Two years ago, the dominant narrative in Silicon Valley was that "Google will be disrupted by ChatGPT." However, over the past year, the market has reassessed this story. From its stock price to its products, from its models to its ecosystem, Google is executing a historic comeback in the AI era in 2026, even launching a dual-TPU strategy for the first time to directly challenge Nvidia.
Today, the annual Google I/O developer conference kicked off at the Shoreline Amphitheatre at its headquarters in Mountain View, California. This marks Google's largest-ever group update of AI products, ranging from the multimodal flagship Gemini Omni and the high-performance Gemini 3.5 Flash to the consumer-grade, always-on agent Gemini Spark, the AI programming platform Antigravity 2.0 positioned against Claude Code, and the new generation of dual-chip TPU series.
This product launch unfolds against the backdrop of Wall Street repricing Google's AI capabilities. Google's stock price has surged from $162 to nearly $400 within a year, with its market capitalization briefly exceeding $4.8 trillion, making it the top performer among the top five US-listed companies by market cap.
Among the extensive array of product launches at this year's I/O, four core products have become the industry's primary focus.
**Gemini Omni: The Most Significant Technical Breakthrough** Gemini Omni represents the most substantial technical breakthrough at this I/O conference. Demis Hassabis, CEO of Google DeepMind and a Nobel laureate in Chemistry, defined it as the starting point for a model that "creates any output from any input," signifying a major leap in world understanding, multimodality, and editing capabilities. Unlike previous models that merely "understand content," Omni can genuinely "create content"—it comprehends physical laws, reasons about "what should happen next" when generating videos, and possesses a deeper understanding of physics, visuals, and complex concepts.
Omni accepts any combination of images, audio, video, and text as input, with video output being the initial release, followed by image and audio outputs. It doesn't just generate videos based on prompts but can also integrate user-provided character images, scenes, or hand-drawn sketches into a unified output. Demonstration scenarios include: conversational editing of user-uploaded real videos (changing environment, style, camera angles), real-time animation generation from hand-drawn sketches, visualizing knowledge as a 'marble world knowledge' physical narrative, and scientific reasoning and display of protein folding.
Omni is deeply integrated into the Gemini App, allowing users to upload any photo or video and edit it using built-in templates. Users can also create AI avatars that look and sound like themselves and place them into action scenes. The initial Gemini Omni Flash is globally available starting today for Google AI Plus, Pro, and Ultra subscribers, simultaneously launching on Google Flow and YouTube Shorts. YouTube Shorts and YouTube Create users receive it for free. The more powerful Omni Pro version will follow.
**The First Consumer-Grade, Always-On Agent** AI agents are undoubtedly a key term for the AI industry in 2026. Gemini Spark is also the I/O release with the most profound impact on ordinary users this year. Google officially positions it as "your personal agent: it takes action on your authorization to help you 'navigate digital life.'"
The official statement: "Spark represents a major shift for Gemini, from an assistant that can answer questions to a partner that proactively completes real work for you." Powered by Gemini 3.5 Flash, Spark possesses cutting-edge agent execution capabilities. It runs on dedicated Google Cloud virtual machines, operating 24/7, continuing to work even when users turn off their computers, and is accessible via the Gemini App on any device.
Official demonstration scenarios: A user entrusts Spark with all information for a wedding, such as the guest list, registry, and Pinterest inspiration board. It assists in tracking RSVPs, contacting vendors, and summarizing preparation progress. Another demo: A user is unavailable to handle emails over the weekend; Spark automatically scans the inbox, summarizes all replies, and sends follow-up reminders to guests who haven't responded—without the user needing to perform any step.
More importantly, Spark seamlessly integrates with Google's ecosystem products: connecting the full Workspace suite of Gmail, Docs, Calendar, and Drive, with plans to expand access to third-party services via the MCP protocol this summer. Users can customize "Skills" to perform periodic tasks.
Spark begins rolling out to trusted beta users today; a Beta version is planned for US Google AI Ultra subscribers next week. Concurrently, Android Halo will launch later this year, displaying agent work progress in real-time as a halo at the top of the phone screen.
**Google's Most Powerful Agent Arrives** Gemini 3.5 Flash is Google's key bet on merging "frontier intelligence" with "agency," serving as the common engine for new features like Gemini Spark, Antigravity 2.0, and Search AI Mode. Google officially positions it as "our most powerful Agent and coding model to date." The official statement: "Outperforms Gemini 3.1 Pro on almost all benchmarks while running at the speed characteristic of the Flash series—four times faster than other frontier models."
