A commonly cited maxim in investing is: “A
good jockey can’t guarantee you a win, but a bad jockey can certainly lose you
the race.” The point is clear: whether in horse racing, venture capital, or
public markets, the individual driving the process often matters more than the
horse they ride. Venture capitalist Arthur Rock captured it well when he said, “I
invest in people, not ideas.”
This same principle applies directly to portfolio
management. While investment strategy and process are essential, at the end of
the day, investors are placing their trust in the manager—the jockey—to execute
with discipline, judgment, and adaptability. A sound process means little if it
isn’t adhered to under pressure.
The data underscores how rare outperformance
really is. In 2025, German researchers analyzed over 3,000 hedge funds and
found that only 6.6% outperformed the market over the 10-year period
from February 2014 through January 2024. Yet, there is encouraging evidence:
smaller funds—particularly those managing under $500 million—tend to deliver
stronger risk-adjusted returns than their larger counterparts, especially in
capacity-constrained strategies. Resonanz Capital, a German investment advisory
firm, has shown that as funds scale, returns often diminish, not because skill
disappears, but because size can be an anchor.
Outperformance is possible—but difficult to
identify in advance and even harder to sustain. This is why betting on the
manager is paramount. If you believe the manager has both the discipline to
adhere to a proven process and the flexibility to adapt as conditions shift,
that conviction outweighs the noise of short-term volatility.
At WCP, LP, we have nearly 18 years of evidence
demonstrating how our approach has navigated multiple market cycles. A central
hallmark of our strategy has been avoiding catastrophic downside—an
element we consider more important than chasing every rally. We do not claim
perfection; no jockey wins every race. But when conditions turn, when
volatility spikes, and when capital preservation matters most, our partners can
be confident they have a seasoned, battle-tested manager at the reins—one whose
sole focus is protecting and compounding their capital over the long run.
Over the past few quarters I’ve become
increasingly persuaded that efficient inference compute is where the capital
and power in AI will concentrate. As Jonathan Ross, founder of Groq,
explained in a recent 20VC interview: “The limiting factor for AI isn’t the
number of chips we can design — it’s the power envelope. Energy is the scarce
resource that determines how much inference you can actually serve.” This
isn’t a theoretical point; we’re already seeing AI labs throttle user access
and rate-limit products, not because of a lack of models or data, but because
they cannot serve additional inference at today’s power and compute
constraints.
Ross makes the case that this scarcity will
eventually force leading AI labs to control their own destiny. “If you’re
OpenAI or Anthropic, you can’t rely forever on NVIDIA’s allocation. To control
your destiny, you need your own silicon.” The logic is straightforward:
when every marginal unit of compute translates into tokens served, revenue
generated, or users onboarded, no lab can afford to be dependent on an external
supplier’s allocation decisions. Vertical integration becomes less about
beating NVIDIA on raw performance and more about securing supply, controlling
costs, and de-risking future scale.
Inference demand is accelerating at a breakneck
pace, fueling both the expansion of data centers and the evolution of
chip design to keep up with artificial intelligence-native applications and
agentic models. According to Andrew Feldman, founder and chief executive
officer of Cerebras Systems Inc., the demand for AI compute continues to grow,
signaling this is not a bubble but a sustained trend. The rush to meet this
demand highlights a new phase in computing, where speed and efficiency have
become non-negotiable. “The willingness of people to wait 10, 20, 30
seconds, a minute, three minutes, nobody wants to sit and watch the little dial
spin while they achieve nothing, Feldman said. By being able to deliver
inference at these extraordinary rates, we found extraordinary demand.” In
other words, inference isn’t discretionary — it is the foundation of user
experience, and demand accelerates in lockstep with responsiveness.
Perhaps the most important shift is recognizing
what the true constraint will be as AI scales. Ross is unequivocal: “In
three to four years, power — not chips — will be the limiting factor.” This
reframes the competitive landscape: the winners will not simply be those with
the best algorithms or the largest model checkpoints, but those who can deliver
the most inference per watt, per dollar, and within the tightest latency
budgets. That requires innovation at the system level — in energy delivery,
cooling, memory bandwidth, and interconnects.
And while much of the public debate fixates on AI
eliminating jobs, Ross argues that perspective is shortsighted. Scaling AI
requires massive infrastructure build-out — data centers, power systems, chip
fabs, cooling technology — all of which are deeply labor-intensive. But more
importantly, AI will create efficiencies that multiply human productivity. He
stresses that in 100 years; we won’t even recognize the jobs that AI has
created. “A century ago, no one knew what a software engineer was, or that
someone could make a living as a social media influencer. A century from now,
the most common occupations will be things we can’t even name today.”
The takeaway is clear: compute supply is
effectively scarce, and the bottleneck is energy. But what makes this moment
extraordinary is the sheer velocity of demand. Every advance in model
capability — from GPT-3 to GPT-4 to today’s frontier models — has not dampened
appetite but multiplied it. Each step change drives orders of magnitude more
inference, not less, as new use cases open and adoption compounds. Feldman’s
point about users refusing to wait even seconds is not anecdote — it’s a
reminder that inference is elastic: the faster and cheaper it becomes, the more
the market consumes.
That insatiable demand means the pressure on
power, efficiency, and throughput will only intensify. We are not heading
toward equilibrium — we are heading into a world where the demand for
compute is breathtaking in scale, and where every marginal watt of energy
translates directly into incremental intelligence served. The winners will be
those who can bend this curve: extracting more inference from each joule, and
scaling systems at a pace that matches the appetite of the market.
At WCP, we intend to be hunting for the winners
at the forefront of this massive growth theme. The AI race will not be evenly
distributed — it is a winner-take-most environment where a handful of
companies will define the future and the rest will fall by the wayside. The
dispersion between outcomes will be extreme. In such a market, backing the
right jockey is just as critical as backing the right horse. Our job is to
allocate capital to the managers, founders, and platforms best equipped to
harness compute, power, and efficiency at scale — and to avoid those that
cannot. This is precisely the type of market we aim to exploit — one driven by
secular tailwinds, characterized by wide dispersion, and offering the potential
to compound capital through focused conviction in a select few leaders within
the theme. Just as in investing, in AI there will be no middle ground: there
will be leaders who compound and laggards who fade, and conviction in the right
jockey will make all the difference.
