Advice for AI Infrastructure Vendors (Especially Startups)

Enterprise customers should choose their technology vendors with care. That’s a maxim with which we can all presumably agree. Selecting the right vendor, offering the right technology for your needs as a customer, yields valuable benefits, including improved efficiencies, increased productivity, cost savings, and maybe even competitive advantage.

Conversely, if you select a vendor that offers the wrong product for your needs, you expose yourself to a range of negative consequences, such as operational inefficiencies, decreased productivity, hidden costs, and even competitive disadvantage. The latter accrues when your competitors make technology decisions better suited to their needs and to market dynamics.

But customers aren’t the only ones who need to tread warily. Just as enterprise customers must exercise discretion and perform due diligence when engaging with a vendor — ensuring that the company in question is solvent and viable as a business — I posit that vendors also need to consider the financial health of customers. While customers don’t want to get stuck with equipment from a vendor that is no longer in business, vendors have no desire to be stiffed by customers who aren’t in a position to pay their bills.

As the AI bubble inflates, expanding to a point where it will deflate elegantly or burst chaotically, vendors of technology infrastructure might wish to vet their prospective customers more rigorously. Not all customers are equal, and not all customers are worth the opportunity cost of meaningful engagement.

We need only look back at what happened when the dotcom bubble burst in the early years of this century (er, millennium).

Looking Back, Looking Forward

I invite those of you of a certain vintage (aging like fine wines, I’m sure) to cast your minds back to the halcyon days of the blitzkrieg Internet buildout. Much was seemingly at stake. Telecom carriers overbuilt fiber and long-haul capacity to satisfy optimistic projections that traffic volumes would double every 100 days. When the dodgiest of dotcoms met their inevitable demise and the fiber remained unlit, bandwidth prices collapsed like cheap umbrellas.

Further, you may recall that some equipment vendors partly financed their own demand through vendor financing with upstart competitive local exchange carrier (CLECs). When those carriers failed (is anybody under the age of 50 even familiar with the CLEC acronym?), vendors were left with excess inventory.

The telecom sector lost about two billion dollars in market value — perhaps that’s not much by AI’s distended standards — and massive job losses ensued. The sweeping layoffs were a shock to a tech industry that hadn’t experienced such hardship.

Multiple bankruptcies occurred among carriers and network operators. The names of those companies are lost to the mists of time and faulty memories, but let’s drag a few back to the service for commemorative purposes: fixed wireless providers Winstar and Teligent; digital subscriber line (DSL) providers NorthPoint (which closed its doors) and Covad (forced to restructure); Internet backbone and hosting companies Exodus and PSINet (assets picked over and bought cheaply by Cogent); Canadian wholesale telecommunications carrier 360Networks; and scores of others.

Perhaps the case that reverberated the loudest, providing fodder for countless business-school dissertations and the inspiration for cautionary books, was the stunning bankruptcy of WorldCom, compounded by the chicanery of fraudulent accounting.

The ramifications of those market ructions spread. As the dotcoms vanished into thin air and carriers and network operators sank downstream, vendors of network infrastructure were next to feel the pain.

Surviving the Storm

Lucent’s headcount fell from approximately 150,000 to 30,000 as its revenue plunged sharply. Nortel took an enormous loss in 2001, cut tens of thousands of jobs, and sold various assets. (Nortel never recovered, though self-destructive management admittedly played a role, eventually submitting to bankruptcy and shuttering permanently later in the decade.) JSD Uniphase saw its stock price fall 95% and shed most of its workforce as losses mounted. Corning’s stock also fell precipitously, and it dramatically slashed its payroll.

The mighty Cisco, which briefly attained status as the publicly listed company with the world’s highest valuation, incurred a massive inventory charge, laid off thousands of employees (something it had never done previously), and its market capitalization plummeted more than 85 percent. Marconi went through a debt-for-equity restructuring that nearly wiped out shareholders. Other long-established vendors also went through the proverbial ringer.

Startup equipment vendors arguably fared the worst. Consolidation was often the best of a poor range of available options. ONI Systems merged into Ciena; Tellium merged into Zhone; Corvis later bought Broadwing and took its name. Companies with names such as Sycamore, Sonus, Avici, Avanex, Oplink downsized precipitously as their valuations plunged as much as 90 percent. Optical component startups were acquired at deep discounts or shut down.

What happened, as we can see in glorious hindsight, is that nosediving demand had calamitous repercussions for suppliers. The vendors who endured, and eventually prospered, were the ones who retained cash and had the least debt, such as Cisco, Corning, and Ciena. Highly leveraged vendors suffered dire consequences. Even for the survivors, the recovery was painfully slow.

Preparation is Salvation

Okay, let’s see whether we can learn anything from that bubble of the past.

Here’s what I think we can learn: Vendors can hope for the best, but hope is not a strategy. What vendors should do is prepare for a range of potential outcomes, including the worst possible scenarios.

This is why AI infrastructure vendors today should target prospective customers carefully, or, at minimum, treat financially suspect customers differently from how they would deal with, say, hyperscalers, who can be counted upon (one would hope) to pay their bills. As an equipment vendor, you might not want to extend generous terms to a neocloud that is living a precarious business existence.

At this point, I return again to the recent Gartner estimate that more than 80% of neoclouds will cease independent operations by 2030. Some neoclouds will be acquired by or merge with other companies while others will fail outright, according to Gartner. As interest rates rise, credit tightens, and loan scrutiny intensifies, Gartner’s prognostication does not seem fanciful. Nobody can foresee the precise contours of the future, but certain scenarios are more likely than others. Preparation is always the right prescription.

I don’t see how anybody can argue convincingly that we’re not in an AI bubble. The questions are when and how the bubble with deflate or burst, and how severe the damage will be for companies unprepared for the eventual pullback.

I often hear the argument that “this time it’s different.”

Indeed, the technology in question is necessarily different from those that preceded it, but that’s not the point. Technologies do not cause market bubbles to form or to deflate or burst.

What Vendors Can Do

Investors, animated by enthusiasm and exuberance, cause bubbles to develop. From one era to the next, technologies change, but investor behavior (or misbehavior) remains strikingly consistent in its proclivity toward periodic outbursts of irrationality and the suspension of critical judgment.

AI might well be a technology that has profound and longstanding utility and market value. Nonetheless, patterns of inveterate investor behavior suggest that it is no more immune to a market bubble than was the internet or, going further back in history, railroads or tulips (which, admittedly, are not a technology or even an artifact).

As I said in my previous post, we are the bubble. The technology is a vehicle, a focus for attention, but market bubbles are created by runaway investor sentiment.

Vendors need to protect themselves from potential market fallout. How exposed are you to highly leveraged necolouds? If you are counting on neoclouds for long-term revenue flow, start devising contingency plans that will mitigate your susceptibility to being swept away by a market deluge.

Diversify your customer base wherever possible, adjust payment terms where appropriate, reduce your own debt, and run your business with discipline and rigor rather than extravagance and incontinence.

If you can’t learn from others’ experiences, you’ll have to pay a greater price when experience pays you a visit.