For anyone who walked the corridors of Silicon Valley in the late 1990s, today’s unprecedented artificial intelligence boom carries a profound sense of historical déjà vu. Back in 1998 and 1999, internet traffic was roughly doubling every six to nine months—an extraordinary rate of growth in digital demand that felt like the absolute outer limit of technological scaling. And yet, that frantic pace of expansion pales in comparison to the seismic shifts currently unfolding across the global technology landscape.
The modern landscape is defined by astronomical investments that dwarf previous tech cycles. Today, the four largest cloud computing providers in the United States are looking at more than $700 billion in capital spending for the year alone, while cumulative infrastructure spending dedicated to artificial intelligence is projected to skyrocket to between $3 trillion and $4 trillion by 2030.
Nvidia arguably fired the starting gun for the modern AI hardware revolution when CEO Jensen Huang hand-delivered the company’s pioneering DGX-1 supercomputer system to OpenAI in 2016. In the years since, Nvidia has transformed from a specialized graphics card manufacturer into a household name and an indomitable force deeply embedded at every layer of the AI ecosystem, operating with near-exclusive supply chains and high-profile investment partnerships.
With such a massive market dominated by a seeming monopoly, a critical question emerges for the broader technology sector: Where can genuine market competition—and the rapid innovation that inevitably follows it—possibly come from?
Part of the answer lies in the structural reality that corporate scale inherently costs speed, though when going up against an execution machine as powerful as Nvidia, navigating that disadvantage is simply the baseline hurdle startups must solve for. Moreover, the pace of technological change in artificial intelligence vastly outruns anything witnessed in previous tech cycles. Model generations now turn over in a matter of weeks rather than years, forcing the underlying hardware to constantly shift and adapt to serve entirely new computational paradigms. Designing from first principles under these conditions carries real merit and opens narrow windows for agile challengers.
Lessons from Davids That Came Before
Hardware suppliers and infrastructure pioneers of the dot-com era traversed very similar peaks and troughs, providing a blueprint for modern upstarts. Cisco Systems, founded in 1984, dominated internet routing for the vast majority of the decade. By March 2000, it stood as the highest market capitalization company in the world, appearing entirely untouchable.
Then came Juniper Networks, a specialized startup where R K Anand served as a founding engineer. Juniper shipped its first core router in 1998, transforming rapidly from a relatively anonymous upstart into a major market force, capturing 17% of the core router market by the end of 1999, and surging to 38% by early 2001.
Today’s market challengers can look to that history for strategic guidance. The primary rule of engagement is simple: Do not build a slightly cheaper version of the incumbent’s established product. Instead, build a far better one. At Juniper, the foundational bet was that the adolescent version of the internet was going to demand exponentially higher throughput than existing technology could provide. By introducing an entirely new architectural approach to how network routers were built, coupled with an innovative design philosophy laser-focused on raw performance and reliability, the company did not set out merely to sell a better router. Rather, it redefined what a core router had to be.
The modern boom in AI inference represents today’s version of that historical strain on data throughput. The initial hyperfixation on training massive AI models is steadily fading, and the industry is collectively pivoting toward putting those models to practical use, a phase known in the industry as inference. Inference is where the promised productivity gains of artificial intelligence actually materialize, putting trained models to work across everyday enterprise tasks, customer service operations, and automated workflows. As enterprise applications for artificial intelligence grow broader and increasingly complex, the underlying demand for specialized inference compute scales right along with them.
Google recently reported that its infrastructure now processes more than 3.2 quadrillion tokens a month, a staggering leap from 480 trillion tokens just a year earlier, as artificial intelligence transitions rapidly from basic chat interfaces to autonomous agents that can plan and execute complex tasks on behalf of users. Serving tokens at such massive scale is a fundamentally different engineering challenge than model training, introducing distinct operational constraints regarding memory bandwidth, electrical power consumption, and total cost.
The core strategic principle remains universally applicable across technology sectors: Challengers have a far better chance of succeeding when they optimize for the market that is about to arrive, rather than the market that is already comfortably well-serviced by the incumbent.
Be Known by Your Track Record
Market challengers invariably begin their journey without an incumbent’s most formidable asset: an established reputation. Purchasing enterprise hardware or software from the undisputed market leader is inherently viewed as the safe choice, meaning a new startup must actively replace institutional trust with undeniable empirical evidence. Every technological edge a young company can demonstrate, validate, and prove in real-world deployments makes the next customer conversation progressively easier. Over time, those validated successes compound into the kind of hard-earned credibility that incumbents often take for granted.
To accelerate this process, new market entrants must make product migration as frictionless as possible. Enterprise customers are rarely wedded to a specific technology platform out of blind loyalty; rather, institutional inertia remains one of the greatest strengths of an incumbent. The sheer ease of continuing to do business the way it has always been done is a powerful deterrent to change.
Unless there are clear, quantifiable economic and operational benefits, along with stark scale and performance advantages, integrating entirely new systems will rarely seem worth the internal friction involved in switching suppliers. By removing every possible technical barrier from the equation, a challenger gives its superior product a genuine fighting chance in the marketplace.
Uncertainty Is Your Friend
The internet ultimately transformed global society in ways that contemporary observers could scarcely have predicted during the late 1990s. There is little doubt that the relentless pace and momentum of the artificial intelligence revolution are producing structural changes that will quickly make modern work and daily life unrecognizable compared to the recent past. That acceleration is only expected to increase. Monopolies in frontier technology often look completely insurmountable—right up until the nature of the underlying work shifts. What artificial intelligence actually does is rapidly morph and evolve, opening unexpected flank attacks for agile competitors.
That inherent volatility offers a source of comfort for startup founders, reinforced by the reality that small, focused teams can now achieve monumental engineering feats. Challenging Cisco at the turn of the millennium meant building a nimble and dynamic organization that was vastly smaller than the market leader—one that was inherently more capable, reactive, and product-focused. Juniper Networks numbered only a few hundred employees by the time it achieved significant market traction, standing in stark contrast to the thousands of workers employed by Cisco.
Today, modern challengers are mirroring that lean philosophy. Companies like Tensordyne operate with modest teams of around 120 people, attempting to accomplish feats that once required an army of corporate engineers. That operational agility and focused intensity are precisely what keep major incumbents watching from the sidelines.
R K Anand is co-founder and chief product officer of Tensordyne, an AI inference company. He was previously a founding engineer and executive vice president of the data center business unit at Juniper Networks and began his career designing microprocessors at Sun Microsystems.
