Cisco’s Patel Hits $9.3B in AI Orders, Threatens to Fire Holdouts

思科高管Patel斩获93亿美元AI订单，警告解雇抵触变革者 AI基础设施并未过度建设，因为目前只有不到2%的知识工作者运行自主Agent，而每个Agent消耗的网络带宽比人类高出450%。 Jeetu Patel详细阐述了思科如何积极转型为关键的“卖铲人”基础设施层，通过定制芯片、网络、网络安全和可观测性为全球AI集群提供动力。Patel驳斥了AI仅用于削减员工人数的观点，他认为执行层面的自动化只会将人力瓶颈转移至战略产品判断层面，最终需要的工程师只会更多而非更少。他还论证了数据中心容量目前仍严重受制于供应而非建设过剩，特别是考虑到目前使用高带宽自主Agent的知识工作者尚不足2%。除技术和运营洞见外，Patel还分享了他从在印度的受虐成长环境和在Sizzler端盘子打工，一路走向高管职位的个人历程，勉励建设者以数十年为期沉淀复利技能，并珍视所获得的难得平台。 思科在两年内将其超大规模AI基础设施业务订单额从零扩大至93亿美元，在630亿美元的企业总体业务规模基数下，远超最初设定的10亿美元目标。 (18:06) Jeetu Patel驳斥了对AI基础设施泡沫的担忧，指出与互联网泡沫时期投机性的暗光纤建设不同，在重度用户渗透率不足2%的情况下，当前每一寸新增算力都会被瞬间消耗殆尽。 (21:27) 由于反复的工具调用、上下文检索和中间Token交换，自主AI Agent在执行相同任务时消耗的网络带宽比人类用户高出约450%。 (21:27) 为了在避免混乱的同时推进大规模企业级执行，思科采用了“三分法则”领导模式，在资深内部领航者、外部规模化高管以及保持战时紧迫感的被收购初创公司创始人之间取得平衡。 (4:48)

Jeetu Patel outlines Cisco’s strategic transition into a primary infrastructure and security provider for the AI era, arguing that sustained enterprise relevance requires pairing startup-level agility with massive operational scale. Rather than viewing AI as a labor-reduction mechanism, Patel contends that the technology will expand compute demands and technical headcount while exposing critical bottlenecks in physical networking and autonomous agent governance. Scaling AI Infrastructure Across Hyperscaler Networks (18:06) Patel positions Cisco as a foundational picks-and-shovels provider for AI, supplying custom silicon, optical transceivers, and switching to support three distinct network tiers: scale-up within server racks, scale-out across data center rows, and scale-across to link geographically separated facilities restricted by local electrical grids. With gigawatt-scale data centers costing between $35 billion and $50 billion to construct, networking infrastructure represents roughly 10% to 15% of total capital expenditure. In this environment, Cisco's hyperscaler AI infrastructure orders expanded from an initial $1 billion target to $2.3 billion in year one, eventually reaching approximately $9.3 billion over two years. Addressing bubble concerns, Patel rejects comparisons to the late-1990s dot-com crash, arguing that modern AI compute is consumed immediately upon deployment rather than sitting idle as dark fiber once did. He emphasizes that less than 2% of knowledge workers currently utilize AI agents, yet a single agent consumes roughly 450% more network bandwidth than an equivalent human user due to continuous tool executions, context retrievals, and token transfers. Running an Enterprise Turnaround in Founder Mode (1:46) Patel approaches executive leadership at Cisco through a business-owner mindset rather than operating as a siloed functional leader. During major platform shifts, he contends that executives must take direct ownership of cross-departmental challenges—actively evaluating issues across product, sales, and marketing while inviting reciprocal scrutiny—to maintain wartime execution speeds. To prevent an enterprise of Cisco's scale from lapsing into disarray, Patel structures leadership around a deliberate Rule of Thirds. The structure balances tenured Cisco operators who understand internal execution systems, external scalers hired from the broader enterprise market to manage complex systems, and acquired startup founders given operational charters significantly larger than the companies they sold. Workforce AI Adoption and Development Cycle Compression (9:04) Patel argues that generative AI will increase overall engineering headcount rather than reduce it, because automating code creation merely pushes the operational bottleneck to code review, and subsequently to strategic product judgment. To accelerate adoption across Cisco's 32,000 engineers, leadership distributed unlimited tokens and established a strict operational expectation that non-adopters would lose relevance. A turning point occurred when an internal engineering team refactored a legacy 500,000-line codebase down to roughly 100,000–120,000 lines within two weeks, leading to an embedded partnership with OpenAI's Codex team. Managing this shift requires synchronizing radically different product timelines across Cisco's stack. While AI applications iterate weekly or quarterly and traditional operating systems operate on 12- to 18-month cycles, hardware manufacturing demands 18 to 24 months, and production silicon requires an unyielding five-year timeline driven by deep intellectual property fabrication and physical SerDes interface constraints. Securing Non-Human Identities and Autonomous Agents (25:10) As organizations transition from supervised prompt-based AI to delegated autonomous agents, Patel warns that untrusted delegation presents existential risks of data poisoning, agent drift, and prompt manipulation. Because modern adversarial attacks operate at automated speeds, Patel asserts that abstaining from AI deployment leaves enterprises defenseless against automated threats. To secure enterprise adoption, Cisco assembled a four-part agentic defense framework that audits model data pipelines, tracks agents as distinct non-human identities, executes algorithmic red teaming against jailbreaks, and applies dynamic runtime interception to clamp deterministic boundaries onto non-deterministic models. This security stack integrates Cisco's core networking fabrics, cybersecurity tools, and telemetry from its acquisition of Galileo alongside Splunk. Career Compounding, Long Horizons, and Platform Leverage (36:40) Reflecting on his progression from waiting tables at Sizzler to executive roles at DocuLabs, EMC, Box, and Cisco, Patel maintains that personal ambition constrains career trajectory far more than innate intellect. He highlights the power of compounding, noting that improving by 1.27% daily produces a 100-fold gain over a single year, while urging leaders to evaluate their careers using 20-year planning horizons rather than quarterly increments—provided their business model aligns with their underlying goals. Citing an encounter at the Taj Mahal with a tour guide who spoke up to 14 languages yet earned roughly $10 a day due to systemic constraints, Patel emphasizes that access to modern tech platforms is a rare privilege. He cautions startup founders joining large enterprises via acquisition against cynicism, advising them to leverage incumbent distribution to maximize technical and societal impact. Patel's thesis rests on the premise that global AI infrastructure is profoundly underbuilt relative to future agentic demand, leaving Cisco uniquely positioned across silicon, networking, and observability. However, this growth remains dependent on whether enterprises can reliably govern autonomous agents and whether multi-year physical silicon manufacturing cycles can keep pace with rapid software-level iteration.

