Driving enterprise transformation with new compute innovations and offerings

In the last 12 months, we’ve made incredible enhancements to our Compute Engine platform. This is driven most notably by new fourth-generation compute instances and Hyperdisk block storage as well as major customer experience enhancements. Across all workloads, Google Cloud’s compute portfolio can help you optimize your performance and costs, while delivering enterprise-grade scalability, reliability, security, and workload consistency, helping you grow efficiently and have more to invest for innovation. Let’s explore what we’ll be announcing today at Google Cloud Next 2025. 

New and enhanced compute for every workload

C4D offers 80% higher throughput per vCPU and stronger performance
Our new C4D VMs are built on AMD’s 5th Gen EPYC processors, paired with Google Titanium’s latest advancements, and have a higher core frequency (up to 4.1 GHz). C4D delivers impressive performance gains over prior generations across a wide set of general computing workloads — up to 30% vs C3D on the estimated SPECrate®2017_int_base benchmark — helping you meet the needs of business-critical applications with fewer resources. 

For databases, C4D achieves an up to a 55% increase in queries per second on MySQL and a 35% performance improvement for Redis workloads compared to C3D. For web-serving workloads, C4D delivers up to 80% higher throughput per vCPU compared to previous generations, driving faster page rendering and a smoother end-user experience. C4D offers confidential computing and is available in 49 industry-standard shapes, with sizes ranging from 2 vCPU to 384 vCPU in three memory configurations of up to 3TB of DDR5 memory, and will include both our first AMD based bare metal offering and our new Titanium LSSD. Now available in preview, try out C4D in Compute Engine and Google Kubernetes Engine (GKE) today.

“AppLovin, a global leader in mobile advertising, is constantly looking for cutting-edge infrastructure innovations to deliver exceptional performance for our clients. Google Cloud’s C4D VMs enable us to do just that — driving a ~40% improvement over the prior generation, which leads to significant efficiency gains and latency reduction.” – Basil Shikin, CTO, AppLovin

C4 VMs enable new capabilities and greater flexibility
For demanding, low-latency tasks such as gaming, inference, large-scale data processing, and real-time workloads, our C4 machine series is expanding to enable new capabilities and configurations, including larger shapes, Local SSD, and bare metal. These new C4 shapes, built exclusively on the latest 6th generation Intel Granite Rapids CPUs, feature the highest all-core turbo frequency of any Compute Engine VM — up to 4.2 GHz.

C4 shapes with Titanium Local SSD offer improved performance for I/O-intensive workloads like databases and caching layers, achieving Local SSD latency reductions of up to 35%. New C4 bare metal instances provide performance gains of up to 35% for general compute and up to 65% for ML recommendation workloads compared to the prior generation. The new, larger C4 VM shapes scale up to 288 vCPU, with 2.2TB of high-performing DDR5 memory and larger cache sizes, enabling better scalability for databases, data analytics, and other memory-constrained workloads. Request preview access here.

H4Ds offer tremendous performance improvements for HPC workloads
Scale your HPC workloads and get insights faster than ever before with H4D VMs. These VMs are built on the 5th gen AMD EPYC CPUs and offer the highest whole-node VM performance of more than 12,000 flops, the highest per-core performance, and the best memory bandwidth of more than 950 GB/s of our VM families. H4D VMs provide 200 Gbps of low latency Titanium RDMA network bandwidth to support clusters with over 10,000 cores and plans for even more scale. Learn more in our Scientific Innovations blog or sign up for the H4D preview.

“The generational performance leap achieved with Google H4D VMs, powered by the 5th Generation AMD EPYC, is truly remarkable. For compute-intensive, highly non-linear simulations such as car crash analysis, Altair Radioss delivers a stunning 3.6x speedup. This breakthrough paves the way for faster and more accurate simulations, which is crucial for our customers in the era of the digital thread!” – Eric Lequiniou, SVP Radioss Development and Altair Solvers HPC

M4 VMs double performance for demanding SAP workloads 
Backed by Compute Engine’s memory-optimized 99.95% single instance SLA, M4 VMs offer up to 65% better price-performance and 2.25x more SAP Application Performance Standard (SAPS) compared to our previous memory optimized M3. Built on 5th Generation Intel Xeon Scalable processors, M4 VMs are certified for business-critical, in-memory SAP HANA workloads ranging from 744GB to 3TB, and for SAP NetWeaver Application Server.

