Techzine TV podcast

What if the best way to cool a data center was to boil the liquid sitting on top of your servers? Reza Azizian, co-founder of Ferveret, joins us to explain how his company is bringing nuclear reactor heat transfer physics to AI data centers. This should make it possible to get up to 35% more compute from the same power envelope.

Ferveret has developed a chassis-level, rack-mountable immersion cooling system called Adaptive Phase Cooling. Unlike traditional immersion tanks that require infrastructure overhauls, Ferveret units drop into existing direct-to-chip rack infrastructure, require no PFAS fluids, operate near atmospheric pressure, and cool the entire rack. The result is a PUE as low as 1.03 and a return on investment measured in roughly one year, Azizian says.

Azizian walks us through the journey from MIT nuclear engineering research to designing cooling for Microsoft HoloLens and NVIDIA GPUs, the physics behind sub-cooled boiling, why air cooling is hitting its limits under Moore's Law, and how the company sees the future of AI data centers.

Key takeaways:
• Ferveret uses sub-cooled boiling (Adaptive Phase Cooling), not full two-phase immersion, eliminating pressure buildup and fluid loss
• Chassis-level form factor fits existing rack infrastructure
• PFAS-free fluid used for electronics cleaning, available off the shelf
• Benchmarks on H200 and Blackwell show ~15% teraflop-per-watt improvement vs. direct-to-chip
• System operates at 45°C liquid inlet temperature, enabling free cooling and eliminating chillers
• Full-rack liquid cooling (compute, networking, power) achieves PUE of 1.03–1.04
• From the same power envelope, up to 35% more compute output is achievable
• ROI approximately one year; cost comparable to direct-to-chip at scale
• Go-to-market through ODMs and OEMs for full-rack shipments
• Scalable beyond 600 kW per rack with no current cooling capacity ceiling observed

Chapters:
0:01 - Introduction: Ferveret and the science of boiling
1:28 - From nuclear reactors at MIT to data center cooling
2:55 - Designing cooling for HoloLens and NVIDIA GPUs
3:56 - Why air cooling is hitting its limits
7:21 - The problem with traditional immersion cooling
8:42 - Adaptive Phase Cooling: the sub-cooled boiling innovation
9:54 - PFAS-free fluid and the proprietary heat exchanger
13:21 - Power efficiency: 15% more teraflops and 35% more compute
23:41 - Go-to-market: partnering with ODMs and OEMs
27:38 - Is AI a bubble? Market outlook and future data center trends
30:35 - Silicon photonics, scalability, and cooling beyond 600 kW per rack

At NetApp INSIGHT in Las Vegas, Techzine TV sat down with Jürgen Hofkens, CTO EMEA at NetApp. We talked about why enterprises are struggling to get the most out of their data for AI. Storage is a major bottleneck, until now. Hofkens talks about the core tension between AI's voracious appetite for data and the practical constraints organizations face when trying to feed that data to GPU clusters at scale.

The conversation centers on NetApp's new Novus architecture, which tackles a fundamental bottleneck: the metadata service. By disaggregating the metadata layer from compute and storage, a concept borrowed from high-performance computing, Novus enables near-infinite namespace scaling using parallel NFS (pNFS) while staying fully ONTAP-compatible. This means enterprises can expose their existing data to neocloud GPU clusters without copying it. They also don't have to sacrifice governance or give up any encryption keys.

