The Ravit Show

Everyone is building AI agents… but very few are talking about the biggest blocker: fragmented data. I sat down with Ravi Marwaha, Chief Product & Technology Officer at Arango, on The Ravit Show to unpack what’s really going on beneath the surface.

This was not a typical “AI hype” conversation. We went deep into why agents that are supposed to reason, decide, and act are still struggling in real enterprise environments. And it comes down to one thing: context.

Here’s what stood out from the conversation:

- Ravi’s journey and why he chose to build at the intersection of data and AI

- How fast the space has evolved and why most architectures are already outdated

- The real problem with fragmented data when it comes to building reliable agents

- The rise of the “context layer” and why everyone is suddenly talking about it

- Where Arango fits in vs vector databases, graph databases, and data catalogs

- Real customer use cases that go beyond theory

- And what the next 18 months could look like for this entire space

One thing became very clear to me:

The definition of a data platform is changing in real time. It is no longer just about storing or querying data. It is about enabling AI systems to actually understand and use that data in context.

If you are building anything in AI right now, this is a conversation worth paying attention to.

#data #ai #context #arango #theravitshow

Just wrapped a great conversation with Ravi Marwaha on-site at NVIDIA GTC.

We talked about what Arango is building and why it is starting to show up more in serious enterprise conversations.

If you have not come across them yet, worth a look: https://arango.ai/

What stood out to me was how they are thinking beyond just being “another database company” and leaning into something much bigger around context and how enterprises actually use data.

Next week, I am sitting down with Ravi for a deeper podcast. We are going to unpack some real questions that I think a lot of people in data and AI are already thinking about:

• Why move from enterprise roles to Arango

• What makes Arango interesting right now

• Are they a database company, a context layer, or something else entirely

• How they evolved so quickly

• Everyone claims to be an AI company, what actually makes them different

• Real enterprise use cases

• What the next 2–3 years look like

This is going to be a candid conversation.

Stay tuned.

#data #ai #AWSPartner #arango #NVIDIAGTC #theravitshow

One thing that became very clear to me at #NVIDIAGTC. AI is not slowing teams down. Data is. I spoke with Molly Presley, CMO and Mike Bloom from Hammerspace on The Ravit Show, the conversation quickly shifted from models to what actually blocks progress. Most enterprises are still dealing with scattered data across different systems, and the way they handle it today is still very manual. Teams are constantly searching for data, copying it, moving it, and trying to make sense of it before they can even start building AI.

What stood out is how Hammerspace is approaching this differently. Instead of forcing companies to move data into new systems, they focus on making existing data usable where it already lives. That means classifying it, organizing it, and preparing it for AI without adding more infrastructure or delay. They also showed how quickly this can happen. Going from fragmented data to something AI-ready in minutes, not weeks.

This feels like a shift. AI is not just about models or compute anymore. It is about whether your data is actually ready to be used.

#data #ai #nvidiagtc #hammerspace #theravitshow

A lot of AI projects look great in a pilot. Very few make it to production. At NVIDIA GTC, I sat down with Glenn Dekhayser from Equinix to understand why. And the answer is not what most people expect. It is not just about models. It is about everything around them. We talked about why so much enterprise data is still unusable for AI.

- Siloed systems

- Lack of structure

- No clear path from data to deployment

And then there is infrastructure. Because scaling AI is not just about training a model once. It is about running it reliably, securely, and close to where the data lives. That is where most projects break. One idea that stood out. Thinking of the data center as an “AI factory.” Not just storage. Not just compute. But a system designed to continuously turn data into outcomes. And that changes how enterprises need to plan.

From day one. If you are serious about AI, this is the shift. From experiments to infrastructure.

More conversations coming from GTC.

#data #ai #equinix #NVIDIAGTC #theravitshow

If you are building anything in AI right now, you need to pay attention to this. I recently sat down with Kirill Gavrylyuk, VP of Azure Cosmos DB, Microsoft on The Ravit Show, and one thing was very clear. The biggest challenge in AI today is not just models. It is the data layer. We talked about where things are actually breaking. Teams are building AI apps, but they struggle with memory, real-time data, and systems that need to scale globally. This is exactly where Cosmos DB is becoming critical.