Google showcased Gemini 3.5 Flash's capabilities with a set of benchmark data:
Terminal-Bench 2.1 (code terminal capability): 76.2%, better than Gemini 3.1 Pro's 70.3% GDPval-AA (economic value knowledge work, Elo score): 1656, significantly surpassing Gemini 3.1 Pro's 1314 MCP Atlas (multi-step MCP workflows): 83.6%, surpassing Claude Opus 4.7's 79.1% CharXiv Reasoning (multimodal chart comprehension): 84.2%, leading all competitor versions...
Gemini 3.5 is fully available starting today: on the Gemini App and Search AI Mode (for all users), Google Antigravity 2.0, the Gemini API (Google AI Studio and Android Studio), the Gemini Enterprise Agent Platform, and Gemini Enterprise. Additionally, Gemini 3.5 Pro is in internal testing, expected to launch next month.
**Google Search: From Information Engine to Intelligent Agent** Notably, the search updates at this I/O represent the most structurally significant transformation since Google Search's inception. Google has fully infused Gemini's agent capabilities into search, evolving Search from an 'information retrieval tool' to an 'intelligent agent' capable of proactive thinking and task execution.
The new search box, leveraging Gemini 3.5 Flash capabilities, predicts intent in real-time while typing and provides direct answers to complex natural language queries. AI Mode is now fully available to all users. The new information agent service can work for users around the clock, proactively integrating information from multiple sources.
Significantly, the agent service can not only search for restaurants, flights, and hotels for users but also directly complete bookings based on user requests. The process is closed-loop within the search interface, completing purchases directly via Google Pay within the ecosystem or redirecting to merchant websites for checkout.
The Gemini App launched globally today on Android, iOS, and Web with a new Neural Expressive design language: fluid animations, vibrant colors, haptic feedback, and new typography. Key information is bolded and placed at the top, embedded with images, interactive timelines, and voiceover videos. Gemini Live is integrated into the core experience, allowing seamless switching between typing and voice conversations without needing a full-screen interface, enabling users to "discuss complex ideas at their own pace without interruption."
Furthermore, Google has upgraded almost every product in its ecosystem with AI.
The Google Daily Brief service automatically organizes a user's Gmail, Calendar, and Tasks, proactively providing a personalized schedule summary and to-do suggestions at the start of each day, listing priorities and proposing next steps. It is available starting today for paid subscribers, expanding to YouTube and Gmail this summer.
AI Inbox automatically analyzes the inbox, distilling to-do items into a 'Suggested To-Dos' list, categorizing important emails as 'topics needing follow-up.' Gmail Live allows users to search emails conversationally, launching this summer for US AI Pro and Ultra users (Android and iOS, English).
Google Pics offers AI image generation and design tools for creating posters, flyers, infographics, and more, directly integrated into the Workspace workflow. Google Labs officially launched Google Flow, an AI video generation and editing tool for creators and professional users, powered by Gemini Omni. Flow adds an Agent mode supporting batch video editing and intelligent material referencing; it introduces character consistency management to maintain consistency of characters, clothing, and scene styles across long video sequences, addressing the "character drift" problem.
The YouTube video platform also introduced an Ask YouTube feature, allowing users to pose complex search queries to Gemini, with the AI precisely locating the most relevant specific clips within each video. Gemini Omni Flash has landed on the YouTube Shorts short-video platform.
Notably, Google also launched a Universal Cart service, allowing users to manage products from major retailers like Nike and Best Buy within a single interface. AI features include: price monitoring (e.g., tracking the lowest price in the past 30 days), compatibility detection (e.g., automatic alerts for incompatible motherboards and processors), and credit card benefit prompts. It will be available to US users this summer in the Gemini App and Search, indicating Google's intent to comprehensively manage user online consumption behavior within its ecosystem.
**First Dual-TPU Strategy to Challenge Nvidia** Google has broken with its decade-old convention of using general-purpose chips, implementing a "dual-chip strategy" for the first time in its eighth-generation self-developed AI chips. The underlying driver for this strategic adjustment is that large models have fully entered the "Agent era" of multi-agent parallelism and complex chain-of-thought collaboration. The traditional single-chip topology can no longer simultaneously handle intensive model training and explosively growing inference traffic. To build an absolute moat in cost and energy efficiency for its enterprise and own services, Google must completely separate the hardware foundations for training and inference.
