The context of this particular edition
Google I/O 2026 took place on 19 and 20 May at the Shoreline Amphitheater in Mountain View, with a main keynote hosted by Sundar Pichai and Demis Hassabis. More than two hours of presentation, dozens of announcements in all directions, and one very clear obsession: positioning Google as the undisputed leader in AI against OpenAI, Anthropic and Microsoft.
Pichai opened with a figure that sums up the strategy: Google now has 13 products with more than a billion users each, 5 of which exceed 3 billion. AI Overviews in search reach 2.5 billion monthly users. This installed base is Google's nuclear weapon: no matter how many users ChatGPT has, Google can inject AI into services the whole world already uses.
As we documented in our preview article of 26 April, many announcements were expected. Let's look at what was actually delivered, what surprised, and what changes for you.
Gemini Spark: the bet on the "24/7 personal agent"
The most strategic announcement of the keynote is probably Gemini Spark, presented as a "personal AI agent available 24 hours a day, 7 days a week". It is Google's answer to ChatGPT and to the race for agentic personal assistants.
In practice, Spark goes beyond the simple chatbot. It is an agent capable of:
- Carrying out multi-step tasks (booking a restaurant, tracking an order, comparing prices)
- Maintaining a persistent memory of your preferences and habits
- Anticipating your needs based on your calendar, your emails, your history
- Acting in the background without manual intervention
This announcement reproduces the philosophy that Anthropic and OpenAI have been deploying for 18 months (the Computer Use and Operator agents), but with one overwhelming advantage: Spark is natively integrated into Gmail, Calendar, Drive, Maps, YouTube. For a Google user already committed to the ecosystem, switching to Spark will be simpler than adopting a third-party agent. As we analysed in our article on the locking-in of super-apps, this is exactly the strategy that allowed Apple to dominate the mobile ecosystem for 15 years.
Spark was not presented by Pichai at the start of the keynote, but positioned strategically after the hardware announcements (Samsung Intelligent Eyewear, TPU 8i). This suggests that Google wants to avoid the trap of "OpenAI invented the personal agent, we're copying". By presenting Spark as the logical conclusion of their integrated ecosystem (hardware + cloud + models + applications), Google implicitly claims authorship of the concept.
Samsung Intelligent Eyewear: the AI glasses finally announced
The most unexpected moment of the keynote is probably the Samsung Intelligent Eyewear, unveiled in partnership with Google, Warby Parker and Gentle Monster. Jay Kim of Samsung took to the stage to present these glasses "that you'll actually want to wear".
The announced features:
- Autumn 2026 release (probably October-November)
- Design in partnership with Warby Parker and Gentle Monster, two premium eyewear brands. The frames shown had a "vintage hornrim" or "cat eye" look, clearly intended as fashion glasses
- Audio focus initially (similar to Amazon's Echo Frames), with a built-in camera
- Native integration with Gemini, Maps, and Nano Banana (Google's image generation model)
- Connection to a Google Watch for visual notifications
The most striking demo: the presenters took a group photo with the glasses, asked Gemini to add generated elements (humorous content), and the result appeared instantly on a smartwatch. Concrete proof of the glasses + AI + watch workflow that Google is promising.
As we analysed in our article on the failure of AI hardware, Samsung and Google's bet is exactly the right one: start from glasses that people already wear, rather than creating a strange new object. It is the Ray-Ban Meta lesson applied to the Android ecosystem. If the execution follows, this product could become massive.
Gemini Omni: the new video model that wants to crush Sora
Demis Hassabis, CEO of DeepMind, came on stage to present Gemini Omni, Google's new video generation model. The announcement comes in a particular context: as we documented in our article on the great AI failures, OpenAI has just shut down its Sora app on 26 April after burning 15 million per day for 2.1 million in total revenue.
Gemini Omni capitalises on this opportunity with a different approach. Hassabis positioned it as a model that builds on DeepMind's unique expertise in "world models", those models that learn the physical laws of the world to generate more coherent videos. The demo showed video sequences where the physics of objects, lighting, and temporal coherence exceeded what Sora or Veo 3 produced.