What stood out to me is how practical this is getting. This is not about concepts anymore. It is about how real teams are building AI agents, how they manage state and memory, and how they design systems that actually work in production.

That is why Cosmos DB Conf 2026 next week matters.

This is not a theory-heavy event. It is focused on what is actually working. Real use cases, real architectures, real lessons. You will hear how teams are building RAG systems, scaling applications, and handling performance and cost in the real world.

Also, it is completely free. Which honestly makes it a no-brainer.

And you do not just watch. The Skills Challenge gives you a way to actually build and apply what you learn.

I will be covering this live on April 28 and sharing the key takeaways.

If you are serious about AI, data, or building systems that scale, this is worth your time.

Register here: https://lnkd.in/ggNcVK6Y

April 28 -- Show up and learn what actually works

#data #ai #azure #cosmosDB #microsoft #api #microservices #theravitshow

Everyone at NVIDIA GTC is talking about GPUs. But my conversation with Greg Matson from Solidigm was a good reminder. AI does not run on GPUs alone. It runs on data. And how fast you can move that data. What stood out is how much pressure new NVIDIA systems are putting on storage. The demand is not just for capacity anymore, but for performance and efficiency at a completely different level.

Solidigm is right in the middle of this shift. They are working with NVIDIA on things like liquid-cooled SSDs inside GPU systems and supporting new innovations like context memory extension. One insight that really stayed with me. If you lose context, GPU utilization can drop from 85% to 40%.

That is a massive hit. Which means storage is no longer just a backend problem. It directly impacts performance, cost, and scale. We also talked about where this is going.

Bigger drives, better efficiency, and designs that allow more GPUs to run within the same power limits.

This is the part of the AI stack that does not get enough attention. But it might be the one that decides how far AI can actually scale.

#data #ai #nvidiagtc #solidigm #theravitshow

Everyone in AI is talking about GPUs. Almost no one is talking about storage. At NVIDIA GTC, I sat down with Greg Matson from Solidigm, and this gap became very clear. Because behind every model, every training run, every inference pipeline… there is data. And how you store, move, and access that data changes everything. We talked about what companies are missing when they say they are “all in on AI.”

Most are focused on compute. Very few are rethinking their data and storage strategy. And that creates problems later.

Slower pipelines

Higher costs

Bottlenecks that no GPU can fix

One insight that stood out. The companies building serious AI infrastructure today are making storage decisions early. Everyone else is going to feel it later. We also touched on a bigger question. Are we heading toward a point where GPUs keep getting faster, but the data layer cannot keep up? Because if that happens, storage becomes the real limiter of AI progress. This is one of the most overlooked parts of the AI stack.

But it might be the most important. More conversations coming from GTC.

#data #ai #solidigm #NVIDIAGTC #theravitshow

AI is no longer just a technology conversation. It is becoming a national strategy. I spoke to with Thierry Pienaar, Chief Tech Officer at Hewlett Packard Enterprise at #NVIDIAGTC, and we discussed something that is quickly gaining momentum: Sovereign AI.

Countries and organizations are starting to realize that where AI is built, trained, and deployed matters just as much as the models themselves. This is not just about innovation. It is about control, data ownership, and long-term independence.

In this conversation, we break down:

- What sovereign AI really means and why governments are investing in it

- The infrastructure required to support sovereign AI at scale

- How sovereign AI workloads differ from traditional AI setups

- The idea of a “sovereign AI factory” and how it comes together

- Real examples of organizations already building in this direction

- Early outcomes and what leaders are starting to see

This shift is bigger than most people think. AI is becoming part of national infrastructure.

Learn more about HPE AI Factory here -- https://www.hpe.com/us/en/ai-factory.html?utm_campaign=FY26_Q2_AI_GE_GO_AMS_AMS_NVIDIA_GTC&utm_medium=DD&utm_source=HIP&utm_content=521216387&crid=%ecid!&plid=%epid!