Among them, the TPU 8i (with deep involvement from MediaTek) is tailored specifically for large-scale inference tasks. By significantly increasing on-chip SRAM and network interconnect bandwidth, it reduces network latency for chain inference by over 50%, serving as the high-concurrency, low-latency inference engine for the Gemini 3.5 service system. The TPU 8t (with deep involvement from Broadcom) is extremely focused on ultra-large-scale pre-training of frontier models, capable of seamlessly connecting 9,600 chips into a supercluster with formidable computing power. It will be fully available to enterprise customers via Google Cloud in pay-as-you-go and reserved instance formats.
To allow this powerful set of TPU chips to completely break free from the physical constraints of the Google public cloud ecosystem and directly challenge Nvidia's absolute monopoly in the computing power market, Google officially announced a major cross-industry partnership with the world's largest asset management firm, Blackstone Group, on May 18. Blackstone injected $5 billion in initial equity capital and holds a majority stake. The two companies jointly established an independent new AI infrastructure company, with Google Cloud veteran Benjamin Treynor Sloss personally serving as CEO.
This joint venture pioneered a unique independent cloud model of "Compute-as-a-Service," allowing global enterprises to directly and flexibly rent Google's TPU computing clusters within Blackstone's vast energy and data center ecosystem. The company plans to launch its first ultra-large-scale computing capacity of up to 500 megawatts by 2027, aiming to directly challenge the Nvidia-backed computing power upstart CoreWeave with a cost-effective, non-Nvidia chip ecosystem. This move signals a comprehensive reshuffle in Silicon Valley's capital-intensive computing power sector.
**More Granular AI Subscription System** Google has comprehensively upgraded its AI subscription system, introducing more granular pricing tiers. The newly added basic-tier AI Plus ($7.99/month) focuses on high value for money, integrating Gemini 3.5 Flash and unlocking Gmail AI Inbox and Daily Brief features. The advanced-tier AI Pro ($19.99/month) adds YouTube Premium Lite, real-time AI collaboration for Docs/Gmail (Live series), and exclusive imaging tools on top of that.
Targeting high-frequency developers and tech enthusiasts, the flagship AI Ultra tier has been fully restructured with price reductions. The newly introduced $100/month Ultra tier includes 20TB of storage and full YouTube Premium, exclusive access to the latest autonomous Agent engine Gemini Spark, and provides 5 times the Antigravity usage limit of the Pro tier (with an extra $100 credit for first-time users before May 25). The original top-tier price has been reduced from $250/month to $200/month, with the Antigravity usage limit skyrocketing to 20 times that of the Pro tier, and exclusive access to all frontier features like the Project Genie world simulator.
**Conclusion: Completing the Comeback in the AI Era** Two years ago, "Google will be disrupted by ChatGPT" was the most popular narrative in Silicon Valley. But over the past year, this story has been reassessed by the market. From stock price to products, from models to ecosystem, Google is completing a historic comeback in the AI era in 2026.
This competitive dynamic is also reflected in Wall Street's repricing of Google. Over the past year, Google's stock price rose from $162 to $408, a gain of over 145%. For the full year 2025, it rose 65%, its largest annual gain since 2009, even surpassing Nvidia and making it the strongest performer among the "Magnificent Seven" tech giants. Google's current market capitalization has exceeded $4.6 trillion.
As Google's core AI product, the Gemini App's monthly active users have experienced explosive growth over the past year, now surpassing 900 million. The integration of AI technology with Google's core search business has also been successful: the "AI Mode" within search has surpassed 1 billion monthly active users, while "AI Overviews" has reached over 2.5 billion people globally.
This massive user base further solidifies Google's unshakable ecosystem landscape. It is noted that Google currently has five super products with over 3 billion users (including Google Search, Gmail, Android, Chrome, and YouTube); its gold-tier product lines with over 1 billion users have expanded to 12, covering all aspects of users' daily lives from Maps and Photos to Drive and Calendar.
This comeback is not because Google defeated OpenAI in a single product point—in fact, ChatGPT's total user base still leads, and OpenAI's annualized revenue exceeds $25 billion.
But Google wins with its unparalleled full ecosystem. It possesses AI Overviews' 2.5 billion monthly active users, Android's 3 billion devices, Gmail's 1.8 billion users, and YouTube's global content moat. Gemini doesn't need to convince users to download a new app; it's embedded in every product users already use daily. The self-developed TPU chip roadmap gives Google a long-term advantage in computing power costs that OpenAI cannot replicate.
The true significance of I/O 2026 is: Google has completed the strategic switch from defense to offense. Gemini Spark targets consumers, Antigravity 2.0 targets developers, Generative UI targets ordinary users, Universal Cart targets commercial scenarios—every product line is using AI to redraw its territory, and every new entry point is funneling competitors' users into Google's AI ecosystem.
Comments