Z3 for storage-intensive workloads
For I/O-intensive workloads such as data warehouses, SQL, and NoSQL databases, our Z3 storage-optimized family now features new Titanium SSDs and offers nine new smaller shapes, ranging from 3TB to 18TB per instance. We are also introducing new storage-optimized bare-metal instance which include  up to 72TB of Titanium SSDs and direct access to the physical server CPUs. Now in preview, register your interest by signing up here.

Nutanix Cloud Clusters are now on Google Cloud
We’re excited to partner with Nutanix, who selected the new Z3-metal instances to launch Nutanix Cloud Clusters (NC2) on Google Cloud. Nutanix NC2 is a hybrid cloud platform that simplifies the ability to run, manage, and operate apps, data, and AI across private and public clouds. NC2’s common operating model makes it easy to manage workloads in a consistent manner, accelerating customers’ migration to Google Cloud and helping them modernize their apps. Learn more and sign up for public preview. 

“We are thrilled to announce the private preview of Nutanix Cloud Clusters on Google Cloud, marking a significant milestone in Nutanix’s commitment to delivering flexible, hybrid cloud solutions. Google Cloud’s Z3 instance types represent a perfect foundation for Nutanix to enable performance and resilience for enterprise applications. We’re excited about our partnership with Google Cloud in empowering our joint customers with greater choice and simplicity in their cloud journey.” – Saveen Pakala, Vice President of Product Management, Nutanix

More options to optimize your VMware environment in the cloud
With Google Cloud VMware Engine, we provide one of the fastest ways to lift and transform your existing VMware estate into Google Cloud. Today, we are offering 18 additional node shapes, bringing the total number of node shapes across VMware Engine v1 and v2 to 26 — six times more node shapes than competitors. Now, you have the industry’s widest range of options to shape your capacity to your workloads’ needs and optimize your TCO. 

Storage and platform capabilities for greater scale and efficiency

Our fourth-generation compute, networking, and block storage portfolio is built on several highly differentiated foundational technologies. Titanium is a system of purpose-built custom silicon and multiple tiers of scale-out offloads that free up the CPU, enhancing performance, reliability, security and maximizing workload efficiency. It is integrated across our compute, storage, and networking offerings, which you’ve seen in a number of the announcements above. 

Recently, we also updated the Titanium ML Adapter to securely integrate NVIDIA ConnectX-7 network interface cards (NICs), providing 3.2 Tbps of non-blocking GPU-to-GPU bandwidth. In addition, Titanium Offload Processors now integrate our GPU clusters with the Jupiter data center fabric, providing greater cluster scale. 

Next-generation block storage with Hyperdisk 
Hyperdisk is Google Cloud’s workload-optimized, high-performance block storage that’s cost-efficient, easy-to-use and that delivers comprehensive data protection capabilities for your workloads. With unique capabilities like the ability to independently tune capacity and performance specific to your workloads, Hyperdisk Storage Pools enable thin provisioning and data reduction, lowering TCO and simplifying management at scale. As customers move larger and larger workloads, we are expanding Storage Pools to store up to 5 PiB of data in a single pool — a 5x increase from before. 

In addition, we are also introducing Hyperdisk Exapools, a new variant of Storage Pools purpose-built for the largest and most demanding AI training workloads. With Hyperdisk Exapools you can provision and manage block storage delivering multiple exabytes of capacity and terabytes per second of throughput for your biggest AI clusters, while leveraging thin-provisioning and data reduction to lower your TCO and simplify management. Exapools will enter preview in Q2, 2025, later this quarter. 

Hyperdisk ML has also added new capabilities, including hydrating from Cloud Storage using GKE volume populator, attaching to the latest Compute Engine instances, and performing data loading acceleration from Hyperdisk ML to run training/inference on the latest TPU VM families. Learn more in today’s AI infrastructure blog. 

Match resources to your usage patterns

Finally, we’re providing you with greater efficiency, flexibility, and control over demanding computing tasks with managed instance groups (MIGs) — collections of virtual machines that you can manage as a single entity. For example, you can now configure MIGs to use multiple VM types and it automatically finds capacity — even during periods of high demand and rapid growth. You can also use stopped and suspended VMs in a MIG with pre-initialized VMs, to save cost and accelerate application startup. We also introduced committed use discounts (CUDs) and reservation sharing with Vertex AI and Autopilot, letting you purchase infrastructure once and utilize it across multiple services.

Invest for innovation with optimized compute

Delivering infrastructure that provides the highest performance and flexibility for all of your workloads is our top commitment. From general-purpose VMs to specialized solutions for HPC, SAP, and databases, we offer workload-optimized solutions tailored to your needs, helping you unlock the innovation your business needs. Got questions? Get in touch!