Entire Industries Might Fall Victim to an AI Bubble Bust
The mad rush into AI systems is reminiscent of how Big Tech recklessly lurched into fiber optics networks 25 years ago and lost billions.
Before leaping into AI, we should take a lesson from the fiber optics financial debacle
The year was 2000. The Optical Fiber Communications (OFC) conference, held in Baltimore, MD, attracted a crowd of nearly 17,000 which overwhelmed the event registration staff and caused a line to wrap around the convention center twice. The show covered 121,000 square feet of exhibit space.
The popularity of fiber optic technology was at its peak, promising a huge new optical market that could sustain long-term growth. Investors were throwing money at anything optical. Companies could not keep up with demand and construction crews were laying fiber optic cable as fast as the fiber could be drawn.
Next year’s OFC conference in Anaheim, CA, drew 38,000 attendees and 270,000 square feet of exhibit space, more than double the previous year. Upwards of $250 billion in junk bonds financed new fiber builds, and write-downs on fiber inventory and builds were becoming common as too many companies were trying to do the same thing with the same generation of technology. The fiber optic “bubble” was about to burst.
As fiber optic transmission speed and bandwidth expanded nearly ten-fold in less than three years, a severe shortcoming became evident. All of the miles of early generation fiber optic cable were performance-limited by distance (a condition known as dispersion). Long-distance networks were unable to handle the new, higher data transmission speeds and frequency bandwidth demands of an internet that was growing at 100 percent per year.
Dark Fiber Networks
At this point, it became clear that the fiber optic networks laid out in a fury over the preceding years were overbuilt, underused, and not forward compatible with advancements in telecommunications software technologies. In 2001, route maps showed that only 10 percent of existing fiber optic networks in Europe and the U.S. carried any signal at all and those that did were limited at around 622 Mb/s. That’s the entire network, not what people could expect at their router.
A few entrepreneurs made a fortune in optical by selling out early, but many companies suffered paper losses in the billions, including Ceina and Corning. Lucent became insolvent and was forced to merge with French Alcatel. Nortel went bankrupt and was liquidated. Creditors began to unload unused fiber networks for pennies on the dollar.
AI Enters “Bubble Territory”
The historical evolution of fiber optic technology and the destructive mania shown by early investors in networks that quickly became obsolete should be a lesson for those watching the rise of the AI “revolution.” There is clear value in increasing the ability of software to generate creative content, or to automate repetitive tasks, or to process verbal and written communication. There is, however, no value in the current investment and scaling of AI systems for most industries.
AI investments that cannot be monetized today become a sunk cost and a perpetual drag on future income. Companies spending billions on AI chips purchased at today’s peak pricing will find that price inevitably drops as new generations of chips are introduced that are cheaper and perform better. Massive AI investments made today without a strategy to monetize AI immediately will create expensive, obsolete systems in the same manner as was the mad rush to embrace fiber optic technology 25 years ago.
Who’s Actually Making Money in AI?
As a leader in AI processor hardware and software, NVIDA has been enjoying quite a spike in business due to its ever evolving graphics processor unit (GPU) chipsets and ongoing development of neural networks intended to replicate human thinking through what is called “machine learning.”
Year-over-year, NVIDIA stock has grown over 200 percent and its market cap is an astounding $3.22 trillion as of this writing. The company’s 750,000 square foot headquarters in Santa Clara is the embodiment of a hard charging, dominant player in the AI and computer gaming business space.
It looks like NVIDIA figured out how to monetize AI as one of the market’s first movers. It’s new Blackwell B200 GPU chip is expected to sell at around $40,000 apiece. The company is poised to sell AI systems to data-driven Google, Microsoft, and Amazon. Those systems will become obsolete in around 24 months, then NVIDIA sells them newer, faster replacement systems. Over and over again. Good for NVIDIA.
Everyone else needs to tread carefully. The purchase of expensive AI systems has to fall in tandem with a solid strategy to make money today, not in two years, not in six months. Today.
It’s a common predicament we find ourselves in with just about all advanced technology (and in the computer world in particular). You pay today’s price for a given level of performance but tomorrow’s solution is almost certain to provide better performance for less cost, thus if you build out something that is going to take a year or so to produce revenue you will likely get crushed by your competition who builds out later when they have a revenue stream that can be generated immediately using better, cheaper technology.
AI is no different, and in fact, it might be worse because there is effectively no revenue model that exists for AI today in most applications. It is highly speculative, based on the assumption that customers will show up and be willing to pay for AI processing in the future.
The Takeaway
The most important lesson from the aforementioned fiber optics mania is that the bigger the profit potential, the less critical judgment for investment is used. Fear of missing out, or FOMO, is a cruel motivator.
Today, AI is an expensive luxury technology intended for deep pocketed corporations that have the financial wherewithal to play in that rarefied air. Someday, of course, the revenue models for profitable AI systems will appear. Smart business leaders should hold off on such investments until a time arrives at which they can generate revenue right away.
Otherwise, they’ll find themselves in a similar – or worse – predicament to those fiber optics investors in 2001. FOMO!