Learn more about HPE Private Cloud AI -- https://www.hpe.com/us/en/private-cloud-ai.html?utm_campaign=FY26_Q2_AI_GE_GO_AMS_AMS_NVIDIA_GTC&utm_medium=DD&utm_source=HIP&utm_content=521216387&crid=%ecid!&plid=%epid!

#data #ai #nvidiagtc #sovereign #hpe #api #infra #theravitshow

Most enterprise AI projects are still stuck in pilot mode!!! But the real shift happening right now is not about better models. It is about where AI actually runs. I had a great conversation with Sadeepa Wijesekara from Hewlett Packard Enterprise at NVIDIA GTC, and one thing stood out:

Private cloud is becoming the foundation for enterprise AI. Not just for control and security, but for actually scaling AI across the business.

In this conversation, we discussed:

-- Why private cloud is becoming the preferred model for enterprise AI

-- How organizations are scaling AI securely from edge to datacenter

-- The rising demand for air-gapped and disconnected AI environments

What it takes to move from pilots to real production workloads

If you are serious about taking AI beyond experiments, this one is worth your time.

Learn more about HPE AI Factory here -- https://www.hpe.com/us/en/ai-factory.html?utm_campaign=FY26_Q2_AI_GE_GO_AMS_AMS_NVIDIA_GTC&utm_medium=DD&utm_source=HIP&utm_content=521216387&crid=%ecid!&plid=%epid!

Learn more about HPE Private Cloud AI -- https://www.hpe.com/us/en/private-cloud-ai.html?utm_campaign=FY26_Q2_AI_GE_GO_AMS_AMS_NVIDIA_GTC&utm_medium=DD&utm_source=HIP&utm_content=521216387&crid=%ecid!&plid=%epid!

#data #ai #hpe #NVIDIAGTC #theravitshow

I saw something at NVIDIA GTC that genuinely changed how I think about AI. Not another dashboard or abstract demo, but a full AI Factory built inside a bus by DDN. You walk in and suddenly AI is not a concept anymore, it is something you can see, touch, and understand end to end. I spent time on The Ravit Show inside the bus speaking with Jyothi Swaroop and Michelle (Pritoni) Scardino from DDN, and what stood out is how intentional this experience is. They are not just showcasing powerful infrastructure, they are trying to make AI feel accessible whether you are an enterprise, a government, or just starting out. Every layer is there in front of you, from compute to storage to networking, all working together in one place.

What really stayed with me was a conversation with VirooPax Mirji, who brought it back to the core problem most teams face today. It is not about building models anymore, it is about dealing with different kinds of data that do not fit neatly into one box. Video, text, multimodal inputs, all of it.

The work they are doing with NVIDIA focuses on making that data actually usable, helping teams search it, summarize it, and take action on it. That is where AI starts to deliver value.

Thomas Jorgensen from Supermicro explained the idea of an AI factory in a way that just clicks. He compared it to a car factory, where you input raw materials and energy, and the output is something functional. In this case, the input is data and power, and the output is intelligence. Underneath that are layers of software, infrastructure, and storage, all supporting different types of workloads like physical AI, reasoning systems, enterprise RAG, and agentic AI. It made the whole stack feel much more practical.

The demos made it even more real. Tim Yung from M2M Tech showed a digital twin of a robotic arm inside NVIDIA Omniverse, where a human interacts with it in VR to generate training data. That data is processed through DDN systems and then deployed into a real-world machine. You can literally see the loop from simulation to reality. And then there was Ameca, a humanoid robot from Engineered Arts, introduced by Leo Chen. It was observing, responding, and interacting in a way that felt surprisingly natural, and yes, we ended up having a dance off.

What I appreciated most about this entire experience is how it removes the noise around AI. It gives you a simple, clear picture of how things actually work, from data to outcomes. No overcomplication, just a real look at how AI gets built and deployed. This is the kind of experience more people need if we want to move from talking about AI to actually using it.

#data #ai #robot #ddn #machines #aifactory #simulation #theravitshow