Cisco executives argue that the distinction between product and model companies is disappearing, and that access to the 55% of enterprise data growth that current AI ignores will separate winners from losers.
VentureBeat recently caught up with Jeetu Patel, Ciscos President and Chief Product Officer and DJ Sampath, Senior Vice President of AI Software and Platform, to gain new insights into a compelling thesis shared by both leaders. They and their teams argue that every successful product company must become an AI model company to survive the next decade.
When you consider how product life cycles are becoming increasingly compressed, combined with the many benefits of digital twin technology to accelerate the time to market of next-generation products, this statement makes sense.
The conversation showed why this transformation is inevitable, supported by solid data points. The team claims that 55% of all data growth is machine data that current AI models have nothing to do with. OpenAI’s Greg Brockman estimates we need 10 billion GPUs to give every human the AI agents he or she needs, and Cisco’s open source security model, Foundation-Sec-8B, has already been downloaded 200,000 times on Hugging Face.
Why the model becomes the product
VentureBeat: You have stated that in the future, every product company will become a model company. Why is this inevitable and not just one possible path?
Jeetu Patel: In the future there will no longer be a distinction between model companies and product companies. Large product companies will be model companies. The close connection between model and product is a closed loop. To improve the product, improve the model, not just a UI shim.
These companies that are currently being formed are a thin layer on top of a model; their days are numbered. The real moat is the model you build that drives product behavior. This requires you to be good at two things at the same time: building great models in domains where you have great data, and building great product experiences powered by those models in an iterative loop where the models adapt and evolve as you have product improvement requests.
DJ Sampath: This becomes even more important when you think about cases moving to agents. Agents will be controlled by these models. Your moat will actually determine how well your model responds to the changes needed.
Harnessing the growth of machine data is critical
VentureBeat: You mentioned that 55% of data growth is machine data, but current models are not trained on it. Why does this present such a huge opportunity?
Patel: So far, models have been very successful at training on publicly available, human-generated data freely available on the Internet. But we’re done with the amount of public data you can crawl. Where do you go next? It’s all locked up in companies.
55% of data growth consists of machine data, but models are not trained on machine data. Every company says “my data is my moat,” but most don’t have an effective way to put that data into an organized pipeline so they can train AI with it and unleash its full potential.
Imagine how much log data is generated if agents work 24/7 and each person has 100 agents. OpenAI’s Greg Brockman said that if you assume every human has a GPU, you’re three orders of magnitude away from where you need to be; you need 10 billion GPUs. If you think like this and don’t train your models effectively with machine data, you’re underestimating your ability to realize AI’s full potential.
Sampath: Most models are trained on public data. The data contained within enterprises mainly consists of machine data. We unlock that machine data. We give every company a starting model. Think of it as a starter package. They will take that model and build applications and agents that are aligned with their proprietary data within their enterprise. We’re going to be a model company, but we’re also going to make it incredibly easy for every single company to build their own models using the infrastructure that we provide.
Why hardware companies have an advantage
VentureBeat: Many see hardware as a risk in the software and AI era. You claim the opposite. Why?
Patel: Many people look down on hardware. I actually think hardware is a great asset to have because when you know how to build great hardware and great software and great AI models and connect them all together, that’s when magic starts to happen.
Think about what we can do by correlating machine data from logs with our time series model. If there is a one degree change in your switch or router, you can predict a system failure within three days, something you previously couldn’t correlate. You identify the change, reroute traffic to prevent problems and resolve the problem. Get much more predictive value for outages and infrastructure stability.
Cisco is the critical infrastructure company for AI. This completely changes the level of stability we can generate for our infrastructure. Manufacturing is one of the top sectors when it comes to the volume of data generated daily. Combined with agentic AI and collected metadata, it completely changes the competitive nature of manufacturing or asset-intensive industries. With enough data, they can overcome disruptions around tariffs or variations in the supply chain, breaking out of the commoditization of prices and availability.
Cisco’s Deep Commitment to Open Source
VentureBeat: Why open source your security models if it seems to give away competitive advantage?
Sampath: The cat is out of the bag; attackers also have access to open source models. The next step is to equip as many defenders as possible with models that make the defense stronger. That’s really what we did at RSAC 2025 when we launched our open source model, Foundation-Sec-8B.
Funding for open source initiatives has stalled. There is an increasing exodus in the open source community, which is in need of sustainable, shared sources of funding. It is the responsibility of the business community to make these models available, and it provides access for communities to start working with AI from a defense perspective.
We are integrated ClamAVa widely used open source antivirus tool, with Hugging facewith more than 2 million models. Each model is scanned for malware. You need to ensure that the AI supply chain is appropriately protected, and we are leading the way.
Patel: We not only launched the security model that is open source, but also a Splunk for time series data. These correlate data; time series and security incident data, to find very interesting results. With 200,000 downloads on Hugging Face, we see resellers starting to build applications with it.
Taking the pulse of the customer after Cisco Live
VentureBeat: How are customers reacting to Cisco Live product launches?
Patel: There are three categories. First of all, completely enthusiastic customers: ‘We have been asking for this for a while. Hallelujah.’
Second, those who say, “I’m going to try this out.” DJ shows them a demo with white gloves, they do a POC and they are amazed that it is even better than what we said on stage in three minutes.
Third, there are skeptics who verify that each announcement appears on the exact days. Three years ago that group was much larger. Now that it has shrunk, we have seen meaningful improvements in our financial results and in the way the market views us.
We don’t talk about things three years from now, just within a six-month period. The load is so big that we have enough to discuss for six months. Honestly, our biggest challenge is keeping our customers informed about the pace of innovation we have.
Obsessed with customers, not hardware
VentureBeat: How do you migrate your hardware-centric installed base without causing too much disruption?
Patel: Instead of fixating on ‘hardware versus software’, start where the customer is. Your strategy can no longer be a perimeter-based firewall for network security because the market has shifted. It’s hyper-distributed. But you currently have firewalls that need efficient management.
We offer you a completely new firewall series. If you want to look at what we’ve done with public cloud: manage egress with Multicloud Defense without trust, not just from user to application, but from application to application. We built Hypershield technology. We have built a revolutionary Smart Switch. All managed by the same Security Cloud Control with AI Canvas on top.
We tell our customers they can go at their own pace. Start with firewalls, move to Multicloud Defense, add Hypershield enforcement points with Cilium for visibility, and add Smart Switches. You don’t need to add more complexity because we have a real platform advantage with Security Cloud Control. Instead of saying ‘forget everything and move on to the new’, which creates too much cognitive load, we start where the customer is and take them along on the journey.
What’s next: Encourage global partners to turn AI into a revenue opportunity
The interview concluded with discussions about November’s Partner Summit in San Diego, where Cisco is planning major partner activation announcements. As Patel noted, “It takes sustained, consistent emphasis to get the entire reseller engine moving.” VentureBeat is convinced that a globally strong partner organization is indispensable for every cybersecurity company to realize its long-term vision in the field of AI.
#Cisco #warns #companies #tapping #machine #data #strategy #incomplete