Key takeaways:
• Enterprises currently use less than 5% of their data for AI due to structural and throughput constraints
• A cluster of 50,000 GPUs needs 100 TB/s of storage throughput. Today's storage typically tops out at 5 TB/s
• Novus disaggregates the metadata service, removing the architectural bottleneck and enabling exascale namespaces
• Novus runs on ONTAP and uses pNFS (so no proprietary protocols). This makes it compatible with existing enterprise data estates
• Neocloud providers can use ONTAP compatibility as a differentiator to attract enterprise customers
• NetApp's strategy centers on delivering intelligent data infrastructure anywhere: on-premises, hyperscalers, and NeoClouds

Chapters:
2:00 The biggest data trends in the AI era
3:54 Which industries lead in AI data maturity
6:27 Technology challenges: data copies and governance complexity
7:31 The GPU throughput bottleneck: only 5% of data accessible
9:34 Introducing NOVUS: disaggregating metadata for exascale storage
13:13 ONTAP meets NeoCloud: enabling enterprise AI workloads
17:13 Sustained performance and flexible hardware with Supermicro
23:33 The rise of CPUs, AI agents, and enterprise file systems
25:41 Future-proofing with a flexible data architecture

At its own Accelerate event, Everpure did several important strategic announcements. The company is expanding its platform on top of what it already has and is going to compete with quite a number of companies outside the storage space. To talk more about this, we talked to Prakash Darji, General Manager, Digital Experience at Everpure and hear more about why the company is expanding well beyond its storage roots.

The introduction of Universal Data Intelligence marks a strategic shift toward helping enterprises manage data independently of applications. Everpure makes this move driven by the architectural demands of AI agents, large language models, and the need for data primacy across the entire enterprise.

Darji walks through the two core components of Universal Data Intelligence. First there's Data Intelligence (formerly 1touch, acquired earlier this year), which catalogs, classifies, and contextualizes data across every source, from block/file/object storage to SaaS platforms like Salesforce, ServiceNow, and Workday. The second component is Data Stream. This vectorizes relevant data and prepares it for AI consumption.

Together, Data Intelligence and Data Stream should form an end-to-end portfolio that also spans FlashBlade S and FlashBlade X for deployment at any scale. A key differentiator is Everpure's middleware approach. Rather than asking customers to move all their data into one ecosystem, the platform builds a knowledge graph of relationships and metadata, pointing back to original sources.

The conversation we have with Darji also gets into the competitive landscape. Because Everpure is going up against some formidable opponents. However, Darji argues that vendors like SAP, Salesforce, Databricks, and Snowflake are all pushing self-serving "give us your data" strategies that don't hold up in real-world multi-system environments. Everpure's answer is data primacy. This means treating data as the primary layer on which applications and AI agents operate, rather than the other way around. The interview closes with a forward-looking discussion on how new applications will be built on shared data schemas, with workflows replacing traditional application integrations.

• Universal Data Intelligence combines Data Intelligence and Data Stream for end-to-end AI readiness
• Data Intelligence catalogs and contextualizes data across all sources without replicating it
• Only ~15% of enterprise data is truly useful. Finding the relevant subset improves AI accuracy and reduces token costs
• Data Stream vectorizes data for AI consumption; integration with Data Intelligence ensures only relevant data is prepared
• A knowledge graph tracks relationships and lineage across systems without storing the underlying data
• Everpure's middleware approach contrasts with ecosystem-lock-in strategies from SAP, Salesforce, Databricks, and Snowflake
• Portworx remains app-centric and will need to evolve toward data primacy to fit the new strategy
• Future applications will be built as workflows on top of shared data schemas, not as isolated systems

0:35 Pure Storage's new direction beyond storage
1:15 Introducing Universal Data Intelligence
2:31 Finding the right data: relevance and primary sources
6:46 Context, ontologies and knowledge graphs
15:00 Data Stream: vectorization and data prep explained
16:30 Navigating a crowded competitive landscape
20:57 Data primacy: moving beyond data-centricity
22:51 The future: data schemas, workflows, and AI agents

At PegaWorld in Las Vegas, Techzine TV sat down with Don Schuerman, CTO and head of marketing at Pega, for a wide-ranging conversation about the state of enterprise AI. We discuss with him why simply dropping AI tools into existing workflows is not enough, and why enterprises must fundamentally reimagine their processes, teams, and application architectures to capture real value from AI.

