Partner Marketing Manager

70th IT Press Tour: DDN wants every GPU earning its keep, and NVIDIA and Google already run on its data engines. Alex Zaharov-Reutt, Global AI and Technology Editor | Published 14 Oct 2026 A GPU waiting for data still draws full power. DDN has built its entire pitch around that one uncomfortable fact, and with bookings growing threefold, a rebuilt executive team and a fresh inference engine, it reckons it's found the money in fixing it. I was at the 70th IT Press Tour in California last week, where DDN gave journalists an impressively detailed briefing session, under the banner "Make AI profitable". Presenting were DDN president and chief revenue officer, Kevin Delane, and DDN chief product officer, Omer Asad. DDN's new chief marketing officer, Adam Berlew, sat in and chimed in as appropriate, despite being on what he told us was day eight in the job. IT Press Tour founder, Philippe Nicolas, said this was the ninth or tenth time the tour had visited DDN. The faces on the other side of the table were mostly new. What's also new is this full video presentation, upon which this article is based, so please watch, and read on! The market maths Delane opened with the market. DDN's slide, citing Gartner's August 2026 forecast, put spending on AI-optimised infrastructure as a service at US$42.3 billion this year, nearly double 2025's US$21.5 billion. This is also the first year inference spending (US$23.3 billion) overtakes training (US$19 billion), with inference taking 55% of the pie in 2026 and 59% in 2027. Gartner senior principal research analyst, Hardeep Singh, said in that forecast that production-scale deployment means models need "continuous, real-time execution rather than periodic training". That's the shift DDN is betting the company on. "The demand is outrunning the supply, so everybody's under pressure," Delane said. "Our goal is to make sure that every GPU out there is running at 99 to 100% utilisation." His summary was short. "Our goal is to make AI profitable." Three times the bookings Delane, who has been at DDN for six months, said bookings are growing at three times year on year. DDN had already hit two times by the time of the briefing, and he said he was "very comfortable" it would finish above three times. DDN claims more than 5,000 customers and more than a million GPUs served, spread across what it calls enterprise AI, strategic AI (AI natives and NVIDIA cloud partners) and sovereign AI. Delane came to DDN in April after stints at Salesforce (via its Own Company acquisition, where he was COO), Cohesity, where he was CRO, and Pure Storage (now Everpure), where he helped scale the company past US$2 billion in revenue, according to DDN's announcement. He's one piece of a heavily restocked leadership bench. The slide showed Asad, chief financial and operating officer Guido Torrini (ex-OneTrust), chief growth officer Rick Scurfield, chief people officer Lauren Bloch, chief legal officer Michelle Rosen, chief customer officer Andrew Kahl and Berlew, who joined from Equinix after earlier marketing roles at Google, Atlassian, Brocade and Dell. Berlew already has a plan for one of DDN's quirks: it's oddly shy about who uses it. "So I'm here now. We're gonna start asking, and we're gonna start publicising," he said. "Next time we meet, you'll see more logos." Google sells DDN as its own The relationship Asad most wanted to talk about was Google's. Google Cloud Managed Lustre runs on DDN's EXAScaler. Asad stressed that it's a native, first-party GCP service: you click for managed training in the console, DDN fires up underneath, Google collects the money and pays DDN on the back end. "When Google looked for how to keep all its TPUs busy and provide the best possible experience for AI out there, EXAScaler, or DDN, was the one that was called up," Asad said. Delane described the Google channel as "light touch for us", adding that DDN is "adding customers on a daily basis there". The August 2026 client slide included Meta, NVIDIA, Alibaba, Tencent, Jane Street, Siemens and Salesforce, plus three names Australian readers will know well: Atlassian, Fortescue and Melbourne health AI company Heidi. Salesforce, training on Managed Lustre, reported 1.5 times faster model training and 42% lower training costs, according to DDN's slides. At Google Cloud Next 26 in April, the service gained a shared KV cache for inference and a performance ceiling of 10TB/s. Asad also noted that many customers don't realise DDN is doing the heavy lifting behind the Google console, which is exactly why Berlew was hired. NVIDIA's own data sits on DDN If Google is the newer love affair, NVIDIA is the long marriage. Asad put it at eight years. Delane corrected him and put it at 10. The slide said 2016. DDN says NVIDIA runs more than 9,000 GPUs internally on DDN, including the Eos (12PB) and Selene (15PB) SuperPODs. Its slide quotes NVIDIA founder and CEO, Jensen Huang: "Without DDN, NVIDIA supercomputers wouldn't be