On 4th May 2023, a leaked memo from a Google employee declared that the company “has no moats” when it comes to AI (artificial intelligence).

A moat, in business strategy circles at least, means a competitive advantage that you can use to protect your business model. It means different things in castle design – in which I am not an expert.

The memo called out a few areas that the author believes are major threats and risks to Google:

  • The ability to train new learning models on devices such as mobile phones and fine-tune AI models on mainstream laptops
  • The free availability, open-sourced, of a wide range of models of different sizes and capabilities
  • Some of the most impressive AI capabilities are non-commercial

Days passed awkwardly without a response from Google. It seemed liked the memo was an accurate reflection of the panic inside Google.

Where was Sundar Pichai, Google CEO, to calm the waters?

This was my reaction to the memo and subsequent silence, which I shared with a friend:

I don’t understand Google’s self-doubt here. They absolutely do have moats – they have more semantic knowledge of everyone’s lives than any other company, activities via GMail, locations via Maps, interests via watching history on YouTube, everything via Search history, etc, they also have a deeper and longer history of developing AI than any big tech company.

Google’s response to ChatGPT should have been simple: they’ve allowed access to the world’s information through a simple text prompt since 1998, all of their products already leverage AI heavily but seamlessly (GMail autocomplete was launched back in 2019) and they have a joined up product ecosystem with millions of users, which is a really hard thing to do.

This isn’t a failure of Google tech capability, this is a failure of PR.

I was severely disappointed in Sundar and Google. This should have been their deadpan, “oh we’ve been doing that for years,” moment.

But I’d forgotten one important thing.

Google I/O 2023

I/O is Google’s annual developers conference and it was scheduled for 10th May, 6 days after the memo was leaked.

I/O has been an evolutionary event for the past few years. 2022 focused on augmented reality (AR) SDKs (software development kits) in response to the AR wearable device from Apple expected soon and a bunch of upgrades to Google’s mobile operating system, Android, accompanied by a refresh of the Flutter cross-platform UI toolkit.

Nothing groundbreaking.

But I/O 2023 had a very different feel this year, clearly in response to ChatGPT and the leaked memo.

Sundar reminded the audience of the existing AI capabilities already baked into GMail, evolved and matured over the last decade, and showed the next iteration of this: “Help me write”.

He then showed a stunning demo of the next generation of navigating with Google Maps.

Next he showed how AI could be used to entirely create Google Docs and Sheets from scratch.

And this really demonstrates where Google’s moats are.

You can watch the entire keynote below:

The GPT-4 large language model (LLM) is mind-blowingly impressive, but it’s a solution looking for a problem – it’s not a “product” by itself, instead it’s infrastructure, like the roads and railways.

There is a concept called the Commoditisation Scale, which describes the trajectory that all technologies take – from novel ideas to ubiquitous and utility (think electricity, or computer processors) and LLMs will very quickly become commoditised and invisible utilities.

The real moats are built on top of these commodities and Google has spent a long time weaving AI carefully into its products.

I also agree with the presenters of the Hard Fork podcast shown below that just because something is open sourced and free doesn’t make it successful, with the popularity of Microsoft Windows over the free, open source Linux being a clear example of this.

The winners of the AI race will be those who can create meaningful products with AI.

Microsoft is being unfairly hailed as a competitor in this space but Microsoft’s track record with AI is horrible. Even today, after a $10bn investment into OpenAI I still get emails going into Junk because I allegedly “don’t usually get emails from this address”, even though I do and I’ve had several meetings with that person which are in Microsoft Outlook Calendar and Microsoft Teams and recorded in Microsoft Stream. It doesn’t take AI to work that out.

Anyone who’s used Microsoft Viva also knows that Microsoft and AI do not go together.

Where Microsoft can pose a challenge to Google (and Amazon) is by offering private, enterprise versions of GPT-4 through their cloud service Azure. This could become a huge revenue stream for Microsoft but again will only be as infrastructure, and the margins on infrastructure of very poor.

The real money is in the application, and Microsoft are not good at apps.

My money remains on Google.