Cisco’s Jeetu Patel: Never fight a megatrend

{"text":[[{"start":6.88,"text":"Cisco was once the world’s most valuable company, briefly eclipsing Microsoft at the top of the dotcom boom on the back of a belief that its routers and networking infrastructure would prove invaluable for the technological revolution that the web spurred. It was hit hard when the bubble burst in the early 2000s, and its shares took more than a quarter of a century to return to the same high."}],[{"start":30.84,"text":"Over time the fibre networks that Cisco connected proved to be the backbone of the cloud that now underpins many of the digital services and apps that we use day-to-day."}],[{"start":40.8,"text":"The company is a key player in the current AI boom, building the equipment that helps link together the graphics processing units (GPUs) within large data centre clusters, supporting the steady flow of data as companies, including OpenAI and Anthropic, train models and deliver applications such as ChatGPT and Claude Code."}],[{"start":61.2,"text":"Jeetu Patel joined Cisco in 2020 under chief executive Chuck Robbins and was promoted to president and chief product officer last year. Here, he talks to the FT’s San Francisco correspondent Rafe Rosner-Uddin about Cisco’s strategy and internal AI push and the broader risks the technology poses."}],[{"start":80.16,"text":"Rafe Rosner-Uddin: Fundamentally, we are discussing Cisco within the frame of AI. You’ve signed deals worth more than $10bn from hyperscalers and so-called neoclouds [start-up cloud providers] this fiscal year. Can you explain the shift?"}],[{"start":95.96,"text":"Jeetu Patel: Let’s take a step back and ask what is happening in the market. As we move from this world of chatbots to agents, the one thing that’s forgotten is that these AI agents are going to be more like digital co-workers rather than tools. Their ability to consume [computing power] is unlike anything we had ever projected."}],[{"start":115.28,"text":"An agent conducting the exact same task as a human consumes 450 per cent more network bandwidth than a human. People talk a lot about [graphics processing units], but GPUs don’t really bring any value unless they have a network. Cisco had a natural advantage . . . because we are the networking company, and what the world needs is massive amounts of capacity."}],[{"start":null,"text":""}],[{"start":135.96,"text":"We make our own silicon for the networking switches for routers and WiFi. We then make our own photonics [a technology that uses light to transmit data on silicon microchips], optics [transceivers inserted into networking equipment such as switches, routers and servers] and optical systems [circuits transmitting data]. On top of that, we’re also a silicon company. And then we’re also a software company. We have a full stack."}],[{"start":159,"text":"So when someone asks, are you a hardware or a software company, we’re actually vertically integrated and the closest analogy I can give is: think about us like Apple but for infrastructure."}],[{"start":170.96,"text":"RRU: Some of your partners are also pushing into hardware. Nvidia for example is making a push into optical switches. How do you handle that dynamic where partners are now competitors?"}],[{"start":180.92,"text":"JP: Six years ago, when I joined, Chuck and I had this exact conversation. We had to make sure that we adopted the mental model where we cannot be arrogant enough to think that all the innovation that happens in the world comes from Cisco."}],[{"start":194.64,"text":"It used to be that you had these moats. You would create these meaningful differentiators, and you would lock in a customer. I fundamentally believe today that’s the wrong mental model to go out and apply. What you should be thinking about is how do I earn my customer’s admiration every single day, rather than trying to lock them in."}],[{"start":213.08,"text":"RRU: AI seems to be lifting a lot of boats. Some of your competitors are also demonstrating quite remarkable levels of growth. How much of what we are seeing from Cisco is Cisco itself rather than this wider context?"}],[{"start":225.28,"text":"JP: I have a lot of respect for all our competitors. I think they’re very capable. We don’t take them lightly at all. What we have to think about is what is our structural advantage where we can do better than everyone else."}],[{"start":236.58,"text":"We make our own silicon for our switches. Some of our competition relies on silicon providers . . . That gives us a huge advantage because we can now control our own destiny. You can have a much better streamlined supply chain if you make all of the pieces and you’re not dependent on someone else."}],[{"start":252.28,"text":"It’s showing up in our numbers. Last quarter, we grew our bookings 35 per cent. We increased revenues 18 per cent. You can just start to see the compounding effects play out."}],[{"start":264.24,"text":"RRU: I want to zoom out into the wider moment we are in. Sam Altman [OpenAI’s chief executive] has expressed worry that there are some data centre operators who either don’t have sufficient revenues or the ability to finance their sites. We are almost certainly in a bubble. The question I have is what type of bubble?"}],[{"start":282.84,"text":"JP: In my mind two things can hold true at the same time. You could be in the middle of a secular shift that’s going to fundamentally change the way humanity operates. And this is the largest of its kind. In that kind of shift you will have some frothy valuations."