possible." Asad said NVIDIA's Nemotron open-weights models train on DDN, and that when NVIDIA leases capacity back from neoclouds such as CoreWeave and Lambda for its own development, the reference architecture there is DDN too. Eos is a neat efficiency story. DDN says the 576 DGX system supercomputer uses fewer than three racks of DDN storage and 100kW of storage power for 18.4 exaflops. Asad, who said he came to DDN from AWS (and before that ran storage at HPE, having arrived there via Nimble Storage), was frank about the depth of the tie. He spent the session close to his phone because two new NVIDIA installs were going live. "If even something blips, I will just disappear from here without even apologising," he warned - although nothing blipped, and he gave a great presentation, as you can watch above! Power is the ceiling, data is the bottleneck Asad's core argument is that AI builders plan around power. A frontier lab or neocloud buys a 50MW, 500MW or 1GW envelope, then rolls in GPU racks. "Before you hit power max, you're already bottlenecked at data," he said. "If it's like having a 747 or a 777 and just parking it, that's not a good investment." Each stage of the AI pipeline wants something different from storage. Training wants massive parallel reads. Checkpointing flips that into bursty parallel writes, and every GPU waits for the slowest one to be acknowledged. Inference brings random reads of wildly varying size, plus model loading on demand. Then neoclouds stack multiple tenants, all at different stages, on shared infrastructure. Asad framed it as a margin question: "What is the margin that CoreWeave has? What is the margin that a neocloud provider has?" His recurring line was that small clusters flatter everyone. "At eight nodes, everybody shines," he said. AI data sets run to hundreds of petabytes, and an average lab bucket is 50PB. EXAScaler, the training workhorse DDN's answer comes as two engines under a single namespace, managed by a "Horizon" control plane. EXAScaler handles training. It's built on open-source Lustre, which DDN gained stewardship of through its 2018 Whamcloud acquisition. Somewhere between 70% and 80% of open-source Lustre developers work at DDN, by Asad's two separate estimates during the session. Asad was keen to separate the two. Lustre, he said, lacks multi-tenancy and quality of service, and needs "75 engineers to install and operate", plus around US$100,000 in professional services. EXAScaler, by contrast, is "push button". The numbers on the slides: 96% network utilisation in NVIDIA's DGX SuperPOD reference design, up to 99% GPU utilisation, and 190GB/s from a two-rack-unit appliance drawing 2.5kW. India's Yotta runs 8,000 NVIDIA B200 GPUs on EXAScaler in its sovereign Shakti Cloud, at 99% utilisation across tenants, DDN says. (Asad described the 190GB/s figure as write speed. Blocks & Files reported the new AI400X3M as up to 190GB/s read and 110GB/s sequential write, so I'd treat writes as the lower number until DDN clarifies.) Asad named Mistral, NVIDIA Nemotron and xAI's Grok (xAI, he noted, now sits under SpaceX) as models trained on DDN. A typical Mistral environment, he said, has 100,000 GPUs hitting DDN through a 4U controller. He also offered a challenge. Against a rival on a 50,000 GPU cluster, he said, "80% of the time, it will finish the job in half the time. Any time, anywhere, test that." Later he called EXAScaler "eight times faster than the nearest competitor". Those are vendor claims, and I'd love to see someone take Asad up on his dare. Infinia, the inference engine The newer engine is Infinia, four years in the making. It's a distributed key-value store underneath, with S3, POSIX file, KV cache and a public SDK on top, and no dedicated metadata appliances. You add nodes, and data and metadata scale together. Asad stripped it back to one idea: a distributed hash tells every client exactly which nodes hold each piece of a file, so clients fetch everything in parallel with no gateway or index tier in the way. Nicolas grilled him at length on where that hash is computed. The answer: on Infinia nodes in the basic case, with the hash map held in memory across the cluster. For NVIDIA GPUs, the calculation moves into an Infinia plugin for NIXL, which ships inside the NVIDIA Dynamo container and covers vLLM, SGLang and LMCache. That means KV cache (the context a model has already computed) moves straight between GPU memory and Infinia over RDMA. "No S3, no NFS, no POSIX," Asad said. "KV cache is the first-class citizen." DDN's slide claims 4.2 times faster time to first token at 100,000 token context with Dynamo. It also tallies the cost of recomputing context on a 4,096 GPU B200 fleet at US$6.52 per GPU hour: roughly US$186,000 in lost revenue every day. In July, DDN and European sovereign cloud Nebul announced early results showing token responses nearly 10 times faster than without a KV cache fabric. Newer Nebul figures were shown under embargo, so they'll have to wait. On