Hassabis took the moment to slip in one of his trademark lines: "AGI is now on the horizon. If we build it correctly, it could enable human progress and flourishing beyond what we can imagine." A line he repeats regularly and which targets regulators as much as investors.
TPU 8i: the hardware that changes the economic equation
Sundar Pichai devoted a long passage to the new generation of TPU 8i (Tensor Processing Unit), Google's proprietary AI chips. A technical announcement but with massive implications.
The claimed figures:
- Faster than the previous generation (performance gains not specified but "significant")
- More energy-efficient, which reduces the cost per token
- Allows Google to distribute its training across the world thanks to a new network architecture
Google's insistence on its TPUs is probably the most important message of the entire keynote for financial analysts. While OpenAI and Anthropic depend on Nvidia GPUs (and therefore suffer their prices and their delivery times), Google makes its own chips. This gives Google a structural cost advantage that compounds over time. Over 5 years, this could represent the difference between 30% and 60% gross margin on AI services. It is probably the main reason why analysts consider Google a likely winner of the AI war, as we analysed in our article on the AI bubble.
Google Search: "the biggest update in 30 years"
Pichai announced that Google Search is getting "the biggest update in almost 30 years", powered by Gemini. AI Overviews now reach 2.5 billion monthly users.
In practice, search is becoming:
- Conversational: you can have a continuous dialogue with Google rather than a succession of queries
- Multimodal: search by image, by voice, by video, mixed freely
- Agentic: Google can now carry out simple tasks directly from the results (book, compare, buy)
- Personalised: integration with your history, your calendar, your preferences
For web publishers and SEO professionals, this announcement is worrying. If Google answers questions directly rather than redirecting to sites, the business model of millions of content creators is under threat. The transition is probably inevitable, but it will leave many economic casualties.
Google Pics: image generation in Workspace
Suz Chambers, director of Google Creative Lab, presented Google Pics, a new image generation tool built directly into Workspace (Google Docs, Slides, Sheets).
The main features:
- Image generation from a text prompt
- Editing of existing images (remove, resize, change the background or foreground)
- Automatic SynthID watermark on every generated image, to identify AI content
- Native integration in all Workspace tools
The "table stakes" angle is obvious here: Google is merely catching up with ChatGPT (which integrates DALL-E) and Microsoft (which integrates Designer into 365). As we analysed in our article on the disruption facing graphic designers, these tools built into office suites could profoundly redraw the market for micro visual services over the next 18 months.
Code Mender: AI cybersecurity arrives at Google
In an announcement that was probably underrated, Google unveiled Code Mender, an AI tool that "automatically finds vulnerabilities and patches them". Hassabis invited experts to come and test the product in closed beta.
This is Google's direct response to OpenAI's GPT-5.5-Cyber and Anthropic's Claude Mythos, whose stakes we analysed in our article on AI cybersecurity. The market for AI-assisted cyber defenders is becoming one of the most strategic enterprise segments. Google could not afford to be absent from it.
Docs Live and AI for accessibility
A particularly interesting announcement on the inclusion front: Docs Live (and its Gmail Live and Keep Live variants) lets you control these applications entirely by voice. You can say "move this section to the top" or "put this paragraph in bold and italics" without touching the keyboard.
This feature, presented as accessibility-oriented, will probably be adopted en masse well beyond people with disabilities. It is exactly the pattern we saw with YouTube's automatic subtitles: designed for deaf people, used by everyone in noisy or silent settings.
Gemini for Science: accelerating research
Probably the announcement that drew the most applause from the technical audience: Gemini for Science, a suite of tools for scientific research. The capabilities presented:
- Hurricane predictions improved on traditional weather models
- Generation of specialised code for scientific simulations
- Acceleration of research workflows in biology, chemistry and physics
This announcement follows on from what we explored in our article on AI and scientific discovery. DeepMind, which has already received a Nobel Prize for AlphaFold, continues to establish itself as the leading player in scientific AI.