[0:00] Jeetu Patel: This is a magical time to be alive in tech. [1:46] Jeetu Patel: I didn’t write a single line of code. [1:46] Jeetu Patel: Everything is your job. And so if something’s not working well in sales, I’ll go talk to the people over there. Something’s not working well in marketing, I’ll go to talk to the people over there, and I fully expect them to come poke holes in my business. [11:47] Cisco engineering leader: Hey, I just want to let you know I was wrong. You were right. [23:10] Jeetu Patel: Because it's fun to check email when you have three emails. [25:10] Jeetu Patel: The difference between trusted delegation and untrusted delegation is the difference between market leadership and complete abject bankruptcy. [36:40] Jeetu Patel: business is a team sport [36:40] Jeetu Patel: it was terrible to go change the world. [39:31] Raj: I want to honor the people that come here... I should not be presumptuous enough to think that they're going to speak my language. I should speak their language. [39:31] Jeetu Patel: If you happen to be afforded a platform, my only ask of people is don't squander it

[0:00] My strong assumption is moving forward [0:02] that we will actually need more [0:03] engineers, not less. The goal is not [0:05] efficiency. The goal is to be able to go [0:07] out and make sure that we benefit from [0:10] the original insights that AI generates [0:12] that allow us to solve problems we could [0:14] have never dreamt of solving before. Do [0:16] you know how many people in their 50s, [0:18] how much money they would pay to [music] [0:19] go back to being 19 if they could? This [0:22] is a magical time to be alive in tech. [0:30] >> Today, I'm talking with Gee Two Patel, [0:32] Cisco's President and Chief Product [0:33] Officer. We're going to discuss how Gee [0:35] Two is bringing a founder's mindset to [0:37] Cisco, [0:39] going all in on AI, what's happening [0:41] inside the massive data center build out [0:43] that they're at the center of, and then [0:44] we'll talk about what founders can learn [0:46] from Gee Two's incredible personal [0:47] story. Gee Two, thanks so much for being [0:49] here. [0:49] >> Thanks for having me. It's great to see [0:51] you. [0:51] >> You have this incredible background um [0:53] from growing up in India, waiting tables [0:56] at Sizzler to ending up as the president [0:58] of one of the world's largest companies, [0:59] but before we talk about all that, um I [1:02] want to talk about like [1:03] Cisco today and your unique role and [1:07] what you're doing there. [1:08] >> The way I think about my role is my role [1:10] is I just make sure that we help build [1:13] the best products in the world [1:15] that all of our customers can use at [1:17] scale. What Cisco does though, which is [1:19] even more interesting than what I do, [1:21] is we Think about us as the picks and [1:23] shovels company during the gold rush. [1:26] We are the AI infrastructure company for [1:28] the AI era during this [1:31] massive movement that's going on right [1:33] now, [1:33] which is a secular shift, and we want to [1:36] make sure that we provide the right [1:38] level of infrastructure so that it just [1:41] AI just runs the way that you want it [1:43] to, [1:44] and it runs securely and safely. [1:46] >> And that description of Cisco, which I [1:49] love, picks and shovels for the AI era, [1:52] clearly you would not have described [1:53] Cisco that way four years ago. So, all [1:55] of this is recent. This is a [1:56] transformation of a very large company [1:58] that's already it was already operating [2:00] at an enormous scale. Um what has it [2:03] been like to come in and try to [2:04] transform an organization at this scale [2:06] and that sort of time frame? [2:08] >> You know, I think in these [2:09] massive kind of shifts that happen in [2:11] companies, uh typically the leaders get [2:14] an [2:16] undeserved level of credit and the [2:18] people that do the work don't get enough [2:19] of the credit. But the reality is is I [2:22] didn't write a single line of code. [2:24] When we were doing this, it is the team [2:26] that did that. Uh but what what we did [2:29] was we made sure that we actually set [2:30] the right direction, [2:32] hired the right people in the right [2:33] jobs, made sure that we had the right [2:35] kind of mechanisms to figure out what [2:38] markets we're going to go pursue and do [2:40] it with everyone feeling like they can [2:42] they can win. [2:44] >> Can you tell the story of how how you [2:46] ended up at Cisco and what you're doing [2:48] there? [2:48] >> So, I ended up at Cisco I was at Box. I [2:50] was uh running product with um you know, [2:52] for Aaron uh Levie um who was a founder [2:55] himself. And so, my entire kind of [2:57] mental model [2:59] got reinforced again when I came to Box [3:01] because I had started my own company [3:02] years ago and then I went to EMC. EMC [3:05] was a very large company. [3:07] I'm like, "Oh, this is like very [3:09] corporate." And then I went to Box I'm [3:10] like, "Okay, this can be fun because you [3:12] can actually operate just like I used to [3:14] operate with my own company." [3:16] And I think there's two kinds of people. [3:17] There's folks that are very good [3:19] functional leaders and they love being [3:22] in their lane. [3:23] Uh no one tell me what to do in my lane. [3:25] I'm not going to tell you what you do in [3:26] your lane. And then there's people [3:28] who've run some kind of business and [3:31] they think of themselves as owners of [3:32] the business and they are completely [3:34] comfortable poking in other people's [3:36] businesses all the time. [3:38] And I tend to be of that ilk. And so, I [3:42] personally feel like when you have this [3:44] founder mode approach, [3:46] it's very good for wartime and it's very [3:49] good for a time when [3:51] the market's going through a shift, [3:53] things are moving at a very fast clock [3:55] speed, and what you have to do is you [3:56] just have to make sure that you don't [3:58] completely alienate the people who might [4:00] not be in founder mode, and kind of help [4:03] them understand the benefit that if you [4:05] do go in founder mode, we can actually [4:06] operate at a very different frequency [4:08] than what we've been operating so far. [4:10] And so that's what we started doing at [4:11] Cisco, and I feel like I tell my [4:14] my peers and everyone in my company, I'm [4:16] like, I feel like I'm a late-stage [4:17] co-founder of Cisco. I was 36 years [4:20] late coming in, but I feel like I own [4:23] the company, and this is my company, and [4:24] if um [4:25] if anyone from the outside tries to um [4:28] screw the company, beware, because I'm [4:30] going to make sure I fight for it. And [4:31] we got to make sure that we we take a [4:33] lot of pride in the craftsmanship and [4:35] what the company stands for. Everything [4:37] is your job. And so if uh something's [4:39] not working well well in sales, [4:42] I'll go talk to the people over there. [4:43] Something's not working well in [4:44] marketing, I'll go to talk to the people [4:45] over there, and I fully expect them to [4:47] come poke holes in my business. [4:48] >> Is this one of the reasons why you'd [4:50] like to bring in former founders at [4:52] Cisco? [4:52] >> You know, I also recognize the [4:55] um and appreciate the value of seasoned [4:58] operators who are systems thinkers. [5:01] Because if you had everyone at Cisco who [5:02] was a founder, it'd be a show. [5:04] >> Yeah. [5:04] >> Uh so you have to make sure that you get [5:06] you get a good balance. So the way we've [5:07] done it is we have a [5:09] I try to call it a rule of thirds. So a [5:12] third of my leadership team are people [5:14] that know how to maneuver Cisco, [5:16] that are at scale operators. The second [5:18] third are people that I bring in from [5:20] the outside, from competition, from the [5:22] market, that are very market in. [5:24] Uh that might not know how to maneuver [5:26] Cisco, but they're systems