Google Cloud Gets More Serious About Infrastructure At Next 2025

Google Cloud's Next 2025: A Focus on Infrastructure, Connectivity, and AI

Google Cloud's Next 2025 conference showcased numerous advancements, too many to fully detail here. For in-depth analyses of specific announcements, refer to articles by my colleagues: Matt Kimball on the new Ironwood TPU chip, Jason Andersen on Google's enterprise AI sales strategy (particularly agents), Melody Brue on the future of AI in the workplace, and Robert Kramer's insightful event preview. This article will concentrate on the most significant developments in connectivity, infrastructure, and AI.

(Note: Google is an advisory client of Moor Insights & Strategy.)

Enhanced Connectivity and Infrastructure

Thomas Kurian emphasized Google Cloud's commitment to connectivity, particularly its new Cloud WAN and Cloud Interconnect offerings. Cloud WAN leverages Google's global network for faster (40% faster, according to Google), more cost-effective performance (40% lower TCO) than traditional enterprise WANs. Cloud Interconnect facilitates high-availability, low-latency connections between enterprise networks and Google Cloud, or even between networks hosted by different cloud providers. Kurian's prioritization of networking at the analyst briefing underscores its strategic importance for Google, reflecting the growing enterprise need for seamless hybrid and multi-cloud connectivity.

Significant infrastructure investments ($75 billion this year) were also highlighted, with Next 2025 showcasing key areas of investment. Beyond AI-specific infrastructure (discussed later), these investments encompassed networking offload, new storage solutions, and a new CPU, all designed to support Google Cloud's strategy of combining hardware and software for high-performance, cost-effective solutions, especially in AI. The emphasis on "low price" is a notable shift for Google.

Strategically, Google is recognizing IaaS as a gateway to PaaS and SaaS revenue. Competitive compute, storage, and a global network attract IaaS customers, opening opportunities for platform-level and higher-value AI service adoption. This proactive approach positions Google to compete effectively with Azure's growing IaaS presence.

Next 2025 marked a more assertive challenge to AWS on the infrastructure front. Google directly compared its Arm-based Axion CPU to AWS's Graviton, highlighting Axion's superior performance. Discussions also included AWS Trainium chips, indicating a welcome increase in competitive intensity.

Ironwood TPU, Gemini 2.5, and AI Hypercomputer

The announcement of Google's seventh-generation Ironwood TPU, slated for release later in 2025, was a significant highlight. This advanced TPU boasts improvements in performance, energy efficiency, and interconnect capabilities. Matt Kimball's detailed analysis provides further insights. Ironwood's support for the vLLM library for inference, combined with Google's Pathways machine learning runtime, enables efficient, cost-effective inference across thousands of TPUs.

Gemini 2.5, currently leading Hugging Face's Chatbot Arena LLM Leaderboard, made its enterprise debut, showcasing impressive visual physics simulations. While I haven't extensively used Google's generative AI products, Gemini 2.5's performance and feedback from my network warrant further exploration.

Google's claims for the Gemini 2.0 Flash model, which balances performance and cost, are noteworthy. The assertion of 24x better intelligence per dollar than GPT-4 and 5x better than DeepSeek-R1, again emphasizes the cost-effectiveness focus. This is achieved through the AI Hypercomputer system, utilizing tailored hardware, software, and machine learning frameworks for optimized performance. Google's adoption of Nvidia's GB200 GPUs further strengthens its AI capabilities, while simultaneously developing its own solutions for optimal price-to-performance.

Strategic Positioning in AI and On-Premises Solutions

Google Cloud's emphasis on delivering high-intelligence at low cost is a strategic shift. Its strength across the computing stack—homegrown chips, high-performing AI models, open software, and improving infrastructure—is a key differentiator. Google Distributed Cloud extends this advantage to on-premises deployments, enabling enterprises to run high-performance Gemini models and Agentspace in secure, air-gapped environments.

While Google's AI agent announcements were positive, they didn't represent a significant strategic shift—at least not yet. Vertex AI and Agentspace simplify model selection, data integration, and agent deployment, and the Agent2Agent open protocol enhances interoperability. However, the current agent landscape lacks clear differentiation among major cloud providers, suggesting further development is needed.

The Hybrid Multicloud Solution

Kurian emphasized Google Cloud's open, multi-cloud platform optimized for AI. The combination of Cloud WAN, Cloud Interconnect, and on-premises Gemini deployments (potentially managed on Dell infrastructure) represents a compelling hybrid multicloud solution. Given that a significant portion of enterprise data remains on-premises, Google's approach addresses the need to leverage this data with generative and agentic AI, offering a comprehensive solution across various industries. Google's capabilities, combined with its track record of execution, position it well for future competition.

The above is the detailed content of Google Cloud Gets More Serious About Infrastructure At Next 2025. For more information, please follow other related articles on the PHP Chinese website!