NetApp's Novus was orderable before PEAK:AIO closed. Is its code included?

The 2 November acquisition adds a metadata specialist alongside the existing Novus offering. The releases do not settle code inclusion, integration timing or commercial returns.

An acquisition does not establish when the acquired technology enters a product. Novus was already orderable. What evidence would show PEAK:AIO code has reached a generally available configuration?

On 2 November NetApp said it had completed the acquisition of PEAK:AIO, a Manchester software company founded in 2021. It had announced its intent to buy on 25 September, so the close took 38 days. NetApp gave no price, and the releases do not quantify integration costs, Novus orders or incremental revenue.

Novus was orderable before the deal closed

On 29 September NetApp announced Novus, a design that splits metadata from the data path. The first version pairs Novus Data Director metadata software, on qualified Supermicro servers, with ONTAP storage on AFF A90 systems. NetApp says that architecture "is orderable today." Orderability shows neither shipments, sales volume nor PEAK:AIO code inclusion.

NetApp says Novus is built to exceed 100 TB/s. Omdia's Tony Palmer, quoted in the release, said audited tests showed near linear scaling as ONTAP clusters were added. Omdia's own modelling projects 100 TB/s of sequential reads, a projection. Mike Leone of Moor Insights & Strategy, also quoted, called Novus a more scalable base for neocloud providers.

The closing release says PEAK:AIO's technology will come into ONTAP and is "expected to enable" new capabilities "such as NetApp Novus." Both releases carry a disclaimer that future plans are no commitment on availability or timing.

What PEAK:AIO brings

Metadata is the index that tells a storage system where each file lives. In a cluster with thousands of GPUs, slow lookups could leave chips waiting even if the data path has capacity.

PEAK:AIO's identified technology includes Lattice, an open source metadata server for parallel NFS. PEAK:AIO built it with Los Alamos National Laboratory and Carnegie Mellon University, and the companies released it in June 2026. StorageReview reports that PEAK:AIO labels Lattice an early release for research and testing, not yet recommended for production.

NetApp also lists Los Alamos, Carnegie Mellon, the NHS AIDE programme, Oxford Robotics Institute and the University of Liverpool as PEAK:AIO deployments. That shows installed experience. It does not say which sites run Lattice. Lattice is MIT licensed and hosted by the Linux Foundation.

Spinnaker took two and a half years to yield a first product

NetApp bought Spinnaker Networks in 2003. The Register reported on 12 June 2006 that Data ONTAP GX, "the first fruit" of that deal, was shipping. NetApp's product vice president blamed the wait on the amount of integration required.

NAND Research's Steve McDowell says full integration of Spinnaker took most of a decade. He judges PEAK:AIO more contained, because ONTAP stays the data layer and PEAK:AIO becomes a metadata tier. He calls the deal "well targeted" but says NetApp "must quickly convert" it.

NetApp also has three integrations in progress. It bought DataPelago in July and JetStream Software a month later. That raises an integration capacity question. McDowell names Lattice stewardship as another open one.

Rivals already ship at scale

McDowell writes that Dell, VAST Data, WEKA and Everpure ship AI scale storage today, and that time to market is NetApp's weakest point. He cites Dell's Lightning File System, generally available since April 2026, and Everpure's FlashBlade//EXA, which already pairs a separate metadata tier with parallel NFS. Metadata separation is therefore not exclusive to Novus.

Inference caches address a different bottleneck. DDN says Nebul, a European AI cloud, stored reusable context on DDN Infinia. The result was 2.5 times more input tokens per second and nearly 60 percent lower cost per million. Nebul ran the test on its own platforms, so it is a vendor result.

NetApp's documentation dated 19 August also describes offloading inference caches from GPUs to shared ONTAP or StorageGRID storage, so NetApp is not absent from caching. But projected sequential reads, inference tokens per second and concurrent users are different measurements. None ranks the products against one another on a common workload.

WEKA gained a new payment route on the same day

Oracle published a post on 2 November saying customers can now apply eligible Oracle Universal Credits to WEKA offerings on Oracle Cloud Infrastructure. A more integrated purchasing and billing experience is still planned. The change concerns eligible payment, not performance.

Oracle's private offer rules cap a customer's partner spend, existing plus new, at 25 percent of its active Universal Credits commitment. A customer already at the ceiling gains no further room to buy WEKA with credits.

The post repeats a joint Oracle and WEKA test on 72 H100 GPUs. WEKA's Augmented Memory Grid supported more than 5,000 concurrent users against about 600 for a DRAM only setup. These are vendor results, and Oracle tells customers to test them on their own workloads.

Measure the bottleneck before choosing a fix

NetApp offers a metadata architecture whose inclusion of acquired code remains unspecified. Oracle offers an eligible payment route for WEKA, while the cited WEKA result concerns inference caching. A customer cannot infer that either technology will solve an idle GPU problem without first measuring the bottleneck.

The missing price, sales volume and revenue contribution prevent a return calculation. A product release could settle code inclusion, though not financial value.

What could happen next

The following scenarios are ranked by likelihood, without numerical probabilities.

Most likely, Novus remains orderable but no generally available product documentation names PEAK:AIO or Lattice code by 31 March 2027. The closing release's future tense and Lattice's research label favour caution, while McDowell's more contained architecture weighs against assuming a long delay.

A limited release or roadmap could also plausibly name PEAK:AIO code before 31 March 2027 short of general availability. A dated release note stating availability terms would separate it from the first scenario.

Least likely, a generally available NetApp product page or release note identifies PEAK:AIO or Lattice code by 31 March 2027. McDowell's assessment makes it possible, while the research label argues against assuming it.

What would prove CTOL's prediction wrong?

With moderate confidence, CTOL expects that no generally available NetApp product will be documented as containing PEAK:AIO or Lattice code by 31 March 2027. It concerns public product documentation, not whether the code sits in any build.

A generally available release note or product page naming acquired code on or before that date would prove the prediction wrong. A roadmap statement, Lattice commit or Novus order would not. Even a qualifying release would show integration, not economic payoff.

INSIGHT ran from 29 September to 1 October and included Novus. NetApp's next earnings are targeted for 1 December, then 25 February 2027. Novus order commentary can inform demand and public Lattice commits can inform project activity, but neither proves code inclusion. The decisive evidence will be the product documentation available in March.