Predictability is a key concept Schuerman returns to regulary. Predictability of the outcome of AI workflows, but definitely also of the cost. Accuracy is also something that needs to be determined properly when it comes to AI. As Schuerman points out during our conversation, an 85% AI accuracy rate is actually a 15% failure rate for a bank. If you compound agent errors, the failure rate will become exponentially worse. We also dig into the cost of tokens. Schuerman gets into how Pega's deterministic orchestration approach should keep AI spending controllable. It makes it possible to offer outcome-based pricing tied to workflows executed rather than tokens consumed.

Schuerman also walks us through the major evolution of Pega Blueprint. That used to be a rather basic ideation tool when it launched in April 2024, but is has been developed into a full design-to-build accelerator. On the new Customer Engagement Studio, he tells us that it brings next best action capabilities to marketing teams who don't have a data science background. The end of the conversation focuses on how to build a clear framework for measuring AI success: better processes, faster delivery, and uncompromised trust.

Key takeaways:
• Enterprises must reimagine workflows, processes, and team interactions, not just add AI tools
• AI accuracy means different things to different people
• Pega's strategy separates non-deterministic AI (design/build) from deterministic execution (run)
• Pega Blueprint now accelerates both the design phase and the build phase of application development
• Applications need to evolve from screen-and-database collections to process, agent, and knowledge collections exposed via MCP
• Outcome-based pricing charges per workflow executed, not per token burned
• Customer Engagement Studio wants to democratize next best action for non-data-science marketing teams
• Infinity 26 supports bring-your-own models, including Mistral for European clients

Chapters:
1:13 - Reimagining enterprise workflows and teams for AI
1:35 - The predictability problem: why 85% accuracy isn't enough
2:25 - Predictability of cost and the tokenomics challenge
3:32 - AI across design, build and run: a smarter architecture
4:41 - Pega Blueprint: from ideation tool to full build accelerator
8:53 - Rethinking application architecture: apps as MCP targets
11:41 - Outcome-based pricing: paying per workflow, not per token
15:49 - Customer Engagement Studio: next best action for all
18:31 - Infinity 26: platform flexibility and bring-your-own AI models
20:05 - Measuring success: better, faster, and trustworthy

David Warshavski, co-founder and Chief Product Officer of Tonic Security, joins Techzine TV to talk about how his company is rethinking exposure management from the ground up. He can draw on decades of incident response experience defending Fortune 500 companies against nation-state threat actors from Russia, China, North Korea, and Iran, and has seen that cybersecurity isn't about finding vulnerabilities, but about prioritizing and fixing the right ones before a breach occurs. This sounds like common sense, but is actually quite rare in the cybersecurity industry.

Tonic Security wants to address the problem of alert overload by building a context engine that taps into both IT/security tools and collaboration platforms like Slack, Notion, Confluence, Teams, and ServiceNow. By extracting business operational context from unstructured data, Tonic can determine whether a vulnerability is exploitable. Not only that, it can also determin whether exploiting it would actually impact critical business processes. That last distinction should fundamentally change how security teams spend their time.

The platform acts as both an insight layer and an execution engine. It features the Tonic Mobilization Coordinator. This is an agentic workflow that can fully automate remediation in low-risk scenarios, Warshavski tells us. We also discuss how Tonic can replace legacy vulnerability scanners like Tenable, Rapid7, and Qualys, how it feeds context to DIY agentic workflows and SIEM solutions, and why opening up to an ecosystem is the company's next frontier.