Google Flow Music: the slightly odd creative bet
The announcement that puzzled the specialist press the most: Google Flow Music. The tool lets you upload a musical recording (you sing a tune, you play a few notes), then Gemini 3.5 generates complementary musical elements.
The demo was described by several journalists as "odd". The presenters waited through several seconds of awkward applause at several points. The use case is unclear: real musicians already use more powerful tools (Suno, Udio), and non-musicians do not know where to start.
This announcement illustrates a persistent weakness at Google: the company knows how to build massive platforms, but still struggles to identify what truly resonates with creative users. As we documented in our article on the AI startups that outperform, it is Higgsfield, ElevenLabs and other specialist players that capture creators' attention, not the generalist giants.
Summary table of the major announcements
| Announcement | Impact | Availability |
|---|---|---|
| Gemini Spark | Major (personal agent) | Gradual rollout 2026 |
| Samsung Intelligent Eyewear | Major (new consumer hardware) | Autumn 2026 |
| TPU 8i | Major (structural cost advantage) | Already in production |
| Gemini Omni | Significant (generative video) | Gradual beta |
| Search "biggest upgrade in 30 years" | Major (SEO disruption) | Immediate rollout |
| Code Mender | Significant (cybersecurity) | Closed beta |
| Google Pics | Standard (catching up) | Available |
| Docs Live | Standard (accessibility) | Available |
| Gemini for Science | Niche but strategic | For researchers |
What Google didn't say (and why it matters)
Several expected topics were not addressed, or were covered only very briefly. These silences are sometimes just as instructive as the announcements.
No mention of Googlebooks in the main keynote. The new premium Android laptop format, despite being teased at the Android Show the previous week, was barely touched on. Surprising for a product that was presented as a category break. Probably because the product isn't yet ready for a real demonstration.
Very few details on Android 17. Google had already covered the ground at the Android Show I/O Edition on 12 May. But the near-total absence of Android from the main keynote says something: for Google in 2026, Android is no longer the subject. AI is. The smartphone has become one platform among others for distributing Gemini.
No concrete pricing announcement for Spark. Google showed the product but said nothing about the business model. Will it be free within Google One? Reserved for premium subscribers? Included in Workspace? These silences suggest the pricing strategy has not yet been settled.
No mention of competitors. For two hours, neither OpenAI, nor Anthropic, nor Microsoft were named. That is deliberate: Google wants to put across the message that "the AI race isn't a race, it's our territory". A classic communications strategy, but perhaps a little too sure of itself given the reality of the market.
What you should take away
If you are a user, an entrepreneur or a tech decision-maker, here are the concrete implications of this keynote.
First, the integrated ecosystem is now Google's main weapon. Spark, Pics, Docs Live, Code Mender, Gemini for Science: none of these features is revolutionary taken in isolation. But together, they weave a web it becomes progressively harder to escape. If you are already with Google, you will probably stay even more deeply embedded. If you are not, the switching cost is rising.
Second, smart glasses are finally becoming serious. The Samsung-Google-Warby Parker-Gentle Monster partnership signals that the industry has learned the Ray-Ban Meta lesson. Autumn 2026 will probably mark the real commercial take-off of this category. If you are interested in this market (as a consumer or as a business angle), now is the time to inform yourself seriously.
Third, web search really is going to change. With 2.5 billion monthly users of AI Overviews and an overhaul presented as "the biggest in 30 years", traditional SEO is going to be turned upside down. If your business depends on Google traffic, you need to start thinking now about diversifying your sources of visitors.
Fourth, AGI is becoming an official topic again. Hassabis's phrase about "AGI on the horizon" is not just a slogan. It is a signal sent to regulators, investors and the public. Google is positioning itself explicitly in the race to AGI, as we explored in our article on prospective AI scenarios.
This year's Google I/O confirms one thing: the AI war is no longer a story of isolated innovations, it is a battle of platforms. And Google, with its 5 products boasting more than 3 billion users, is playing its trump cards. It remains to be seen whether the execution will live up to the promise. The next 18 months will be decisive.