Alphabet Inc. concluded its annual Google Cloud Next 2026 conference in Las Vegas today, unveiling a comprehensive suite of artificial intelligence hardware and software designed to transition enterprise customers into what CEO Sundar Pichai termed the "agentic era." The announcements centered on the debut of eighth-generation custom silicon, a deepened infrastructure partnership with Nvidia Corp., and a unified platform for autonomous AI agents.
The company introduced its eighth-generation Tensor Processing Units (TPUs), bifurcating the product line into the TPU 8t for model training and the TPU 8i for inference. According to technical specifications, the TPU 8t scales to 9,600 chips per superpod with two petabytes of shared high-bandwidth memory, while the TPU 8i features 288GB of high-bandwidth memory and 384MB of on-chip SRAM. Mark Lohmeyer, Vice President of Compute and AI Infrastructure, stated that these new systems deliver up to an 80% improvement in performance-per-dollar compared to previous generations. Alphabet also confirmed an expanded co-design agreement with Broadcom Inc. for the TPU v8 "Sunfish" training chip, targeted for 2027, and a new collaboration with Broadcom on "Cloud Network Insights," a service for monitoring end-to-end performance in AI-driven networks.
In a significant expansion of its third-party hardware offerings, Google Cloud announced the general availability of A5X bare-metal instances powered by Nvidia’s new "Vera Rubin" architecture. These systems are designed to reduce inference costs by 10x while increasing throughput per megawatt. The infrastructure utilizes the new "Virgo Network" architecture, developed with Arista Networks, which allows Google to scale GPU clusters to 80,000 units at a single site. Additionally, the company launched the first Confidential Computing service for Nvidia Blackwell GPUs, enabling encrypted processing for sensitive enterprise data.
On the software front, Alphabet rebranded and consolidated its AI development tools into the Gemini Enterprise Agent Platform. This new ecosystem replaces Vertex AI and includes "Agent Studio," a no-code environment for building autonomous agents capable of executing multi-step business processes. Thomas Kurian, CEO of Google Cloud, demonstrated agents performing complex tasks such as supply chain reconciliation and automated threat remediation. The platform is supported by the new Agent-to-Agent (A2A) protocol, which facilitates cross-platform communication between different AI systems.
To support this technological shift, Pichai reiterated that Alphabet’s total capital expenditure for 2026 is projected to be between $175 billion and $185 billion. The company also established a $750 million fund to accelerate the adoption of agentic AI among its global partner network. Internal data shared during the keynote revealed that paid monthly active users for Gemini Enterprise grew 40% quarter-over-quarter in the first three months of 2026, with approximately 75% of Google Cloud customers now utilizing the company’s AI products.