thinkers, and [5:28] they've operated at scale. And then the [5:29] third third [5:30] are co-founders and CEOs of companies [5:33] that we've acquired. [5:34] But those people, we give them charters [5:37] that are larger than the company that [5:38] they sold us. [5:39] >> You said earlier [5:40] that this way of operating is especially [5:43] important in wartime. Do you feel Cisco [5:45] is in wartime? [5:46] >> I feel like Cisco is in wartime right [5:48] now [5:49] um in the sense that we have [5:51] >> [snorts] [5:51] >> we have made one of the most successful [5:53] turnarounds that you've seen from a [5:55] product perspective. Um I think we are [5:58] now squarely in the AI infrastructure [6:00] game. [6:01] Uh which is something that no one would [6:02] give us credit for three or three, four [6:03] years ago. But [6:06] if you look forward the next three or [6:09] five, seven years [6:11] um the amount of [6:13] uh opportunity that exists in front of [6:16] us is so [6:18] massive compared to anything that we've [6:20] experienced before. [6:22] That we have to make sure that we can we [6:23] can capitalize on all of that. And that [6:25] means that we have to operate at I like [6:27] I like I like to tell my team internally [6:29] we should operate like the world's [6:30] largest startup. That means you have to [6:32] operate at speed but with scale. [6:34] If you just operate at ski at speed then [6:36] you're just a startup. If you just [6:38] operate at scale then you're just [6:39] largest company. If you want to operate [6:41] as a world's largest startup that has to [6:43] be with speed and scale. [6:45] >> What are the key things you need to get [6:46] right in the next few years to seize [6:48] that opportunity? [6:50] >> Let's assume right now that we have sub [6:52] 2% of the [6:54] of humans that are using agents in any [6:56] kind of power user capacity. [6:58] And [7:00] think about the fact that the entirety [7:02] of infrastructure that gets produced [7:04] today [7:05] is getting consumed. [7:06] We are short on infrastructure. We are [7:08] supply constrained with only sub 2% of [7:11] the people using agents. [7:12] >> Imagine when 25 30, 40, 50% of the [7:16] people start using agents. [7:18] The amount of infrastructure you're [7:19] going to need is going to be at a very [7:20] different scale proportion that we have [7:22] right now. And we want to make sure that [7:23] we can help solve that problem. That's [7:24] problem number one. [7:26] Problem number two [7:27] is these agents in fact this this week [7:30] was a very interesting week because [7:32] these agents are going to go rogue. [7:34] At times they might actually have a mind [7:36] of their own. It's actually harder and [7:38] harder to distinguish between when an [7:39] agent has [7:41] uh is operating with uh intent because [7:44] of [7:45] an adversary influencing the agent or [7:47] poisoning the agent or an agent just [7:49] thinks it's the right thing to do and [7:52] they're just kind of, you know, [7:53] diligently going out and [7:55] following your slash goal command. [7:57] >> The paperclip optimizer problem. [7:58] >> Exactly. [7:59] >> Yeah. [7:59] >> And so what we have to have is you have [8:02] to have some kind of safety, security, [8:04] and observability capabilities in place. [8:07] Um and so that's the second big problem [8:09] that we want to solve is we want to make [8:10] sure that AI is safe [8:12] to use for everyone and that that's that [8:14] doesn't become this constant fear [8:16] because of which people don't use AI. [8:18] Uh in fact, I think it's the other way [8:20] around. If you if you don't use AI, [8:22] you'll be less safe. [8:24] And and that is that is true because [8:26] most of the attacks are happening at [8:27] machine scale, so you have to have the [8:29] defenses at machine scale as well. So, [8:31] we're one of the largest, you know, [8:32] networking companies in the world [8:34] that provide the infrastructure. [8:36] We're also one of the largest security [8:38] companies in the world. [8:39] We want to make sure that we do that and [8:40] we're also one of the [8:42] you know, one of the most significant [8:43] kind of machine data companies in the [8:44] world with Splunk. [8:46] >> Oh, that's right. [8:46] >> And so the combination of those three [8:49] should give us uh all of the underlying [8:52] apparatus that's needed so that we can [8:54] become critical infrastructure for the [8:57] critical infrastructure companies for [8:58] the AI AI era. [9:00] >> You've gone all in on AI both in terms [9:02] of investing the AI and data center [9:04] market, but also in terms of bringing AI [9:06] into Cisco to make all make everything [9:09] more efficient. I'm actually really [9:10] interested to learn what it looks like [9:12] to adopt AI at a company the size of [9:15] Cisco. [9:15] >> It can't be around efficiency. [9:19] The goal is not efficiency. The goal is [9:21] to be able to go out and make sure that [9:23] we benefit from the original insights [9:25] that AI generates that allow us to solve [9:27] problems we could have never dreamt of [9:29] solving before. [9:30] Which is very different from just get to [9:32] be 20% faster at doing this or make sure [9:35] that you have less people. In fact, my [9:36] strong assumption is moving forward that [9:38] we will actually need more engineers, [9:40] not less, [9:41] as we move forward. Because every single [9:43] time I've seen a step function [9:44] improvement in AI, [9:46] I've seen a bottleneck that's emerged [9:48] that's a human bottleneck. You know, [9:50] coding is automated, code review becomes [9:52] a bottleneck. Code review is automated, [9:54] judgment on what to build and what not [9:56] to build becomes a bottleneck. So, I I [9:58] do feel like we we're going to need to [10:00] have more people. But, what what was [10:02] interesting was getting 32,000 engineers [10:06] and the rest of the company one to taste [10:09] the honey, and two not be fearful of the [10:12] fact that if you use this, then you're [10:13] going to actually be creating a trap for [10:15] yourself where you might lose your job. [10:17] Instead, what we did was we did [10:19] something completely opposite. We [10:21] guaranteed our employees that if you [10:23] don't use AI, you will lose your job. [10:26] And so, it it was a very different way [10:28] to kind of approach the problem. [10:29] That wasn't a very controversial [10:31] statement to make because think about [10:33] today, if you have someone that didn't [10:36] know how to use a PC or didn't know how [10:37] to use the internet. Like, would you [10:39] ever hire them in a knowledge worker [10:40] job? You would never. So, it's not that [10:42] controversial to say that if you are [10:44] seeing a secular shift and if there's [10:45] this massive platform that's going to [10:47] get refactored [10:49] with AI, [10:50] then anyone not using it is just not [10:53] going to be that relevant. [10:55] And so, we always believed that there's [10:57] going to be only two kinds of companies [10:59] and two kinds of people. Ones that are [11:00] very fluent and dextrous with AI, and [11:02] the others that are going to be really [11:04] struggling for relevance. [11:05] And so, we just made it very, very clear [11:07] for people that if you happen to be in [11:08] that second camp, [11:11] because you're not using AI, chances are [11:13] that Cisco's the wrong place for you. [11:15] And therefore, what you should do is go [11:17] all in, and we will give you all the [11:18] tooling and all the support to go all [11:20] in. [11:21] And initially, we actually gave everyone [11:23] unlimited tokens. To say, "Just start [11:25] using it." Because the way I think about [11:26] it is you have to get It's like riding a [11:28] bicycle. You have to first get familiar [11:30] with something, then you have to get [11:32] good at it, and then you can become [11:34] efficient with it. [11:35] >> Yeah. [11:35] >> But if you try to go efficiency first, [11:37] you'll