2025 Year-In-Review: AI Bubble or Breakthrough?
From power-law breakouts to bubble concerns, my final post of 2025 takes a retrospective look at the most consequential AI debates of the year and examines what these signals may reveal for 2026.
AI hype continued in 2025, but the rising tide did not lift all ships equally, as this was a year dominated by AI power-law winners. “Power-law” (an extreme distribution where a few standouts account for the lion’s share of impact) is a familiar concept in VC. AI, however, has pushed this phenomenon into overdrive, with AI power-law leaders compounding performance, momentum, and valuations to truly historic levels.
AI is the ultimate kingmaker in the power-law game
We saw this dynamic play out in the public markets. AI beneficiaries decisively outperformed the broader market (chart above), but a closer look within the AI cohort reveals that mega-cap ($50B+) companies captured a disproportionate share of gains, and have in fact been doing so over the past three years post-ChatGPT:
As value consolidates around mega-cap hyperscalers, AI is not just reshaping global tech leadership, it is fueling the rise of the largest companies in history (exhibit below). With power-law dynamics deepening, concentration risk has surged: just ten mega-cap firms, representing 2% of S&P500 constituents, now account for ~40% of the index’s market capitalization, far exceeding Dot-Com era peaks.
Power-law effects are just as pronounced in the private markets, where AI is likewise anointing the world’s most valuable private companies. In the startup world, the cloud era’s “triple-triple-double-double-double” growth gold standard now feels lethargic compared to the blistering velocity of the fastest private AI companies surpassing $100MM ARR in under two years:
In pursuit of the fastest-growing AI startups, VC capital is also concentrating in power-law fashion, with nearly half of all venture dollars flowing to a handful of AI companies (chart below). Power-laws beget extremes, and the extreme is record-shattering in the AI super-cycle: OpenAI overtook SpaceX this year as the world’s most valuable startup at a $500B valuation, a number larger than most national economies and the highest ever for a private company.
As power-law dynamics become increasingly entrenched, top startups are able to attract outsized capital and stay private for longer. The result is a new class of $50B+ mega-cap private companies:
These companies are redefining the upper bound of scale in private markets and their expansion ambitions are far from complete. Mirroring the public hyperscaler playbook, AI giants in this cohort are compounding value and further cementing dominance through vertical integration across the stack. For example, in 2025 alone, OpenAI has made major moves toward:
Hardware layer through acquisitions such as Johnny Ive’s io. Even space has come up as a potential expansion vector!
Application layer through acquisitions such as Sky and Statsig (where the founder was instituted as OpenAI’s new CTO of applications).
Compute layer through co-development partnerships with Broadcom and investments in Stargate.
Bubble trouble?
With power-law forces shaping the markets, generational behemoths are seemingly breaking away in this AI super-cycle. But meteoric hypergrowth attracts intense scrutiny, and 2025 delivered this in full force. Questions around AI ROI, margin profile, and growth durability permeated the conversation all year, magnified by the reality that AI startups are among the most capital-hungry businesses in history:
Correspondingly, chatter about an “AI bubble” found its footing in 2025:
The AI bubble debate is multi-faceted, so I’ll focus on three “hype vectors” that involve the AI power-law players:
I. AI infrastructure investment: As Big Tech continues to invest record levels of capex into AI infrastructure (chart below), several key questions have been raised around useful life estimates, fungibility of this capex, and potential risk of overbuilding if revenue from “pent-up” demand fails to keep pace or if a glut ensues.
Complicating matters further is the emergence of “circular deals” involving creative financing structures from Big Tech giants, which injects additional risk into the ecosystem and raises questions about quality of revenue:
As I explored in my deep dive on the AI infrastructure buildout debate, I believe that demand is likely to continue outpacing capacity in the near term, since the inference market could be larger than any of us may expect. This naturally leads to the next discussion point on AI adoption.
II. AI adoption: As a B2B VC, my focus on this vector of the bubble debate has centered on enterprise AI rather than consumer AI (yes, we all know that ChatGPT is the fastest-growing consumer app in history). Opinions here are sharply divided. Some argue that the pace of enterprise adoption is plateauing (chart below), while others see it accelerating. Forecasts for AI’s impact on productivity and the economy also vary widely. Where you land on this debate largely depends on your fundamental view of total addressable use cases: is AI a one-time innovation shock confined to marginal co-pilot roles, or will autonomous AI agents reshape the landscape? I fall squarely in the latter camp, recognizing that it will take time for an AI-driven future of work to unfold.
III. AI valuations and fundraising: Focusing on the private markets, AI rounds are getting bigger and happening faster than ever. Valuations are now reaching levels that look astonishing on both an absolute and multiples basis (chart below). Against this frothy backdrop, questions about a valuation bubble have been swirling. I dove into this topic with Carta’s Peter Walker on The Data Minute podcast. While a reckoning (or AI wildfire as some call it) may be on the horizon to clear out noise in an overheated market, the funding environment could remain exuberant in the months ahead due to the intense fear of missing out on the next generational company, particularly given the extreme power-law dynamics at play.
So, bubble or breakthrough?
Heading into 2026, we’re likely straddling both. This tension is exactly what you’d expect as we experience one of the most consequential tech cycles in history, where a rapidly evolving landscape creates unprecedented opportunity alongside staggering Cambrian volatility.
The AI race is likely to get hotter in 2026. Even as power-law leaders pull away from the pack, competition is intensifying rather than abating. Staying at the top has become an obsession for leadership teams, with speed of execution serving as a critical moat. AI power-law leaders are now setting a breakneck pace for the industry, with iteration cycles measured in days, not quarters, and relentless leapfrogging driving the frontier race in AI.
Wishing everyone a very happy new year and be sure to tune in for my next post with 2026 predictions!