Cisco CEO on the AI ‘bubble’: why the crash can still leave winners

/ Technology / By

Summary: Cisco CEO Chuck Robbins says AI could be bigger than the internet, but he expects a painful shakeout first—“winners will emerge, and there’ll be carnage along the way.” That’s not a throwaway quote. Robbins lived through the dot‑com boom as Cisco became the most valuable company in the world in 2000, then lost roughly 80% of its value when the bubble burst.

The core message is nuanced: AI is real and transformative, but today’s investment cycle is overheated, and not every company (or job role) survives the transition.

What Robbins is actually saying (not the headline version)

From the report:

  • AI will “change everything” and may be bigger than the internet.
  • The current market is “probably” a bubble.
  • Some companies “won’t make it.”
  • Some jobs will change or be eliminated, especially in areas like customer service.
  • The bigger risk for workers is not “AI taking your job,” but “someone using AI well taking your job.”
  • AI will improve cyber attacks and scams; security becomes more important.
  • The UK has “pretty good odds” of becoming an AI superpower if it embraces AI.

The theme is not “panic.” It’s “embrace the shift, but be honest about disruption.”

Why the dot-com analogy is useful (and where it misleads)

The dot‑com bubble is often invoked lazily, as if it means “hype now, crash later.” The more useful lesson is structural:

What the bubble built

Even though many internet companies died, the era built:

  • data centres
  • fibre networks
  • software infrastructure
  • consumer behaviours around online services

The underlying technology won.

What the bubble destroyed

Speculative equity value in companies without durable products or distribution.

Robbins’ “carnage” framing is basically that: capital and companies get wiped out, but the platform shift still happens.

Cisco’s vantage point: infrastructure, not apps

Cisco is not primarily an “AI app” company. It sells and builds infrastructure that enables AI to run:

  • networking
  • security
  • data centre systems

So when Cisco talks about AI, it’s often speaking from the layer that survives bubbles.

App companies come and go; infrastructure and distribution winners often persist.

Why networking becomes a bottleneck in AI

Large models are trained across many accelerators. The more chips you use, the more your system becomes limited by:

  • network latency (how fast nodes can coordinate)
  • bandwidth (how much data can move)
  • reliability (a single failure can slow or interrupt training runs)

That’s why companies talk about “AI clusters” like they’re supercomputers. In that world, networking isn’t plumbing—it’s a competitive differentiator.

The jobs angle: what changes first

Robbins points to customer service as a category where companies may need fewer people.

Why customer service is the first target

Support workflows often have:

  • high volume
  • repeated questions
  • known policy rules
  • text-based inputs and outputs

That makes them a natural fit for AI triage and partial automation. The first wave is usually “deflection” (answers without a human), followed by “agent assist” (humans supported by AI), and then automation of the easiest end-to-end cases.

That’s plausible because customer support has many:

  • repetitive questions
  • standard workflows
  • text-based interactions

But the deeper point is: AI doesn’t replace “jobs.” It replaces tasks.

A typical job is a bundle of tasks:

  • some automatable (drafting, summarising, triage)
  • some not (judgement, empathy, accountability, negotiation)

So the workforce impact is uneven:

  • people who adapt become more productive and more valuable
  • people in roles dominated by repeatable tasks face displacement pressure

“Someone good at AI will take your job” — what to do with that

This line is uncomfortable because it’s true.

In practical terms, it suggests a survival strategy:

  • learn the tools early
  • build workflows and checklists
  • become the person who can combine AI speed with human judgement

The competitive advantage is not knowing prompts—it’s knowing:

  • what to ask
  • what to verify
  • what the output should look like
  • where the risk lives

Security: AI makes scams and attacks better

Robbins warns that AI will make cyber attacks better and phishing more convincing.

That’s already visible in:

  • better-written scam emails
  • more believable impersonation
  • deepfake audio/video used for fraud

So the “AI revolution” has a parallel revolution in:

  • identity verification
  • fraud detection
  • secure communications

Security is not a side topic. It’s one of the primary battlegrounds of AI adoption.

What does “AI bigger than the internet” actually mean?

If you strip away rhetoric, “bigger than the internet” could mean:

Another interpretation: the internet connected people and systems, but most work still required humans to translate information into action. AI reduces that translation cost. If that’s true, AI becomes a general productivity layer the way electricity became a general capability—visible in every sector, not just “tech.”

  • AI becomes the interface to information (search changes)
  • AI becomes the interface to work (agents in workflows)
  • AI becomes embedded in every product (from banking to healthcare)

The internet connected systems. AI changes what those systems can do.

So the claim is that AI isn’t merely a new app category; it’s a capability layer that rewrites software economics.

The UK as an AI superpower: conditions that matter

Robbins says the UK has “pretty good odds” if it embraces AI.

What “embrace” usually means in policy terms:

  • make it easy for responsible experimentation to happen in business and government
  • invest in skills so adoption isn’t limited to a small elite
  • fund research and commercialisation bridges (not just academic work)
  • maintain credible regulation that targets harms without freezing innovation

The UK has strengths (research, talent, finance, a strong startup culture) but also constraints (compute access and competition with the US/China for top labs). The likely path to “superpower” status is not beating the US and China at scale, but building competitive clusters and high-value specialisations.

In practice, “embrace AI” means:

  • research strength + talent pipelines
  • compute access (or partnerships)
  • supportive but realistic regulation
  • adoption in government and industry

Countries that adopt earlier may gain productivity advantages and attract investment.

What a “healthy” AI buildout looks like

A bubble narrative is compelling, but it’s also possible to have both hype and real progress at once.

A healthier buildout tends to show:

  • clear ROI use cases (cost reduction or revenue lift you can measure)
  • consistent deployment in workflows (not just demos)
  • improving safety practices (monitoring, evals, red-teaming)
  • consolidation around standards and platforms

That’s different from a frothy market where most value is in announcements and fundraising.

What to watch next (signals of a real bubble vs a healthy buildout)

If this is a bubble, you should expect:

  1. Crowding in apps
    Many similar products competing on thin differentiation.

  2. Margin pressure
    Companies spending heavily on compute without clear revenue payback.

  3. Consolidation
    Stronger players acquiring or outlasting weaker ones.

  4. Infrastructure winners
    Network, chip, cloud, and security providers benefiting regardless of which apps win.

  5. Regulatory shock
    A major incident (fraud, deepfakes, model misuse) can accelerate rules that change economics.

Practical advice: how to be on the winning side of the transition

For individuals:

  • learn how to use AI tools in your domain (not just generic prompting)
  • build verification habits (what do you trust, how do you check it?)
  • focus on tasks where humans are still accountable: judgement, relationships, strategy, safety

For organisations:

  • start with measurable use cases (support, analytics, code review)
  • invest in security and fraud defence early
  • treat AI as a process change, not a “tool rollout”

Bottom line

Robbins’ message is not anti-AI. It’s a realistic diagnosis of platform transitions:

  • the technology will reshape work and security
  • the investment cycle is overheated
  • many firms will fail

If you’re building or investing, the long-run winners will be those who turn AI into durable distribution, trusted products, and measurable value—not just bigger demos.

Bottom line (one sentence)

AI is a real platform shift, but the market is behaving like a bubble; the winners will be the companies that turn AI into trusted, measurable value while surviving the shakeout.

Sources

n English