never get familiar or you'll [11:39] never get good. And so we started doing [11:41] that. [11:42] Got a lot of initial resistance [11:44] from a lot of folks. The first few [11:46] months were rough. [11:47] And the the seminal moment for me was I [11:50] have this one engineer who's a [11:53] engineering leader on my team. One of [11:55] the most amazing people. [11:57] And he is always a skeptic. [12:00] He's always going to tell me, "G 2, [12:01] those ideas are crazy. I don't think [12:03] it's going to ever happen." [12:04] And he called me one day in the evening, [12:07] and he was as sincere as he could be, [12:09] and he said to me, "Hey, I just want to [12:11] let you know I was wrong. You were [12:13] right. [12:14] The models are moving at a pace now. [12:16] And I had some benefit because, you [12:18] know, Kevin Wheels on our board, and I [12:19] had some, you know, kind of I I knew [12:21] some people within the ecosystem that [12:23] every time I talked to someone in the [12:24] startup world, um I was getting a very [12:27] different set of data than I was I was I [12:29] was getting internally. [12:30] And so Will says to me finally, [12:32] "This is now moved where we have now [12:34] taken [12:36] and refactored like, [12:38] you know, half a million lines of code, [12:40] and we were able to refactor it into [12:42] like [12:43] 100 120,000 lines of code or something. [12:46] And so 5x improvement, and we were able [12:48] to do that within like 2 weeks. [12:50] >> 2 weeks, yeah. [12:50] >> You know? And so one, [12:53] you are right that this is actually [12:54] going to move really fast. So I can't [12:57] just think about what we have today. I [12:58] have to think about the fact of where is [13:00] this going to be in 3 months from now? [13:01] Not a year from now, 3 months from now. [13:03] And assume that and start making the [13:05] changes today. [13:06] But second is I'm also feeling very [13:09] anxious about the fact that this might [13:11] mean a lot of bad things for people, and [13:13] we might have to make a lot of changes [13:14] in our staff. [13:15] >> Mhm. [13:16] >> And I'm like, "Or [13:17] we can actually motivate everyone [13:19] >> Mhm. [13:20] >> to make sure that we can um we can all [13:22] get better as a result of the use of [13:24] this." Then, you know, it just kept [13:25] compounding. We actually became the [13:27] first design partner with uh with OpenAI [13:29] and Codex. [13:31] They gave us some FTEs. And then I to [13:33] this date [13:35] um every um I think it's every two or [13:37] three weeks now [13:39] uh I sit down with my entire leadership [13:41] team of engineering and then their FTEs [13:43] and we talk about like what went right [13:45] and what went wrong in the past 2 weeks. [13:47] And um it it was just very clear to [13:50] people that I'm just not going to let [13:52] this one go. And I think that's in large [13:55] companies, there's a few things where [13:57] you have to go top-down. Most things [13:59] it's going to be pretty hard to go [14:00] top-down, but a few things that you [14:01] think are religion, you have to go [14:03] top-down and you have to just make sure [14:04] that you're unrelenting on those and AI [14:06] was one of those. [14:07] >> We've all seen how AI is transformative [14:10] for software development, but Cisco does [14:12] a lot of hardware development. [14:14] >> Yep. [14:14] >> Are you seeing the same impact on the [14:17] hardware development side or is that [14:18] still a a in the in the future? [14:21] >> So, the Cisco does [14:23] software, hardware, and silicon. And [14:26] each one of them have a different kind [14:27] of adoption cycle. [14:29] So, on the hardware side, you're [14:31] starting to see um good kind of design [14:34] patterns that are starting to emerge [14:35] now. [14:36] Uh [14:37] but it's not quite as much as software. [14:40] Software, the clock speed is really [14:41] fast. [14:42] Silicon, [14:44] there's a level of intentionality of not [14:46] going all in [14:47] by the team because they're very [14:49] concerned about making sure that [14:51] the data and the IP [14:53] >> Yeah. [14:54] >> does not leave and then before you know [14:55] it someone else has replicated it. [14:57] >> Right. [14:57] >> You know, and so on the silicon side, we [14:59] had gone a little bit slower, but now um [15:02] I think with the models getting [15:04] better [15:05] um with uh open weights models being [15:07] there, having our own infrastructure, [15:09] all of that, I think it it's going to [15:10] change, but we we have to compress the [15:12] cycle time [15:14] for silicon. The the thing that's [15:15] interesting in our business is we are a [15:16] full stack, right? Like we build [15:18] silicon, we build photonics, we build [15:19] systems hardware, software, [15:21] the entire stack. Each one of the [15:24] elements and layers of the stack have a [15:25] different planning cycle. So, silicon is [15:27] a 5-year planning cycle. What I do [15:29] today, [15:30] uh I have to make sure that I know [15:32] what's going to happen for at-scale [15:33] production 5 years from now and then [15:35] make the appropriate investments today. [15:37] >> That's bananas. How can you possibly do [15:39] that in the AI? AI is only three 3 years [15:41] old. [15:41] >> Think about the entire [15:43] um [15:44] manufacturing cycle of not just the [15:46] silicon, but silicon to systems to SDKs [15:48] to taking it out to the market. [15:51] You have to make sure that you start [15:52] because in order to build the silicon, [15:53] you have to build the IP for the silicon [15:56] like serdes and all of those pieces. [15:57] Those take actually um [16:00] you have to build them up ahead of time. [16:01] So, like there's a sequencing that needs [16:04] to happen in silicon that cannot happen [16:05] within a matter of like, you know, you [16:07] can't just say go build a chip in 12 [16:09] months. [16:09] Uh you can build a chip, it just won't [16:11] be that good. Hardware tends to be um [16:14] you know, 18 to 24-month planning cycle, [16:16] maybe 30 months in some cases because of [16:18] supply chain. Because we Think about [16:20] we're doing this at scale [16:22] for very large volumes of um of units. [16:26] You go to software [16:27] an operating system, that is a 12-month [16:30] planning cycle, maybe 18-month planning [16:32] cycle. [16:33] Then you go into models and agents and [16:36] apps. Now you're into a 3-month planning [16:38] cycle, maybe a weekly planning cycle. [16:41] And we have to make sure that that [16:42] co-designed full stack actually has all [16:44] of those things working in harmony. [16:46] And if you can figure that out as a [16:48] company, you have a very strategic moat [16:52] in the sense that you've been able to [16:53] figure out all these different layers [16:55] that work well together [16:57] from an IP That's That's very hard for [16:58] other people to do and replicate. [17:00] >> Is it possible to speed up the cycles [17:02] now that AI is here? [17:04] >> Absolutely. All of them will get [17:05] compressed. But when I say planning [17:07] cycle, I don't just mean how long does [17:09] it take to build? It I mean from idea to [17:12] at-scale production with billions of [17:14] dollars in revenue. [17:15] >> Speaking of chips and [17:17] hardware and this whole [17:19] supply chain, I want to talk about the [17:20] data center build out because [17:22] there's this incredible historic event [17:24] happening. We're building more stuff [17:26] than we built in in a in a a generation [17:29] and you're at the center of it. Can you [17:31] talk about the role that Cisco is [17:33] playing in this huge data center build [17:35] out? [17:36] >> So, in data center build outs, by the [17:37] way, everyone knows the role of GPU [17:39] plays. [17:40] And what had happened was when you just [17:43] started [17:44] seeing these transformers come about, [17:46] these transformers would fit on a single [17:47] GPU. [17:48] And so, the the training runs