Superpower Daily

Superpower Daily

Your reading space

Sign in
China-Linked Posts Do Not Explain America’s Data-Center BacklashGoogle’s Ironwood Takes an FP8 Cost Lead in New Serving Benchma...Grove City Holds 310-Acre Data Center Plan, Exposing Limits of...Morgan Stanley Raises Oracle Target by $3, Keeps Equal Weight a...Matt Clifford to Leave ARIA After Anthropic Role Draws Conflict...TeraWulf’s $3.2B AI Campus Exposes Gaps in Safety and Power Acc...China-Linked Posts Do Not Explain America’s Data-Center BacklashGoogle’s Ironwood Takes an FP8 Cost Lead in New Serving Benchma...Grove City Holds 310-Acre Data Center Plan, Exposing Limits of...Morgan Stanley Raises Oracle Target by $3, Keeps Equal Weight a...Matt Clifford to Leave ARIA After Anthropic Role Draws Conflict...TeraWulf’s $3.2B AI Campus Exposes Gaps in Safety and Power Acc...
Swipe

Infrastructure AI News

Infrastructure
Showing 1-18 of 23Results matching "GPU-as-a-Service"

Today

1 story
PartnershipInfrastructure

HPE Issues Oracle Warrants in AI Network Deal With Financing and Telemetry

The newly detailed terms add stock warrants, support, financing capabilities and network-monitoring work around Oracle’s prospective global Juniper deployment. HPE has issued Oracle warrants to buy HPE common stock under an AI networking collaboration that also includes financing capabilities, multi-year support services and joint telemetry work. Large GPU clusters rely on the network to move data among machines.

2 min readRead story

Last week

5 stories
Infrastructure

HyperVault Targets $7.4 Billion for a 1-Gigawatt AI Campus in Hyderabad

The proposed campus would give TCS’s infrastructure unit a large physical base for GPU-heavy AI workloads. The next test is converting a secured site and planned capacity into service for AI companies and hyperscalers. Tata Consultancy Services subsidiary HyperVault and its partners are targeting up to 700 billion rupees, or $7.41 billion, for a 1-gigawatt AI data-center campus in Hyderabad, Telangana.

3 min readRead story
PartnershipInfrastructure

Crusoe Signs Reported $13 Billion, Five-Year AI Cloud Deal With Jane Street

The reported agreement gives the trading firm access to Crusoe’s cloud infrastructure for AI training and inference, but it does not connect the $13 billion commitment to a specific site or amount of capacity. Bloomberg reported the private agreement September 3, citing people familiar with the matter who requested anonymity. The reported deal arrives after Crusoe’s June capacity disclosure.

2 min readRead story
LaunchInfrastructure

Broadcom Validates Five AI Model Families for VMware’s On-Premises Model Service

The update gives enterprises a path to run models from NVIDIA, Google, NEC, Alibaba Cloud and Z.ai inside their own infrastructure, with new sharing, routing and code-isolation services around them. Enterprises using VMware Cloud Foundation now have a validated route to run five additional AI model families on premises and offer them internally as a model-as-a-service.

2 min readRead story

August 2026

12 stories
FundingInfrastructure

Blue Owl Leads $2.4 Billion IREN GPU Financing Tied to Hardware Delivery

IREN has a large debt commitment for its Canadian AI-cloud buildout, but the amount becomes available alongside delivered and commissioned hardware—not as one immediate cash payment. Funds managed by Blue Owl Capital have committed $2.4 billion to IREN’s AI-cloud expansion at its Mackenzie campus. The headline sum is not an upfront payment: IREN will draw the facility as equipment is delivered and commissioned.

2 min readRead story
LaunchInfrastructure

OpenRelay Routes AI Inference Across GPUs, TPUs and Trainium Through One API

The early-access network aims to let developers buy performance rather than a particular accelerator, but its cost-savings claims and ability to deliver consistent service across mixed infrastructure remain unproven. OpenRelay has launched an inference API that sends AI workloads across GPU, TPU and Trainium capacity through a single endpoint. OpenRelay is in early access, though its /v1 API is listed as stable.

