly:

1) AI will not replace your job, People using AI (better) will replace your job.
2) New kinds of jobs will appear, and I believe the uplift in productivity, thus in economic results, will be so great that the net effect will be positive (which I formulate here as a conservative understatement - the potential is far greater).

Now, could Microsoft have done that without buying a full 49% of OpenAI?

Sure, but at what cost?

And they would have benefitted only modestly from a financial standpoint of the position they will bring OpenAI to.

What's in it for OpenAI?

  • getting the compute then need (ie server time from Microsoft Azure where their stack runs) / currently spending $3M/day!
  • further/deeper work with Microsoft for a purpose-built AI stack using Azure, improving its AI performance and scalability..
  • a safer play than going on their tiny own against Google & Co.
    but most importantly I think:
  • a shortcut to get from “hype product used by a niche” to “dominating AI on the largest share of business computers”, as quickly as possible.

How long will it take?

Considering Microsoft's size, and the risks involved in deploying something too early to a mass of non-tech-savvy users, I would say a couple of years until we see a true impact. I might be (way) off though 🤷🏻‍♂️