multi-tenancy, one API call sets a tenant's keys, quota and access, and weighted fair share stops noisy neighbours. Asad's road analogy was the best of the day: "If there is no car on the road, go nuts. But if there are other cars, then stay in your lane." 700PB in a single namespace The headline Infinia deployment is a frontier AI lab that DDN runs as a managed service. It spans four data halls, each with six racks and 240 nodes, for 700PB in total, more than 90% full, with over 10 billion objects in one hall. It serves about 200,000 GPUs from multiple silicon vendors. Peak throughput hit 6.4TB/s of GETs, and with the new Infinia 2.5 release DDN says it sees 4TB/s GETs, 2TB/s PUTs and five million small-object GETs per second. Oracle Cloud Infrastructure audited Infinia at 225,000 GETs and 194,000 LIST operations per second. Asad singled out the LIST number, since data preparation means endless crawling of huge buckets. Infinia is still young: about 15 production deployments, ranging from seven nodes to more than 300. A 500TB inference store needs three nodes. An exabyte namespace needs 300 or more. Asked whether Infinia will eventually replace EXAScaler, Asad said there were no plans. "It's a workhorse," he said. If a customer leads with object, which he put at 99% of new use cases, it's Infinia, and EXAScaler can be switched on in the same namespace when file access is needed. Next year brings native analytical interfaces, including Kafka and vector databases served straight from the system. How DDN stacks up The competition is fierce and well funded. VAST Data raised about US$1 billion at a US$30 billion valuation in April. WEKA launched NeuralMesh 6 and WEKApod in July with its own KV cache Augmented Memory Grid, claiming 10 times higher token throughput in OCI deployments. Hammerspace, which presented on the same tour, argues the fix lives in the standard Linux kernel's parallel NFS. Every vendor is now selling tokens per dollar. DDN's money story is interesting too. Blackstone invested US$300 million at a US$5 billion valuation in January 2025. In June, Bloomberg reported (via Blocks & Files) that DDN wanted strategic investors, and Forbes reported DDN expects revenue to roughly double to about US$1 billion this year. Asked for an update, Delane kept his cards close. "We feel we have great opportunities in the market, and we'll continue to execute and grow," he said. "You saw the rate we're growing at." Beyond NVIDIA, and down to the edge I asked whether DDN works with AMD and the hyperscalers building their own chips. Asad said DDN actively works with Google TPUs, NVIDIA and AMD. "Wherever optimisation libraries are available, we jump into it right away," he said. Delane was more pragmatic. "Our fastest way to growth is NVIDIA," he said. I also asked what DDN's role is when powerful AI runs on phones and laptops. Asad said Infinia can scale down to a single node with a single SSD, and that robotics and autonomous cars are "a very big use case". Phones, he admitted, he hadn't contemplated yet. DDN eats its own cooking Asked by Nicolas and me about AI in DDN's own engineering, Asad said the company is a heavy user. It runs locally fine-tuned models (he named Kimi) inside its telemetry data lakes to spot I/O patterns, training stalls and GPU failure rates, and uses Claude licences too. "We're currently on a smaller than a three-month release cycle," he said, crediting automated code review, test case generation and unit tests. What it means for Australia Power was the thread that tied the session together. Berlew, freshly arrived from Equinix in the Netherlands, said that while more power will come, "raising that level of utilisation is so critical for the industry". That's acutely relevant here. Gartner forecasts Australian data centre power demand reaching almost 1.5GW in 2026 and 3.8GW by 2030. In September, NVIDIA grouped eight local operators, including NEXTDC, AirTrunk, CDC, Firmus and Sharon AI, into a 2GW sovereign AI program, while AEMO data shows grid connection queues that can take years. Delane noted that building data centres is easier outside the US. "The growth and opportunity right now in Asia Pac is moving faster than it is here and in Europe," he said, adding, "we do well in Asia Pac." DDN already has local roots, from Pawsey's radio astronomy storage to NSW Government on its customer slide, plus Atlassian, Fortescue and Heidi via Google Cloud. The verdict I think DDN's pitch is sharper than it's ever been. A GPU that sits idle wastes capital and power that Australia's grid can't easily spare, and DDN's claims about utilisation, checkpointing and KV cache go right to that cost. The caveats are the usual ones. Most numbers are DDN's own, Infinia is young, and VAST and WEKA are chasing the same tokens. But with NVIDIA and Google both running on its engines, DDN has proof points rivals would gladly trade for. As Asad put it, "Performance at scale. It's simplicity at scale. It's capacity at scale."

