Long-form writing: reference architectures, traffic-flow deep dives, and cost models.
How to stand up a Pilot Light DR cluster for NC2 on AWS: protection tiers, sizing for Prism Central and MST, the network and ports, keeping MST off the NAT Gateway with an S3 Gateway Endpoint, bucket prep, and proving failover.
Real-time AI endpoints are sized for the peak, so their GPUs sit idle most of the week. NAI 2.8 batch inference and a hibernate-and-resume schedule put those hours to work.
NC2 on AWS now deploys clusters onto the Dedicated Hosts you have already paid for, using idle hosts already in the account before it provisions new capacity.
Nutanix Enterprise AI 2.8 shipped fine-grained authorization and per-model sharing. That's the feature that answers whether fifty engineers can share one instance without seeing each other's models, keys, and endpoints.
MST Instant Restore splits recovery into two clocks. The guest can boot while most of the disk is still in the bucket. Background hydration finishes the copy. Missing blocks that the guest needs early are fetched on demand.
Most token skills change how Claude behaves. Token Optimizer measures what your setup is already wasting, then fixes it: bloated configs, unused skills, stale memory, compaction loss and model misrouting.
Prism Central 7.6 made Projects a hard boundary. A non-default project VM fails the protection task. The same VM saves clean into a recovery plan and breaks at failover.
Keep NC2 on Azure DR traffic private for Zero Compute and Pilot Light: lock down the MST storage account, front Blob with a Private Endpoint, fix DNS, and cut data-transfer cost ~78%. Both directions of the private path are now live-tested.
A step-by-step build for Zero Compute MST DR on NC2 for Azure: the VNet and storage account, the Private Endpoint and DNS zone, the DNS resolver and conditional forwarder that keep replication off the public internet, and the manual failover and failback procedures.
Reasoning-trace distillation plus a strong open-weight base model turned "build a competitive AI model" from a multi-billion-dollar lab problem into a bounded engineering project. That is why private and sovereign AI are suddenly real options for people who could not compete before.
Files 5.4 is organized around four areas: public cloud extensions, data migration, admin operations and file system security. The common thread is operational control, aimed at the problems that show up after deployment, not the ones on a spec sheet.
Nutanix ships an actual migration tool for moving legacy protection domains onto entity-centric protection policies and recovery plans. Here's what it maps, what it can't carry over, and both ways to run it.
I run nine AI personas on my machine, with specialists handling video, design, research, content, and project management. This is the story of what broke, what I changed, and what it costs to add a cheaper independent reviewer.
A local documentation wiki that has to stay current, and DR automation against Prism Central and Nutanix Cloud Manager. Both come down to reading the web, so I pointed a tool whose search and fetch cost nothing at real work.
Pre-stage a migration target and the same prefix gets advertised from two places at once, which is how you end up chasing asymmetric routing through a stateful firewall on a Saturday. Disconnected overlay subnets let a subnet exist before it is routed, so exactly one path is advertised at any moment.
A step-by-step technical walkthrough for deploying Nutanix Government Cloud Clusters in accounts where the operator can't touch IAM: precreate every role first, deploy last.
Nutanix Government Cloud Clusters used to require the person deploying to hold full IAM rights. GC2 1.2 splits that into two roles, and it's a bigger deal for teams than it sounds.
Comparing pocket-tts against voicebox's chatterbox backend for cloned-voice narration, on the same laptop CPU, same reference clip, same test sentences.
Default Claude Code leaves the token meter running hard. Four skills cut the bill from four different angles: the words it writes, the code it builds, the motion it wastes, and the context it drags around.
The GPU gets all the hype, but the model runs on a GPU while the agent lives on a VM. Long-running agentic work is a stateful, always-on workload, and the hard parts - uptime, state, rollback, secure multi-tenancy, blast-radius containment - are classic virtualization problems Nutanix already solved.
Every packet in and out of your NC2 user VMs crosses the Flow Gateway, so for real workloads you run a scaled-out set. There are two ways to spread traffic across them. One bills you by the hour for a Route Server and a pair of BGP VMs. The other is an internal load balancer you already understand - and NC2 builds it for you.
The US government pulled Anthropic's Fable 5 just 72 hours after launch. It is an important governance precedent - and a non-event for the models you actually build on. What happened, what is speculation, and what comes next.
GPU inference, enterprise databases, and cross-site protection on AWS — five Nutanix and AWS products composed into one coherent platform.
How Nutanix Flow Virtual Networking moves traffic on AWS: overlays, NAT vs no-NAT, ERPs, VPC gateways, and Transit Gateway routing for NC2.