What the AI Boom Looks Like From Inside Cisco with Sam Badri | The Real Eisman Playbook Ep 77

Summary

Steve Eisman and Cisco IR head Sam Badri discuss how the AI boom is showing up inside a legacy networking company. Badri says Cisco’s core business is connectivity hardware—switches, routers, Wi‑Fi, and its Silicon One chips—and that the company’s growth has accelerated from mid-single digits to roughly 18% recently, with FY27 guidance at 15% midpoint. The main drivers, he says, are hyperscaler networking demand and a large campus refresh cycle replacing old unsupported gear. The conversation also zooms out to the infrastructure layer: grid allocation audits in Texas, rising community resistance, and data-center power density moving from tens of kilowatts per rack toward hundreds of kilowatts and even megawatt-level designs. On the AI side, Badri argues adoption is still early, token consumption is the best demand signal, and most users have not moved beyond basic prompts into skills files or agents.

Notes

• Cisco’s core franchise is enterprise connectivity hardware: switches, routers, Wi‑Fi access points, and related networking infrastructure.
• Silicon One is described as a network processing unit, not a GPU.
• Cisco’s chip stack sits in the connectivity layer that moves information from point A to point B over light.
• The specific Silicon One chip is new, but Cisco’s silicon program has been in place for about 10 years.
• Badri frames Cisco’s pre-AI growth as normalized single digits to mid-single digits, followed by an inventory digestion period.
• Cisco’s most recent quarter was around 18% revenue growth.
• FY27 guidance is 15% revenue growth at the midpoint.
• Cisco’s hyperscaler business is roughly doubling in FY27 but still represents less than 20% of company revenue.
• Cisco sells hyperscalers two main products: switches and coherent pluggables for optical connectivity.
• Cisco’s campus refresh cycle is replacement demand for old deployed gear that is no longer supported.
• The AI inflection for Cisco came when orders for Silicon One systems began arriving in much larger sizes.
• Badri places Cisco’s AI inflection roughly six to nine months before the interview.
• Cisco and Arista mainly differ on silicon strategy: Cisco uses its own silicon, while Arista typically uses third-party chips and software.
• Arista’s faster growth is attributed to a revenue mix that is more concentrated in hyperscalers.
• Cisco still has sizable slower-growth businesses, including security, collaboration, and services.
• Badri says the AI demand story is being pulled by consumers of frontier models looking for tokens.
• He compares the market structure to a private Google Search, with a few extremely compelling model providers.
• Badri says the key demand vector he watches is token consumption or token growth.
• Most users, he says, have not progressed beyond simple prompts.
• Badri defines skills files as structured instructions for guiding an LLM through work like stock research.
• He says he built a skills file by uploading old research reports into an AI harness and turning them into editable instructions.
• Cisco uses both open-weight and frontier models internally.
• For intensive work, Badri says Cisco is willing to pay for frontier models because best-in-class intelligence speeds productivity.
• Badri argues hyperscaler capex is not new in kind, only larger in scale.
• Texas is no longer the easy data-center state it once was; power requests are now being audited.
• Data-center power requests go through a queue, and that queue is being audited for double-counted or speculative capacity.
• Badri says the backlog is multiples larger than what is actually under construction.
• Pushback includes sound and water concerns, and he cites a water-equivalence claim of 38,000 prompts per almond.
• Badri says community opposition to data centers is not new, citing years of experience covering Equinix and Digital Realty.
• What counted as a big data center in 2021 was 100 to 200 MW; today projects are discussed in gigawatts.
• Physical data-center size is rising only modestly, while electrical density is rising much faster.
• Older data centers ran around 35 to 40 kW per rack; current roadmaps point to 150 to 650 kW and eventually megawatt-level racks.
• 800-volt DC is part of the power-architecture discussion for higher rack density.
• Badri describes the field as peak electrical engineering.
• At Cisco IR, he says he had to learn what investors wanted, change disclosures, and recalibrate communication cadence.
• He describes his IR role as external representation and internal shareholder advocacy.
• Cisco says fiscal 2026 delivered record revenue, record operating margin, and record earnings per employee.
• Badri says AI is still early from a token-consumption perspective and that token growth data looks explosive.
• He says most people have not even reached the skills-file stage, let alone deployed AI agents into production.
• Software developers are the main power users today, with financial services next and broader sectors later.
• Eisman says Cisco’s revenue growth accelerated from 5% to 18%, with much of the gain tied to hyperscaler AI-related business.
• Eisman’s main durability concern is a price war among Anthropic, OpenAI, and Chinese models.