The article argues that what many thought was a decline of “Big Tech” was premature. Rather than dying out, leading technology companies have transformed themselves — evolving from traditional internet-era platforms into modern AI-first platforms. The narrative that “startups will take over” lost steam as companies like Google, Amazon, Meta Platforms and Microsoft doubled down, leveraging their massive existing assets to dominate the new landscape.
Traditionally, tech-platform dominance rested on three pillars — user base, data, and cloud infrastructure. In the AI era, though, those same pillars have morphed: data has become “training fuel,” users provide signals and feedback, and cloud infrastructure powers massive compute for AI model training and deployment. Put simply: the underlying architecture of value creation has shifted from “content and interaction” to “intelligence and compute.” This shift gives enormous advantage to those who already have scale — making the biggest players more entrenched, not less.
The essay highlights that AI is no longer just about neat software or gimmicky features — it’s an infrastructure business. Running cutting-edge AI requires massive capital expenditure: hyperscale data centers, high-density compute clusters, memory-heavy architectures, energy, cooling — the works. These requirements make it infeasible for most small players to compete at the frontier. In effect, the “arms race” for AI infrastructure has turned into a moat: whoever controls compute, data, and deployment capability, controls the AI landscape.
This transformation doesn’t necessarily kill innovation — but it changes how and where it happens. For smaller companies or startups, competing purely on features or ideas becomes harder. The leverage is now with firms that control infrastructure and massive data-scale. That said, smaller players might still find opportunity by focusing on niche verticals, building on top of big-tech AI platforms, or specializing in trust, compliance, domain-specific workflows or deep vertical integrations.