would be [17:50] pretty simple and then [17:52] uh you had more parameters and those [17:54] GPUs needed to scale and so, you had [17:56] eight GPUs within a server and so, you [17:58] now went into a server, but those GPUs [18:01] were all interconnected within a server. [18:03] Then you had racks of servers. [18:06] GPUs got interconnected within a rack of [18:08] server working coherently. [18:10] Like one, you know, kind of giant [18:12] machine with um [18:14] things going. And then, racks started [18:17] scaling to rows [18:19] because you now had multiple rows of [18:21] racks [18:23] that needed to be networked together [18:24] within a large data center. [18:26] Um and that required what they call [18:29] scale out networking, not just scale up [18:31] networking for um [18:33] for for rack level scaling. [18:36] And now, what you have is because [18:39] there's such a huge power constraint and [18:41] power shortage, [18:42] you can't always bring all the power to [18:44] where the data center is. So, what you [18:46] do is you go to wherever the power is [18:47] available. [18:49] And so, you might have data centers that [18:50] are hundreds of kilometers apart [18:52] that need to then work [18:55] like one completely [18:57] um logical unit. [18:59] >> Oh. [19:00] Two different physical data centers [19:02] separated by hundreds of miles that need [19:04] to be networked together as if they were [19:06] in the same building. [19:06] >> As as if they were in the same building. [19:08] As if they were the same computer. [19:09] >> Wow. [19:10] >> That requires a whole different family [19:12] of networking called scale across [19:14] networking. [19:15] For within a data center, you might have [19:17] scale out networking. And for within a [19:19] rack, you have scale up networking. We [19:21] are largely focused on scale out and [19:23] scale across. [19:24] And we do not just the silicon and the [19:26] switch trays for connecting the GPUs [19:29] together, [19:30] but we also do the photonics, which is [19:33] the optics. We started our business [19:36] selling to hyperscalers 2 years ago for [19:39] AI workloads. And we thought we'd do [19:41] we'd give the market this tree of [19:44] guidance and said we'll do a billion [19:46] dollars in orders taken. [19:48] We ended up at 2.3 billion in year one. [19:51] >> Uh so the next year, we said, "Well, [19:52] let's double it. Instead of 2.3, let's [19:54] do [19:55] like 5 billion." Last year, which was a [19:58] Q4 that just ended a month ago, [20:00] uh we ended up at 9 billion. [20:02] >> Wow. [20:02] >> Both years you doubled the very [20:04] optimistic projections that you made. [20:06] >> Yeah. And so we basically went from zero [20:08] to 9.3 billion. [20:10] >> in a matter of 2 years. [20:12] >> Wow. [20:12] >> Right? And so that's the speed at which [20:14] this is moving. We still have so much [20:16] share to go out and capture. It's a [20:17] complicated business, 63 billion dollars [20:19] in revenue. [20:22] Hundreds of products [20:23] um that we have to kind of make sure [20:25] that we maneuver in the right way. [20:26] >> But the 9 billion out of the 63 billion [20:28] total is like quite material for a brand [20:31] new business line. [20:32] >> 9 billion in orders. That takes a while [20:33] to get to revenue, but yeah. [20:34] >> I see. Yeah. [20:35] >> Got it. [20:36] >> But but you're right. It's it's it's [20:37] starting to get to be meaningful. And [20:39] it's exciting because it's you're right [20:41] in the thick of things. [20:43] And the market is so huge. A a gigawatt [20:46] of capacity [20:48] would cost about anywhere between It [20:51] used to be 35 billion, now it's like [20:53] close to 50 billion dollars. [20:55] Uh is what a gigawatt of data center [20:57] capacity would would cost to build. And [20:59] about 10 to 15% of that tends to be [21:01] networking. [21:02] Largely it's in the GPU and [21:05] But there's a huge amount of complexity [21:07] in you know, you have to build out the [21:08] shell and you have to build out the [21:10] power permits for that. And [21:12] the you walk into a data center, it [21:15] impresses upon you on how complicated [21:17] when you make a query on your prompt to [21:20] say give me a pizza recipe, [21:22] what happens behind the scenes to get [21:24] that to you within a second or two? [21:27] >> Because the build out is so extreme, [21:29] some people [21:31] have wondered whether we're [21:32] overbuilding. Are we? [21:34] >> Categorically not. Let me tell you why. [21:37] And people try to equate this Firstly, [21:38] in AI, there are no priors. [21:40] So I do fundamentally believe that and [21:42] this is one of the big mistakes people [21:43] make is they keep going back to an [21:45] experience they've had. I think one of [21:46] the most important skills to learn [21:49] um in AI is unlearning because your past [21:52] experience can be a huge liability. But [21:55] people typically will it try to equate [21:57] this to the dot com boom days. Oh my [21:59] goodness, there was so much [22:00] infrastructure build out that happened. [22:02] What was happening during that time is [22:03] you were building out infrastructure and [22:05] you were hoping that demand would catch [22:07] up to the supply. Right now, no matter [22:10] how much infrastructure you're building, [22:12] it's getting consumed almost instantly. [22:14] So it's there's no catch up time like [22:16] the your supply constrained despite the [22:18] exponential growth that looks like a [22:20] vertical line. So I do feel like from [22:22] that perspective, I don't think that [22:24] there is a um [22:25] there's a huge um overbuild. In fact, I [22:28] think it's the other way around. [22:30] If sub 2% of the people are using agents [22:32] and these agents are so consumptive on [22:34] average an agent is 450% more [22:36] consumptive on network bandwidth than a [22:38] human would be for conducting the same [22:40] exact task. Now think about trillions of [22:43] agents. [22:43] >> Right. [22:44] >> Um and now think about most of your [22:45] tasks getting conducted by agents where [22:47] they're going to have to make sure that [22:48] they keep enriching context. You have a [22:50] skills file, you have a markdown file. [22:52] You know, you've got the G stack from [22:54] Gary and it's going to keep going back [22:55] and forth. Like there's going to be a [22:56] lot of back and forth that happens. [22:59] >> It's going to be a lot tokens. [23:00] >> Yep. [23:00] >> Yeah. Speaking of agents, I heard that [23:03] you've recently become a fan of using [23:05] your own personal agents. I'm curious [23:07] what is what is what is your usage of [23:08] agents look like? [23:10] >> So I've [23:11] been using it for a while, but Instinct [23:14] was the force multiplier for me and this [23:15] is just like 4 days ago. I I suck at [23:17] email. [23:18] Really suck at email. Like I I had [23:21] 171,762 [23:23] emails or something just like 5 days [23:25] ago. 146,000 of them were unread. [23:29] >> Okay. [23:29] >> I used Instinct. I [23:31] took [23:32] the leap of faith and said, "You know [23:33] what? I'm going to go give it my [23:35] personal email [23:36] access. [23:37] I have 200 emails." [23:39] >> Wow. It blasted through your entire [23:41] inbox. [23:41] >> It took care of everything. It told me [23:43] what to do. It actually moved things. It [23:45] archived them in the right way. If I [23:46] needed something, it would bring it [23:47] back. It actually summarized my news [23:49] news for me from all the sources that I [23:51] was getting. I I was feeling [23:54] even though I didn't really do email [23:55] that much, I there was a part of me [23:57] subconsciously that was feeling [23:59] overwhelmed. [24:01] And I just feel better that wow, I'm I'm [24:04] on top of that. Now all of a sudden I'm [24:05] checking email. I'm like, "Why am I [24:07] checking email?" [24:07] >> Mhm. [24:08] >> Because it's fun to check email when you [24:10] have three emails. [24:11] >> It