Google I/O 2026 – Blocking and Tackling

A solid, if unexciting outing.

• There are no fireworks at Google I/O this year, and what we find are a series of launches and changes that aim to keep users engaged, revenue growing and keep its rivals in their relative boxes.
• Top launches and changes include:
  • First, Usage growth: which remains very rapid and has grown to 3,200tn tokens per month in 2026, or up 6x from 2025.
  • There are now 8.5m monthly active developers, and the Gemini App has 900m MaUs, up 2x on 2025.
  • Crucially, search is at an all-time high in terms of usage, adding weight to Google’s claim that AI mode entices users to use search more rather than less.
  • There are now 15 Google Digital Ecosystem services with more than 1bn MaUs and 5 that have more than 3bn MaUs.
  • Second, Gemini Omni: which Google claims advances its journey towards superintelligent machines, but in reality is an iterative improvement aimed at keeping its rivals from invading its patch.
  • Omni is what is now termed a “world model”, which aims to simulate reality so that it can more accurately understand the world and offer better answers and more intuitive services.
  • The new model shows improvements against the usual benchmarks and offers greater ability to generate a wide range of content.
  • However, the real action at I/O was in Gemini 3.5 and Spark.
  • Third, Gemini 3.5 and Spark: which are a new model family and an agentic system to keep usage of Google AI inside the ecosystem.
  • The first model to roll out is Gemini 3.5 Flash, which is already available worldwide and also offers enough of an improvement to keep it in contention with OpenAI and Anthropic.
  • 3.5 Flash has clearly been optimised for coding, which fits with where most of AI usage is at the moment, again demonstrating Google reacting to the market environment and moving to keep competitors at bay.
  • Gemini Spark is essentially a consumer product and is the first iteration of a personalised AI agent that executes tasks for users rather than just recommends them.
  • Initially, this requires the user to live his or her digital life entirely within Google, but with time, Spark will be connected to more and more external services making it much more useful.
  • This is important as this is Google’s offering for the “age of agents”, and if it works well, it could see a lot of traction.
  • As always, the demos were great, but this is a very far cry from what it is like in the hands of real users.
  • For example, it takes Claude 20x longer to search my Office 365 database and update my calendar (multiple permissions needed) for a meeting location than if I just do it myself.
  • Fourth, Search: which has been updated to mash together search, AI overviews and AI mode into one user experience that takes the usage of search much further.
  • Users can now set complex search queries with complex criteria and have them running in the background, and only send alerts when all of the criteria have been met.
  • Search is now also able to generate content, such as pictures and animations, to answer questions and explain concepts.
  • Shopping functionality has also been updated with a universal commerce protocol, agents that can make payments as well as a universal shopping cart.
  • These aim to deliver the ability to shop from Google services as well as monitor prices and have agents make purchases under the tight control of the user.
  • Crucially, Amazon has signed up for this, as well as several payment players and other social networks, meaning that 3^rd party engagement is better than I would have expected at this stage.
  • Fifth, Rate limits: which were not announced at I/O but sent round to AI Pro subscribers after the event.
  • There will now be a usage limit that resets after 5 hours (just like Claude).
  • This limit now applies to all Gemini users, but as a Pro subscriber, the 5-hour limit is 4x higher than that of a free user.
  • AI credits in Flow and Antigravity have been removed and replaced with usage limits, but Google promises that the amount of usage that one gets remains broadly the same.
  • This is about trying to control the runaway usage growth that results from unrestricted usage, for which Google needs to spend $190bn in capex in 2026 alone.
  • It is also about limiting the stuff that Google has to give away for free and encouraging those users to pay to use its AI services.
  • This obviously does not apply to the AI compute that is going into the new search functions, which will remain monetised by advertising.
  • Sixth, AI Glasses: which is developing at a very slow pace.
  • Here, Google continued to tease the display glasses with no product launch date, but did launch an audio-only version with Samsung, Gentle Monster and Warby Parker.
  • This is the equivalent of the Meta Raybans, and shows that Google is recognising that the market for smart glasses is growing fast, where it does not want to be left behind.
  • Again, a great demo of how useful these glasses can be when used with Google ecosystem services was made, but there is no way to tell how good they really are until they are in the hands of users (Q4 2026).
  • I think that Google has a large advantage in this space over Meta, as its ecosystem is far stronger for the use cases of smart glasses than Meta’s is.
  • However, it needs to execute faster, as at the moment, Meta completely dominates this market.
• The net result is that this I/O is almost entirely commercially driven.
• With free cash flow sinking to decade lows as a result of the capex, Google needs to show a return on investment, which is what all of these product launches and the new usage limits are all about.
• It is also about keeping Anthropic out of its territory (focus on coding) and cementing its grip on the consumer AI ecosystem before OpenAI can get a foothold.
• Google has come a long way from its very shaky start in generative AI in 2023 and looks to be in a good position.
• It must now continue the fast revenue growth and earn a real return for shareholders who are footing most of the $190bn in 2026 (and more in 2027) of capex investments being made.