}],[{"start":296.84,"text":"Why are there frothy valuations? Because when you have a secular shift, a bunch of different business models start to get experimented on and get funded. And as those experiments are being run, some of them succeed and become the OpenAIs and Anthropics of the world and others don’t do as well."}],[{"start":312.24,"text":"When you ask, are we in a bubble? [Of course] there are going to be certain areas that are overvalued. But that does not mean that we’re not experiencing a sustained secular shift."}],[{"start":null,"text":""}],[{"start":322.84,"text":"RRU: We are watching vast amounts of capital being allocated towards OpenAI and Anthropic. They are not profitable and are subsidising consumption. How do you view this dynamic?"}],[{"start":333.16,"text":"JP: It takes about $3bn to $5bn to train a frontier model [an AI system at the cutting edge of capability]. And that model has a certain window [of relevance in the market] and the consumption of that frontier model happens during that window. Each one of those models tends to be pretty profitable on their own. But what ends up happening is, while you’re training the one model, you start training the next model because there’s such a race."}],[{"start":357.08,"text":"RRU: You joined Cisco not that long ago but it is a company that came to notoriety because of the dotcom bubble. That history must be quite pressing for you in the current era. How do you move defensively to ensure we do not get a repeat of the past here?"}],[{"start":371.56,"text":"JP: We are now in a global race. There is a certain tempo and momentum at which AI advancements are happening, and that is set by the collective market. I don’t think that slows down."}],[{"start":382.48,"text":"What do we do to make sure that the turbulence we had experienced in previous waves doesn’t happen? Some of that is bound to happen in the sense that when you have this much experimentation going on, not everyone is going to succeed."}],[{"start":395.12,"text":"What we don’t discuss are the companies that tried to avoid the internet . . . How many of them are still relevant?"}],[{"start":402.24,"text":"RRU: I assume internally at Cisco you do look at this and ask, what are some of the risk factors here? How do we navigate them?"}],[{"start":409.6,"text":"JP: The danger scenario . . . is we get into a massively constrained state of undercapacity for a sustained amount of time. It’s very hard to see the signals [in AI demand] that say this is going to materially slow down."}],[{"start":421.4,"text":"RRU: It seems like you don’t see any downside risks to this bar not keeping up with the race?"}],[{"start":428.12,"text":"JP: What I’m saying is the risk of going slow is higher than the risk of going fast [in AI model development]."}],[{"start":null,"text":""}],[{"start":434.08,"text":"RRU: What has the AI deployment looked like within Cisco?"}],[{"start":437.76,"text":"JP: Most people think of AI as a productivity enhancer. I like to think of AI as something that makes the impossible happen. We have one product that has been written 100 per cent with AI [and no human-written lines] and by the end of this calendar year we will have six more."}],[{"start":453.32,"text":"By the end of next year, 70 per cent of all our [software] products will be 100 per cent written with AI. In two years, my hope is that we have no legacy [outdated] lines of code."}],[{"start":463.28,"text":"RRU: These are the soundbites that scare people because they think about job displacement. But all of that code generation has people behind it."}],[{"start":471.36,"text":"JP: This narrative that AI is going to take all our jobs doesn’t make sense to me. Because every time there’s a step-function improvement [a massive leap forward] with AI in some area, a human bottleneck emerges in another."}],[{"start":484.24,"text":"RRU: You’ve spoken [in the past] about the fact that Cisco will have net new additions this year in terms of headcount but we also got a lay-off announcement. It feels like these decisions feed a [frenzied news] cycle and companies are slow to correct the assumption that AI is coming for people’s jobs."}],[{"start":null,"text":""}],[{"start":502.52,"text":"JP: We were very clear about this when we did the restructuring. We said we want to make sure we balance the portfolio and put our resources wherever there’s the strongest amount of demand."}],[{"start":514.08,"text":"We had a tremendous amount of demand in silicon. A tremendous amount of demand in photonics. We needed to make sure that we hired more over there. A lot of the people [who] were part of the restructure could get jobs in those areas but we need to make sure we’re doing the right sizing of skills. It’s just a portfolio management exercise."}],[{"start":532.4,"text":"RRU: Did those people get reallocated? Did you keep them?"}],[{"start":536.56,"text":"JP: Many of them we did. What you have to look at is the total amount of people and spend in R&D. We’re going to spend around $8bn in R&D this year. If you compare that to where we were two, three years ago, it’s gone up quite a bit."}],[{"start":550.48,"text":"In fact, R&D as a percentage of revenue . . . is pretty high. Our revenue is also growing at a pretty fast pace. You simply can’t hire people as fast as revenue is growing at times."}],[{"start":561.2,"text":"RRU: Are you experiencing pushback internally around AI deployments?"}],[{"start":565.32,"text":"JP: We did when we first started. Now everyone has [bought into the technology’s benefits] . . . [We said] stop worrying about AI taking a job because what is guaranteed is that if you don’t use AI your job will go away."