really ties back to the the point [24:13] about the data center build out never [24:15] being sufficient cuz like think about [24:17] the tokens to go through 140,000 emails [24:19] and figure out what to do with them. [24:20] >> Totally. This is where I feel like I [24:21] just don't think [24:24] it is unfortunate in some way that the [24:27] the level of kind of [24:29] um exaggeration that happens on oh my [24:32] goodness, this is a bubble. Everything's [24:34] going like you might still have like two [24:36] conditions can hold true. [24:38] You will have a secular shift where this [24:40] is going to completely fundamentally [24:42] change the way that people live. And [24:44] that's going to require a lot of [24:45] infrastructure. And you will also have [24:46] some companies that are very frothy in [24:48] valuation. [24:49] Because from time to any any kind of [24:51] massive platform shift like this, [24:54] you will have a ton of experimentation [24:56] that goes on and only a few of those are [24:57] going to win. Not not all of them win. [25:00] And so, there will be some frothiness in [25:01] the valuation, but that does not [25:02] constitute my mind a bubble. A bubble is [25:05] if that secular shift itself is [25:06] questioned. [25:08] I think it's very hard to go out and [25:09] argue for that. [25:10] >> So, um it's been in the news recently [25:13] this um open AI agents went rogue, [25:16] compromised hugging face and and other [25:17] places. What are What are your thoughts [25:20] on what's happening with AI safety now? [25:22] And also, how does that play into like [25:24] deploying AI at a large like [25:27] mission-critical enterprise? [25:28] >> Yeah, I think it's a great question. One [25:30] of the things that [25:31] I would say right now is it is going to [25:33] be the defining um you know, kind of [25:37] topic of our time right now in the next [25:39] um the next few years because [25:42] the constraint and infrastructure was a [25:44] really big deal. [25:45] And we were talking about a lot about, [25:46] you know, is there compute available? [25:47] That's still a problem. [25:49] But a much much larger problem is [25:51] whether or not I can trust these systems [25:52] that I'm going to delegate all my work [25:54] to. [25:55] And all my personal life too. The [25:57] difference between trusted delegation [25:59] and untrusted delegation is the [26:00] difference between market leadership and [26:02] complete abject bankruptcy [26:04] in my mind. So, like I think we have to [26:06] make sure that we actually get through [26:07] that. [26:08] And that requires [26:10] a a bunch of things. And we are actually [26:13] Cisco is working a lot on making sure [26:15] that we can provide a whole agentic [26:17] security platform [26:19] that we'll we'll we'll we're making [26:21] available to the market that everything [26:23] from [26:24] having full visibility in what goes into [26:26] a model to identity for non-human [26:29] identities and inventorying what agents [26:32] are running within your environment [26:34] to actually doing validation on the [26:36] models itself, um [26:38] which includes algorithmic red teaming [26:40] to say, can you [26:42] uh jailbreak this model? And if you do [26:44] jailbreak it, and you ask to the [26:46] question that you didn't want it to um [26:48] have it have a response, or if you had [26:50] the agent conduct an action that you [26:52] didn't want it to conduct? How can you [26:53] intercept that and make sure that [26:55] dynamically at runtime you can put you [26:57] can enforce guardrails? [26:59] That entire apparatus we actually make [27:02] available including the security side [27:04] but also the observability side. [27:06] Um and we fuse it together and I think [27:08] that's it's one of the most important [27:11] issues of our times. [27:12] >> It's very cool that you're building it [27:13] not just internally but you're actually [27:15] going to build it as a product. [27:16] >> No, we have it as a product to make it [27:18] available in the market. [27:19] >> In fact, we have a product called the AI [27:20] defense which is the core foundation of [27:22] that that we launched a year and a half [27:23] ago. It's doing ex- exceptionally well [27:25] and um we just need to make sure that we [27:27] get to more and more of it. And that was [27:29] only done for the chat era. So, we did [27:31] that initially for the model and then we [27:33] extended that to agents. [27:35] And so now what we have is dynamic [27:37] monitoring of agents at all times and [27:40] you know, if you ask an agent for a [27:42] refund, it shouldn't be giving you three [27:44] times the amount that you paid for the [27:47] for the pair of shoes, you know, and [27:48] that's something that [27:50] happens all the time in the in the [27:52] retail industry. So, that's we acquired [27:53] a company called Galileo. [27:55] Um there were folks from DeepMind and [27:58] and Uber that um they came along and [28:00] then those folks have done a great job. [28:02] They're part of the Splunk team and [28:03] they've done a great job on that. [28:05] >> Can you take us back to your early [28:06] career and [28:09] help us understand how growing up where [28:11] you did and the early jobs that you [28:13] worked like led you to be in this unique [28:15] position to have such a transformative [28:17] impact now? [28:18] >> When I look back now, I was very [28:20] fortunate. At the time it didn't seem [28:23] like I was very fortunate. But my dad [28:25] was a high-stakes con man. [28:27] He was like Bernie Madoff. [28:29] Uh he would swindle people for money. [28:31] Uh at some point it got dangerous for me [28:34] to live in India um and there were [28:36] people who were looking to kidnap me. He [28:38] had a mistress. That mistress had [28:40] committed suicide. I didn't want to [28:42] stick around. [28:43] >> Oh my goodness. [28:44] >> You know, it was it was it could make a [28:45] movie out of that that that that part of [28:47] my life. And so what I decided to do and [28:50] my he was very abusive to my mother and [28:52] I was very close to my mother. And so [28:55] both of us escaped [28:57] and she decided to [28:59] um [29:00] you know, uh live in um a place that was [29:03] undisclosed and I said, "Let me just go [29:05] out and check out America and see if I [29:06] can check out some schools and all of [29:08] that." Came over here. [29:09] >> are you? [29:09] >> I was 19 at the time, 1991. Right? And [29:13] so [29:14] uh I came here [29:16] my uncle put me up for a while, waited [29:18] on tables at a fine dining establishment [29:20] called Sizzler. [29:21] Um average tip of a dollar. [29:24] And um you know [29:26] there were a few seminal things that [29:27] happened during that time that [29:30] really gave me a mental reframing [29:32] because I was a terrible student in [29:33] school. [29:34] Um I wasn't really that and I was a [29:36] black sheep of the family in the sense [29:38] that a lot of my cousins were all very [29:40] smart academically and I was not. And so [29:42] I always had this kind of inferiority [29:43] complex. I used to stutter and so I [29:46] would always stay quiet. Um I couldn't [29:48] say hello when I picked up the phone and [29:50] so there was all of these kind of things [29:52] that were happening. And then I started [29:53] waiting on tables and I'm like, "Boom. [29:56] You have to figure out a way to [29:57] entertain or you will not make money." [29:58] And then I came um you know, when I was [30:01] over here, I was in [30:02] undergrad. I came across this company [30:05] DocuLabs [30:07] and they wanted to give me an internship [30:08] because my professor said to me, "Hey, [30:10] you might want to get an internship [30:12] here." She was my mentor in school and [30:13] she was great. And so she she um made an [30:16] introduction and so I I got a job as an [30:18] intern. Like few months later, the guy [30:20] comes to me, the founder, and says, [30:22] "Well, we're going to my original [30:23] investor is going to uh [30:25] leave. We're going to buy him out." I'm [30:27] like, "Oh, I want in." [30:29] And they're like, "Well, each one of us [30:30] put uh putting in a quarter million [30:31] dollars. You can put in a quarter [30:33] million dollars and you can have in." I [30:35] went to one of my the people I knew and [30:38] uh he was um Uh, he was a banker and I [30:41] asked him, "I like I need a quarter [30:42] million dollars. [30:43] And he said, "Why do you need a quarter [30:44] million dollars?" I'm like, "I'm going [30:45] to buy this company." [30:47] He goes, "What are you going to do when [30:47] you buy the company?" I'm like, "I'm [30:48] going to run it." [30:50] Now, remember, I wasn't a founder of [30:51] this company. I was an intern. [30:52] >> You were an You were an intern. You were [30:54] still in college at the time. [30:55] >> I had just The first loan I took was [30:57] 10,000, and the second loan I took was [30:59] 250,000. But the first loan I took, I [31:01] was still in college. I was minimum [31:03] wage. [31:03] >> Yeah. [31:03] >> Right? And I said, "Give me [31:06] Give me some money, and I'll pay it back [31:07] to you." He said, "10% interest." I ran [31:10] that business for 17 years. And now it's [31:11] like a [31:12] sub $5 million company. [31:14] >> You stayed there for 17 years. [31:16] >> It was in some ways my biggest mistake [31:18] in my career. [31:20] Um [31:21] because [31:23] not because the company wasn't good, [31:25] it's because the business model did not [31:27] align [31:28] with what I wanted to do in life. And [31:30] what And sometimes, you know, like, our [31:33] CEO, Chuck, tells me, he's like, "Gitu, [31:34] once you actually go public, um you [31:37] can't alter your business model that [31:38] much. Your business model is your [31:39] business model. Takes it Takes a lot of [31:41] effort to alter your business model. [31:43] But when you're when you're in a [31:44] services business, this was kind of a [31:47] market research company in a services [31:48] business, [31:50] and I loved software, and I loved [31:53] Silicon Valley, and I used to come here, [31:55] and I had an apartment in Russian Hill [31:57] for a while, and I would come here, and [31:58] I'd just be enamored with the people in [32:00] Silicon Valley. And I'm like, "I want to [32:01] go work with these folks because the [32:03] decision they make in that start button [32:05] in Windows is going to be affecting [32:07] billions of people, and I want to be [32:09] that person, you know?" [32:11] Um but I I never really was getting that [32:15] same level of uh satisfaction from that. [32:17] But I stayed on there because of ego. [32:20] And this is the other lesson I'll give [32:21] to founders. I stayed on there because I [32:23] had this ego that I will never work for [32:24] someone else. [32:26] This is my own company. I'm my own boss. [32:28] I don't want to work for someone else. [32:30] Biggest mistake I made. [32:31] >> Mhm. [32:31] >> Because ego drove my decision, rather [32:34] than learning driving my decision. So, [32:36] finally, I decided to go work at EMC. [32:39] And the guy who hired me at EMC is still [32:41] my coach. Um [32:43] and we were having dinner. [32:45] >> EMC, huge company. [32:46] >> Huge company. [laughter] 60,000 people. [32:49] Right? I'm going from like [32:52] 25 people to 60,000 people. He's hiring [32:54] me as a CTO. [32:56] Um and we're having dinner. [32:58] Uh like this across the table and I'm [32:59] like his name is Rick Devine and I said, [33:01] Rick, uh you got to pay me more cuz, you [33:04] know, I've had 17 years of experience. [33:07] He said, G3, you don't have 17 years of [33:08] experience. You have 1 year of [33:10] experience 17 times over. [33:13] Come work for me. [33:14] And in the first year I'll give you 17 [33:16] years of experience. [33:17] I learned more from him in the first [33:19] year [33:20] than I did in the 17 years running my [33:22] own business. [33:23] It was [33:24] absolutely magical. I hated working for [33:26] him during that time. [33:28] But after we left and I went to Box, [33:32] there's this person that I is my [33:34] co-founder in life I think of. Like, you [33:36] know, she is [33:37] um she's my head of operations. I don't [33:39] take a job without her. [33:40] Her and I would always be like, what [33:41] would Rick do at this point? [33:43] And we kept asking ourselves that. And [33:45] so, [33:46] that was a great mentor to have. Um [33:50] then I went to Box. [33:51] Box, we took that business from 200 to [33:53] 800 million during my time there. I I [33:55] ran all product by the end of it. Then I [33:57] was going to go run advertising for a [33:59] fan company. [34:02] Not at all anything to do with Cisco. [34:05] I accepted the job, but the kind of [34:09] offer they were going to give to Jesse, [34:11] who was with me, was not the right offer [34:13] for her. And she was my co-founder in [34:15] life. Like, I was not going to go [34:17] anywhere without her because she stuck [34:18] with me the whole time. [34:20] Um and so, I'm like, okay, so I need to [34:22] figure out what to do. [34:24] Um [34:25] and literally happenstance, [34:28] Chuck calls me the next day. [34:31] And [34:31] >> says, hey, you should probably think of [34:32] coming here." And I told him, I'm like, [34:34] "Hey, look, I'm really I've really [34:35] accepted this job." [34:37] And he said to me, he's like, "That [34:39] company will do well with or without [34:40] you. But if you come to Cisco, the [34:43] trajectory will be shaped based on what [34:44] you do." And then it was like so old, he [34:46] gave us two offers within 9 days. So, [34:50] you knew that the company was willing to [34:51] work at a very fast clock speed. The [34:54] question I asked him was I'm like, "How [34:55] much freedom are you going to give me to [34:56] just be very product obsessed?" [34:58] Uh and he said, "Yeah, I would love you [35:00] to be product obsessed." [35:02] And we did that, and before you know it, [35:04] like, you know, Cisco um um became the [35:06] place where I think I got some of my [35:10] in my career, the largest amount of [35:12] growth I've had [35:14] is probably in the past 6 years. [35:16] >> What are some lessons that you'd share [35:17] with, [35:18] you know, someone who is like you at [35:21] 19 or so, perhaps even growing up in in [35:25] in in India or some place outside of of [35:27] the valley where they're not where [35:29] they're [35:29] they're looking in from the outside. [35:31] >> I think the the first lesson I would say [35:33] is um your upper limit is typically [35:36] bound by your own ambition [35:39] rather than [35:41] um [35:42] the um [35:43] any other kind of factor. [35:45] It's not generally your ability. I'm a [35:47] very averagely intelligent person. [35:50] Um I didn't have a fancy degree from a [35:52] fancy school. [35:54] Uh but I was hungrier than everyone [35:56] else. And so, [35:58] I still um I'm hungrier than most people [36:01] I work with. Not because they're not [36:02] hungry, but because I just feel like I [36:04] don't want to let them down, and so I'm [36:06] just going to keep keep going at it, and [36:07] there's no one who can tell me that I'm [36:09] not putting in [36:10] the time. [36:12] And that I think is the first thing is [36:14] put in the time. Think really big, and [36:16] have very bold [36:19] uh ambitions. [36:20] Because if your imagination like [36:22] imagination can tend to curtail us much [36:24] more than we give it credit for. [36:26] You know, if you have your goal that's [36:27] this big, chances are you'll never get [36:29] here, but if you have your goal that's [36:30] over here, and if you get here, that's [36:31] pretty good, you know. And so, that's a [36:33] pretty important [36:35] uh you know, dimension to keep in mind. [36:37] The second one is [36:40] business is a team sport. Take pride in [36:42] longevity of relationships [36:45] that you can build where you have [36:47] actually