3 min readRead story
Enterprise adoptionInfrastructure

Salesforce Uses SageMaker Placement Controls to Meet 2-AZ Agentforce Standard, Keep 8x Savings

The configuration separates model-level resilience from an endpoint’s infrastructure footprint, but balanced placement still depends on GPU capacity in each target Availability Zone. Salesforce has used new placement controls in SageMaker Inference Components to make its production Agentforce models support two Availability Zones, a requirement that an endpoint spanning multiple zones did not previously guarantee.

3 min readRead story
Open releaseInfrastructure

Snowflake Open-Sources Semi-Persistence With Sub-Second Single-GPU vLLM Swaps

The release treats GPU model copies as disposable while retaining weights in CPU memory, aiming to let many specialized models share costly accelerator capacity without long reload delays. Idle models can give back GPU memory only if they can return before the next request waits too long. Avoiding the slow path on every swap The target is a tradeoff in vLLM, the inference engine Semi-Persistence currently wraps.

3 min readRead story
BenchmarkInfrastructure

Heidi Cuts ASR GPUs From 16 to 4 With CUDA MPS, Holding Sub-Second Latency

The production design turns spare capacity in small speech-inference requests into concurrency, but it depends on workload-specific scheduling and protections against MPS failure modes. AWS, Nvidia and Heidi Health have published a speech-recognition serving design that cut Heidi’s required GPU instances from 16 to four while maintaining sub-second transcription latency. Source: aws.amazon.com.

3 min readRead story
LaunchInfrastructure

Deepgram Sends SageMaker Billing and GPU Metrics to CloudWatch Without Container Egress

The release makes vendor-level cost and capacity signals visible inside AWS tools, but teams still need different metric streams for billing, feature usage and an individual endpoint’s GPU headroom. Keeping speech AI inside a customer’s AWS account can support data-residency and compliance work. The design needs no agent, sidecar or additional IAM permissions.

3 min readRead story
PartnershipInfrastructure

AWS Plans 2 Million Nvidia GPUs for 2027–28 and Federal AI Factories

The partnership reaches beyond GPU supply into CPUs, networking and AWS’s own Trainium chips. The largest capacity commitment remains a future deployment, making execution on the shared infrastructure the central test. AWS plans to deploy 2 million additional Nvidia GPUs across its global infrastructure in 2027 and 2028 and plans U.S. government AI factories with 100,000 GPUs on secure AWS infrastructure.

3 min readRead story
PartnershipInfrastructure

Cisco Adds Supermicro Rack-Scale Systems to Its NVIDIA AI Factory for October Rollout

The partnership turns Cisco’s AI infrastructure offer into a more complete rack-to-fabric package, combining dense GPU servers, networking and liquid cooling for customers building larger AI clusters. The move brings the compute layer more directly into Cisco’s AI infrastructure offer as customers confront the power and cooling demands of larger clusters. Cisco announced the partnership with Supermicro on August 25.

2 min readRead story
LaunchInfrastructure

AWS Brings Ray Into SageMaker HyperPod With Recovery Tools and Tiered Cache on EKS

The launch moves several operational tasks behind SageMaker Studio while tying Ray teams to a HyperPod-on-EKS setup and its required add-ons. AWS has integrated Ray into SageMaker HyperPod on Amazon EKS, putting cluster management, job submission, dashboards, and hung-job-detection configuration into SageMaker Studio. Original project Python package Open package ↗ KubeRay Open repository ↗ in this interactive demo.

3 min readRead story
LaunchInfrastructure

Nvidia Puts Groq 3 LPX Into Production for Faster AI-Agent Responses

The first Nebius deployment will test Nvidia’s argument that specialized token generation can make long, tool-using AI sessions more responsive while leaving GPUs at the center of the system. Nvidia has put its Groq 3 LPX inference chip into full production, turning technology acquired from Groq into a rack-scale product for AI systems where delayed output can make an agent feel slow.

3 min readRead story
5 more next