}],[{"start":null,"text":""}],[{"start":578.8,"text":"The two cohorts that have worked very productively with AI are those very early in their careers and the very experienced. The people [who] have taken the longest time to come up to speed are in the middle."}],[{"start":589.84,"text":"RRU: What are you doing to bring the middle layer of people up to speed?"}],[{"start":593.56,"text":"JP: You have to do a combined approach of giving them all the tooling, incentives and training apparatus so they can get more skilled and dexterous with AI. Also, make it very clear that if they don’t use that tool they will not make it through the transition."}],[{"start":609.58,"text":"This was one of those few things where I did not engage in a debate with people on whether or not we need to use AI. We decided we’re going to do it. If people debated it, saying “I don’t believe this should be the case,” that’s fine. They can go somewhere else."}],[{"start":623.48,"text":"RRU: We’ve reported on instances at companies like Amazon where employees have hated some of their AI tools and management has ploughed ahead regardless, with dire consequences. How do you avoid being blinded by conviction?"}],[{"start":636.84,"text":"JP: There are very few times when a megatrend is happening in front of your eyes. One thing you have to use as a cardinal rule is to never fight the megatrend. Always use it as a tailwind."}],[{"start":648.24,"text":"You have to know the difference between a megatrend and a hype cycle. A hype cycle is a passing fad."}],[{"start":653.84,"text":"RRU: Did you buy Bitcoin?"}],[{"start":656.8,"text":"JP: Not enough to materially change my life. But Web 2.0 [the emergence of social media networks and apps] was a classic example of a hype cycle."}],[{"start":666.6,"text":"At some point in time, I became fully convinced that AI was a megatrend and companies had to act decisively. We didn’t seek input on whether or not we should use it. We only sought input on how we could be the most successful in our use of AI."}],[{"start":680.8,"text":"RRU: Did you see things like “tokenmaxxing” [where workers maximise AI consumption without thinking about costs or outcomes] from your staff?"}],[{"start":689.8,"text":"JP: We actually encouraged it initially because at first you have to . . . get familiar with [AI] before you get good at it. Now we’re at the point where, like with everything else, employees have a budget and have to make sure their tokens are used efficiently within that budget and produce outcomes with them."}],[{"start":707.44,"text":"RRU: A big part of Cisco’s business is security. One of the largest acquisitions in this company’s history was [cyber security company] Splunk. We recently had the Hugging Face incident [where OpenAI bots independently attacked the Hugging Face’s production infrastructure]. What were your takeaways?"}],[{"start":723.76,"text":"JP: Defences can no longer be human scale . . . This is the first time that a defender [someone who protects systems from cyber threats] could actually have an advantage compared to the adversary. When [OpenAI’s] Astra came out or when [Anthropic’s Mythos] came out, a few companies were given a head start to get ahead and protect our critical infrastructure [from these models being used by attackers to identify vulnerabilities]."}],[{"start":740.28,"text":"I am equally paranoid and excited. And I’m paranoid about the fact that, gone wrong, this thing could cause a lot of damage."}],[{"start":null,"text":""}],[{"start":747.68,"text":"RRU: The thing about the Hugging Face incident is that it demonstrates that people making these models do not necessarily have a full grip or control. If Cisco had suddenly hacked a partner or a competitor, I think the headlines would have been very different from what we saw around Hugging Face."}],[{"start":764.32,"text":"What is your perspective on the degree of trust and confidence that we can have in the companies that are producing these models?"}],[{"start":769.96,"text":"JP: It is going to get harder and harder to know the difference between a malicious attack or a breach by an external entity and an agent just exercising poor judgment."}],[{"start":779.96,"text":"No one actually poisoned [hacked or manipulated] that OpenAI agent that broke containment and went to Hugging Face to steal evaluation scores [an AI model’s performance on tasks like accuracy, safety or prompt completion]. It just happened because it was following the goal given to it, which is to figure out a way to pass the evaluation."}],[{"start":793.76,"text":"We have to build agentic controls that will be done in real time so that when you start detecting a pattern of anomaly you can immediately go out and apply guardrails."}],[{"start":802.6,"text":"RRU: Do you feel like we’ve caught up with the capabilities?"}],[{"start":806.92,"text":"JP: Can we get ahead? Yes. Have we caught up? No. The [security] risk is the highest it’s ever been."}],[{"start":813.4,"text":"RRU: Do these companies need to pause development?"}],[{"start":816.44,"text":"JP: You can’t because you’re in a race. The only way you can [prevent harm] is by making sure that people have access to these technologies and that there’s asymmetry between defenders who can use frontier models and adversaries who might not be able to use them yet. We need to make sure that gap does not converge."}],[{"start":834,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1790254556_9910.mp3"}