made other people successful [36:50] as a function of you. And that I think [36:52] will go a really long way. And if you [36:53] think that way from the time you're 19, [36:57] boy, the advantage that you have [36:59] is tremendously high, [37:01] you know. [37:02] Um and then the third area that I [37:04] usually tell founders about is [37:07] this notion of power of compounding. [37:09] Like, you just have to get 1.27% better [37:12] today than yesterday. [37:14] And within a year you're 100x better. [37:15] And if you do that for 10 years, [37:17] like [37:19] that's pretty good. In 10 years in the [37:20] grand scheme of things, [37:22] it feels like a long time when you're [37:24] 19. [37:25] Like Sam Altman had this blog he had [37:27] written once, "The days are long, but [37:29] the decades are short." I read that [37:31] every every year. It's an amazing blog [37:32] post. It is one of the best blog [37:34] Jeff Bezos will say like [37:37] the ability for someone for a founder to [37:39] think in a 20-year window [37:42] versus uh [37:44] 3-month window [37:46] changes the quality of decisions you [37:48] make materially. [37:50] And if you think about the the the [37:52] really [37:53] big entrepreneurs that we all respect, [37:55] think about Jensen, think about Elon [37:58] Musk. [37:59] Um none of them were accused of thinking [38:01] in very short-term windows. I mean, they [38:04] grinded it out for a long time. [38:06] The only difference is grind it out when [38:09] the business model actually makes sense, [38:11] but don't do it like what I did and [38:13] grind it out in a business which doesn't [38:16] actually fit your ambitions, because my [38:17] business, the one that I ran, was great [38:19] for for business. [38:21] It was terrible to go change the world. [38:24] I wanted to go change the world. You [38:26] know, that's the wrong business to be in [38:28] to go change the world. But, if you [38:29] wanted to be a good lifestyle business [38:31] and have it be a thing that provided [38:33] well for your family, [38:34] fantastic. Keep doing that. That's [38:36] great. And by the way, none of those are [38:38] wrong answers. [38:39] It just happens to be that the one that [38:41] I [38:42] my ambition did not align with the with [38:44] the model of the business. Which is the [38:46] other thing you have to keep in mind is [38:47] does your ambition align with what [38:49] you're actually doing right now? [38:51] Where you are going to get fulfilled [38:53] when you look back when you're 80 and [38:54] say, "You know what? [38:56] I feel very good that this is how I [38:58] spent my time." [39:00] I think that's a question we should keep [39:01] asking ourselves and [39:03] for the younger folks, I always tell [39:05] them time's going to pass by so fast. [39:08] I never thought I'd be 55. [39:10] You know, I never thought I'd be called [39:12] uncle. I never thought I would have a [39:13] kid. [39:15] And I never thought that she would say, [39:16] "Dad, you're old." [39:17] You know. [laughter] [39:18] >> And bald. [39:19] >> Uh and so like those things sometimes [39:22] give you a lot of humility that you have [39:24] to make sure that you don't take time [39:27] for granted and spend the time in the [39:29] places that actually really matter. [39:31] >> What advice would you have for [39:33] entrepreneurs or for people who are [39:35] coming into an organization like Cisco [39:36] and trying to transform it the way that [39:38] you did? [39:38] >> I'll give you the story. So, when I I I [39:40] hadn't gone back to India from the time [39:42] I [39:43] um [39:44] left India. In 1991, I left. My dad died [39:46] in 2004. I'd gone to see him for 48 [39:49] hours once before he died. But then I [39:52] hadn't gone back until 2017. So, for any [39:54] kind of practical purpose from 1991 to [39:57] 2017, I hadn't spent that much time in [39:59] India. I took my daughter there. We went [40:01] to the Taj Mahal. [40:02] And there's this tour guide. His name is [40:04] Raj. [40:05] He's showing us the Taj Mahal. The guy [40:07] seemed like a really good product. He [40:08] knew every single detail about his [40:09] product, which was the experience he was [40:11] selling about the Taj Mahal. [40:13] But then people would keep walking by [40:15] and he would actually bust out in [40:16] different languages, Mandarin and and [40:19] German and French and Spanish and so at [40:21] some point I'm like, "Dude, how many [40:23] languages do you speak?" [40:25] He gave me like some ridiculously large [40:26] number, like 12 or 14 languages or [40:28] something like that. [40:29] And so I said, "That's that's crazy." [40:32] He's like, "Well, you know, I want to [40:32] honor the people that come here [40:34] because [40:36] um they come here and I should not be [40:38] presumptuous enough to think that [40:39] they're going to speak my language. I [40:40] should speak their language." [40:42] I mean, that's very noble. [40:44] Clearly he's smart, dude. [40:45] Uh I'm thinking to myself, I'm like, [40:47] "This guy is smarter than anyone I used [40:49] to work at Boxed at the time, anyone [40:50] that I work with." [40:51] He's making probably $10 a day. [40:54] I'm sitting here in Silicon Valley. [40:56] Uh and why is that? Because I happen to [40:58] have a platform and he doesn't. Right? I [41:01] have a platform of education, I have a [41:02] platform of America, I have a platform [41:04] of tech, I have a platform that just [41:06] gives me the affordance of opportunity [41:09] that he didn't get. [41:11] And so, no matter how smart he was, no [41:13] matter how hard he's working, [41:15] he's having a hard time [41:18] breaking away from the pack. [41:20] If you happen to be afforded a platform, [41:23] my only ask of people is don't squander [41:25] it. [41:26] And [41:27] as a founder, um you know, like we can [41:30] all look at all the things as glass half [41:32] empty and these are the things that we [41:33] don't have. [41:35] But man, we are all living in one of the [41:37] most special times. And if you're 19 or [41:40] 20 [41:41] or in the earlier part of your career [41:44] in this moment in time during the [41:46] movement in AI, do you know how many [41:48] people in their 50s, how much money they [41:51] would pay to go back to being 19 if they [41:53] could? [41:55] This is a magical time to be alive in [41:56] tech. [41:58] This is a great time to actually make an [42:00] impact on the world. [42:02] And if you are spending your time [42:06] associated with YC, which is one of the [42:08] finest organizations I think, which is [42:09] why I love [42:11] working with you folks, You would be [42:12] nuts not to just bust your tail and do [42:16] everything you can [42:17] not for any other reason, not for making [42:19] the money, but because the impact you'll [42:21] have on society at large [42:23] will be so gratifying when you're older [42:27] that it'll be something that's worth [42:28] doing. And if you happen to sell your [42:31] company to a large company [42:33] don't get cynical going to a large [42:34] company. Go learn things that the large [42:36] company teaches you because all those [42:39] large companies at some point in time [42:40] were startups. And if you are successful [42:43] you will eventually be a large company. [42:46] And [42:47] don't diss large companies. Actually [42:49] learn how to make large companies [42:50] operate with a level of, you know, kind [42:52] of chutzpah. [42:54] And then you'll you'll actually be able [42:56] to benefit from that as well. And go in [42:59] with an extreme level of curiosity and [43:01] regardless of which way you go, but but [43:02] know that you're afforded the platform. [43:04] And very few people in the world [43:07] look at the 8 billion people, very few [43:09] people are afforded a platform like we [43:10] have. [43:11] >> [music] [43:11] >> Gigi, that's such an inspiring note to [43:13] end on. Thank you so much for joining [43:14] us. [43:14] >> Thank you for having me. It was great.