Sep 28, 2026 · 45 min
Cisco’s AI inflection tests the durability of the infrastructure boom
What the AI Boom Looks Like From Inside Cisco with Sam Badri | The Real Eisman Playbook Ep 77
Cisco’s accelerating growth offers a view into whether hyperscaler spending, networking upgrades, and AI adoption can support a lasting investment cycle.
- 1Cisco’s revenue growth has accelerated to roughly 18% as hyperscaler demand and campus networking refreshes converge.
- 2Silicon One gives Cisco a proprietary networking advantage, while Arista benefits from greater exposure to hyperscaler customers.
- 3AI usage remains early for most people, even as power constraints and gigawatt-scale data centers raise the cycle’s costs.
Don't miss
Sami Badri contrasts today’s gigawatt-scale data centers with older 100-to-200-megawatt facilities, highlighting the engineering challenge of rising power density.
The brief
Steve Eisman speaks with Sami Badri, Cisco’s investor-relations chief, about the AI boom from inside the networking industry and the infrastructure connecting devices, data centers, and models.
Cisco’s growth has accelerated from normalized mid-single digits to roughly 18%, driven by hyperscaler demand, campus refreshes, and a surge in orders for its Silicon One technology.
The Cisco-Arista comparison turns on business mix as much as technology: Cisco has proprietary silicon and broader operations, while Arista is more concentrated in faster-growing hyperscalers.
Badri argues that AI adoption is still early: power users build repeatable workflows with skills files and agents, but most people remain at the simple-prompt stage.
The infrastructure opportunity comes with physical limits, as data centers move from 100-to-200-megawatt facilities toward gigawatt-scale projects with sharply higher rack power density.
The episode’s central tension is whether extraordinary hyperscaler spending reflects durable demand or an expensive bet whose productivity gains have yet to reach most users.
What was said on this episode
26 statements · 13 positive · 3 negative · 4 mixed · 6 neutral
Cisco’s majority business consists of invisible network connectivity infrastructure.
“the majority of our business is stuff that people do not see.”
Listen at 4:23
Cisco’s specific Silicon One chip is new, but its silicon program has existed about ten years.
“This specific chip is very new. This was announced in February of 2026, but the actual silicon program is not new at Cisco. It's been around for about 10 years.”
Listen at 6:19
Cisco’s hyperscaler business is expected to nearly double in fiscal 2027.
“Almost doubling in FY27.”
Listen at 9:22
Cisco’s hyperscaler business remains below 20% of company revenue.
“It's still below, you know, below 20%.”
Listen at 9:31
Cisco’s campus networking opportunity is also affected by AI dynamics.
“the campus opportunity does draw in some of the AI dynamics as well.”
Listen at 13:03
Larger Silicon One system orders marked Cisco’s AI-relevance inflection.
“As we started seeing product orders come in and much bigger size than we historically have seen them come in, that really became the AI relevance factor of Cisco.”
Listen at 14:05
Cisco uses proprietary Silicon One in switches, unlike Arista’s typical merchant-silicon approach.
“we have our own silicon so silicon one goes into our switches and form factors whereas for rista they typically are buying someone else a silicon and putting it into their switches”
Listen at 15:23
Arista’s revenue mix is more heavily weighted toward hyperscalers than Cisco’s.
“their revenue mix is much more indexed and weighted towards hyperscalers as an entire company.”
Listen at 16:03
Demand for frontier-model tokens is accelerating broadly across consumers.
“You're seeing a very widespread collective demand acceleration for tokens”
Listen at 19:50
Token consumption growth indicates high-quality AI demand.
“The quality of the demand vector seems to be very positive, which is token consumption or token growth.”
Listen at 21:24
About 90% of AI-tool users remain at the beginner stage.
“The fair majority of people that use these tools today are in the beginner stage. It's probably 90% of the users”
Listen at 24:44
A skills file steers an LLM through stock research and analysis.
“It's literally a Word doc of instructions to basically steer the LLM and the harness on how to research and how to analyze a stock.”
Listen at 25:48
Using a skills file accelerates Badri’s research productivity.
“And it just accelerates my productivity.”
Listen at 27:04
Some work processes will not require Anthropic or OpenAI frontier models.
“there will be work processes that do not need the latest and greatest from Anthropic and OpenAI”
Listen at 27:24
Demanding teams will pay more for frontier-model intelligence to improve productivity.
“You want best in class intelligence to accelerate productivity. And for that, you will pay up.”
Listen at 28:07
Hyperscalers’ heavy capital spending reflects belief in a significant future opportunity.
“if these companies are willing to do this, they must really believe in, you know, the journey or the destination they're solving for.”
Listen at 30:49
Texas data-center construction now faces pushback and grid-allocation audits.
“There has been pushback. There is now grid allocation audits.”
Listen at 32:15
Today’s large data centers operate at roughly ten to twenty times prior power scale.
“Today, you're talking about gigawatts. So we're talking about 10X, 20X, the size.”
Listen at 35:38
New data centers are only several times larger physically, but have much higher power density.
“They're probably doubling, tripling, or 4 or 5x-ing in size. But the difference is the electrical density, the power density in the facilities are really going up.”
Listen at 35:56
Data-center rack power roadmaps reach 150–650 kilowatts and potentially higher.
“Today, you have roadmaps that are talking about 150 to 650 and eventually megawatts at the rack level.”
Listen at 36:13
AI adoption remains early when measured by token consumption.
“you need to believe that we are still early from a token consumption perspective.”
Listen at 40:50
Broad professional adoption will require everyone to learn AI tools.
“And everyone's going to have to learn these tools.”
Listen at 42:37
Effective AI-tool use will generate substantially more token consumption.
“if you are going to use these tools and you're going to use them in the right way, that equals a lot more token usage.”
Listen at 42:41
The durability of AI growth remains uncertain.
“The question is, how durable is it?”
Listen at 43:41
AI growth is durable as long as token growth continues increasing.
“From Sam's perspective, it's very durable as long as token growth keeps going up.”
Listen at 43:46
A price war among Anthropic, OpenAI, and Chinese models could threaten the AI story.
“The risks really are, will there ever be a price war between anthropic open AI and the Chinese models?”
Listen at 44:18
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.