Varonis and Cohesity: improving cyber resilience with data security. Varonis and Cohesity close the backup dataset security gap in one integration. Last updated September 30, 2026 Backup datasets often contain sensitive information - from PII to intellectual property - and a wide range of business-critical data. Without visibility into their backups, security teams are left with a blind spot that increases the blast radius and impedes cyber recovery. Together, Varonis and Cohesity close that gap, pairing deep insight into backup data with stronger security and faster recovery. Varonis integrates with Cohesity to combine proactive data security with cyber resilience, surfacing what's inside backup data, automating remediation, detecting threats, and enabling secure recovery. How Varonis and Cohesity reduce risk and strengthen resilience Varonis and Cohesity address the gap between data security and cyber resilience, from discovery through recovery. Each layer builds on the last: Discover and classify sensitive data Varonis automatically discovers and classifies sensitive and regulated data across Cohesity Data Cloud, helping organizations identify risk, reduce data exposure, strengthen cyber resilience, and accelerate recovery efforts. Continuous visibility into sensitive data supports compliance, retention requirements, and AI readiness. Lock down access and shrink the blast radius Overexposed data is a breach waiting to happen. Varonis maps your entire permissions structure, providing visibility into where data resides, who has access to it, and who's using it. From there, you can enforce least privilege: managing access at a granular level and safely automating permission changes, while shrinking the blast radius with minimal effort. Detect and stop threats in real time Varonis builds baseline behavioral profiles for users and non-human identities, then continuously monitors for the activity that signals an attack, including suspicious data access, lateral movement, and privilege escalation. When something looks wrong, Varonis can alert your team and leverage automated responses to prevent data from being affected, like shutting down user sessions and changing passwords, with no wait for manual triage. And for organizations that want 24x7 coverage, Varonis Managed Data Detection and Response (MDDR) provides expert-led monitoring, investigation, and response to help stop threats before they impact critical data. Contain ransomware and recover fast If ransomware reaches your data, the two platforms work together to contain and recover. Varonis flags the abnormal file activity that signals encryption and triggers an automated response to lock it down. Cohesity hardens the data itself with AES-256 encryption, multifactor authentication, DataLock for WORM and FIPS-compliant encryption. And because Cohesity keeps unlimited, fully hydrated snapshots, you can mass-restore files, VMs and application objects almost instantly, cutting recovery time from days to minutes. Optimize cost Varonis Data Lifecyle Management continuously identifies redundant, obsolete, and trivial (ROT) data, helping organizations safely archive or dispose of data that no longer provides business value. By shrinking the amount of data stored and protected in Cohesity, customers can improve storage efficiency, reduce infrastructure costs, and maintain a cleaner, more manageable data estate while preserving compliance and security requirements. Complete visibility, protection, and recovery Security teams work faster when the picture is unified. By continuously identifying sensitive, stale, and overexposed data, Varonis helps organizations reduce cyber risk before an incident occurs and provides the context needed to prioritize recovery efforts when one does. Cohesity gives you a single UI and enterprise search across silos, VMs, backups, data centers, remote sites, and multiple clouds. Varonis layers on the context that makes that data defensible: classification, automated remediation, threat detection, and ROT reduction. Together, Varonis and Cohesity deliver a unified approach to data security and cyber resilience, helping organizations protect sensitive data throughout its lifecycle. Ready to secure your backup datasets? See how Varonis and Cohesity work together. Schedule a demo with Varonis Systems, Inc. to see Varonis in action. Varonis Systems, Inc.'ll personalize the session to your org's data security needs and answer any questions. See a sample of its Data Risk Assessment and learn the risks that could be lingering in your environment. Varonis' DRA is completely free and offers a clear path to automated remediation. Follow Varonis Systems, Inc. on LinkedIn, YouTube, and X (Twitter) for bite-sized insights on all things data security, including DSPM, threat detection, AI security, and more. Dan Petrillo Dan Petrillo is the VP of Product Marketing at Varonis.

A New Era in Protecting the World’s Data: How the Combination Will Benefit Customers and Partners