Key takeaways:
• Most security breaches involve vulnerabilities that were already known but not properly prioritized or remediated in time
• CVE and CVSS scores alone do not reflect real business impact and create massive, misleading backlogs
• Tonic Security ingests unstructured data from collaboration tools to determine true business criticality of assets
• The platform differentiates between exploitability, reachability, and actual business blast radius
• The Tonic Mobilization Coordinator automates remediation in high-confidence, low-risk scenarios
• Security teams report significant time savings, reduced burnout, and freedom to focus on strategic priorities
• Tonic can displace legacy scanners and act as a context engine for other security tools via API and MCP server
• Fully on-premises deployment is available for heavily regulated industries

Chapters:
0:54 - David Warshavski's background in incident response
2:15 - Why Tonic Security was founded
5:32 - The prioritization problem and CVE noise
9:03 - Understanding business impact and context
17:17 - Tonic as a decision and execution engine
20:53 - Reducing alert fatigue and saving security teams time
27:44 - Where Tonic fits in your security stack
33:12 - Flexible deployment and next frontiers

At Cisco Live in Las Vegas, Techzine TV sat down with Gordon Thomson, president of EMEA at Cisco, for a wide-ranging conversation about the explosive pace of market change, and some of the most significant product and strategic moves Cisco has made in decades.

A major focus of the conversation is Cisco's Sovereign Critical Infrastructure (SCI) initiative. Launched in September 2025 with limited scope, SCI has since evolved into a formal legal framework giving customers guarantees around air-gapped operation, no backdoors, and continued product functionality even during embargoes or conflict. Thomson talks about how SCI has expanded to 13 product families, including data center, campus, wireless, collaboration, and Splunk on-prem, and how Cisco plans to allow customers to recertify existing installed-base devices as sovereign toward the end of the year, switching them from subscription-based to perpetual licenses.

The interview closes with a deep dive into Cisco Cloud Control, which Thomson, a 29-year Cisco veteran, describes as one of the most fundamental things the company has ever done. By unifying networking, security, and observability into a single AI-assisted platform with agentic automation, Cisco Cloud Control represents the culmination of years of incremental innovation. Currently rolling out in the US before expanding to Europe, it is a cloud-only service, a deliberate trade-off that Thomson acknowledges some customers will need to weigh carefully.

Key takeaways:
• Why Cisco shifted from cloud-first thinking to Sovereign Critical Infrastructure
• How SCI moved from a market signal to a binding legal framework with perpetual licenses
• The plan to recertify hundreds of millions of existing Cisco devices as sovereign
• The sovereignty debate: emotion, trust, chip-level demands, and the price customers are willing to pay
• AI's role in reducing interoperability costs for sovereign cloud deployments
• Cisco Cloud Control: unifying networking, security, and observability in one platform
• Why Cloud Control is cloud-only, and what that means for sovereign customers

Chapters:
1:08 - Navigating a world of generational change
3:06 - Cisco's cultural and organizational transformation
5:01 - Sovereign Critical Infrastructure: from intent to legal framework
8:50 - The sovereignty debate: emotion, trust, and chip-level demands
11:28 - 13 product families and recertifying existing sovereign devices
16:47 - AI, sovereign clouds, and on-prem dashboard trade-offs
20:14 - Cisco Cloud Control: networking and security unified

Rob Bunger, Global Director of Data Center Solution Architecture at Schneider Electric, sits down with us to talk about why the industry is moving from traditional 48V AC to 800V DC power distribution for next-generation AI workloads. As AI racks scale from 150kW to potentially 1MW per rack, the current power architecture simply cannot deliver the required density without consuming all available rack space with cabling and power supplies.

The solution involves leveraging 800V DC technology (partly coming from the EV charging industry), deploying power through sidecar systems, and eventually centralized DC distribution. This shift addresses critical challenges including space constraints, safety requirements, and the need for live-swap capabilities in high-density environments. Bunger discusses real-world deployment timelines, with pilot projects expected in 2024, and explains how this represents a fundamental shift in data center power architecture.

Key takeaways:
• Why 48V AC power distribution cannot scale beyond current AI rack densities
• How 800V DC reduces cabling complexity and saves rack space
• The role of sidecar systems in transitioning to DC power
• Safety testing and breaker technology for high-voltage DC systems
• Timeline for commercial deployment of 800V DC infrastructure

Chapters:
0:15 - Introduction to 800V DC power systems
1:38 - Current 48V power architecture explained
3:21 - Power density challenges with AI racks
8:12 - Why 800V DC is the solution
12:12 - Sidecar deployment architecture
15:01 - Deployment timeline and future outlook
17:24 - The return to DC power distribution
19:01 - Safety considerations and testing

Charlie Sutton, Chief Design Officer at Atlassian, discusses the evolving role of design in the AI era. From managing non-deterministic experiences to controlling token costs, he shares insights on how Atlassian approaches AI product design with intention and structure.

Sutton explains why good AI design isn't about maximizing outputs but achieving meaningful outcomes. He reveals how Atlassian uses the teamwork graph, structured data, and design tokens to keep AI costs in check while maintaining quality. The conversation explores the tension between deterministic control and generative possibilities, and how context-rich work environments offer advantages over consumer applications.

Key takeaways:
• Why companies need chief design officers in the AI age
• How AI is changing both the tools and materials of design
• The importance of user agency over direct control in non-deterministic systems
• Strategies for managing AI token costs through structured data
• Design principles Atlassian applies to AI products
• Moving from AI experimentation to value-driven adoption
• Balancing the familiar with the novel in product experiences
• Why AI will become assumed, just like the internet

Chapters:
0:14 - What is a chief design officer?
2:34 - How AI is transforming design work
3:08 - Balancing control in AI-powered experiences
6:16 - Using the Teamwork Graph in design
8:27 - Managing AI token costs effectively
11:39 - Moving beyond the experimentation phase
13:48 - Principles for AI product design
19:18 - Balancing the familiar with the novel

Some time ago we sat down with Martin Elwin, Technology Director for Northern Europe at AWS, and talked about the massive infrastructure investments needed for AI adoption, of course also from AWS. However, simply buying GPUs doesn't deliver AI value. More is needed. We also talk about new on-prem offerings AWS introduced for organizations wanting to control their hardware while benefiting from managed services.

Other key topics include production AI deployments using Amazon Bedrock AgentCore for evaluations and drift detection, the importance of continuous testing in AI applications, and how AWS applies its own Amazon learnings to customer challenges in logistics, supply chain optimization, and regulated industries.

0:50 - AI driving infrastructure capacity needs
2:37 - Introducing AWS AI factories
3:49 - Getting business value from AI
10:13 - European sovereign cloud discussion
18:11 - Amazon Bedrock AgentCore for production AI
20:57 - Security agents and continuous testing

At NTT Upgrade 2026, Paul Bloudoff, senior director of Edge AI at NTT Data, discusses with us how edge computing, private 5G, and physical AI are transforming manufacturing operations. We talk about the symbiotic relationship between edge AI and private 5G connectivity, and focus on how manufacturers can achieve real-time insights and automation on the factory floor.

During our conversation, Bloudoff talks about real-world examples, including Cargill's deployment of private 5G across 50+ manufacturing facilities, and explains how NTT Data delivers full-stack solutions from sensors to cloud infrastructure. The conversation covers predictive maintenance, task verification through machine vision, and how foundation models can be trained with just 40 hours of video data instead of requiring months of preparation.

Key takeaways:
• Edge AI focuses on computing as close as possible to data sources in OT environments
• Private 5G enables low-latency, high-bandwidth connectivity for real-time manufacturing data
• Physical AI models can be deployed in weeks rather than years with modern foundation models
• Multiple use cases are typically needed to justify private 5G network deployments
• Safety and preventative maintenance are often the first use cases that enable adoption
• Full-stack solutions include sensors, connectivity, edge compute, algorithms, and managed services
• Partnerships with Nvidia, Qualcomm, and ISVs enable comprehensive edge AI solutions

Chapters:
1:35 - Defining the edge in enterprise environments
2:59 - Full stack edge AI solutions
4:54 - Private 5G and edge AI synergy
6:23 - Use cases for edge AI and physical AI
7:41 - Real-world private 5G deployments
13:36 - AI partnerships and ecosystem approach
13:59 - Measuring value and ROI
22:23 - Automation and the future of edge AI