Creekstone is powering the race for AI dominance, securing the largest solar lease in Utah's history. The first phase will contribute over 1 GW of renewable energy in 2027.
For more details, you can read the full announcement here: https://t.co/MhPUHdO4Tp
He's my recent interview on Wired West discussing the expansion of nuclear power in Utah. The conversation focuses on how this initiative aims to support long-term growth in gigawatt-scale AI data centers. https://t.co/LVNS4l4nJr #AI#datacenter#nuclear
Is R Bigger Than G in the Post‑AI World?
In the late 1990s, as a young partner at Oak Hill Capital, I had the privilege of working with Myron Scholes. Scholes, already a Nobel laureate for his work on option pricing, joined our team after his time at Long Term Capital Management. One day, curious about the real frontier of his field, I asked him what he considered the greatest unanswered question in economics. Without hesitation, he replied: “I have no idea how to value equities.”
Stunned, I pressed him. His explanation was simple and disarming: “I don’t know if R is bigger than G.”
For readers who don’t live in valuation jargon, R is the discount rate—the rate used to convert future cash flows into present value. In practice, it’s often framed as a risk‑free rate plus an equity risk premium, and it embeds everything from inflation expectations to perceived uncertainty. G, by contrast, is the perpetual growth rate of those cash flows—dividends, buybacks, or distributable earnings—used in simple equity valuation frameworks such as the Gordon Growth Model.
The punchline of that model is not the formula; it’s the condition. When you write equity value as:
Value ≈ Cash Flow₁ / (R − G),
you’re forced to assume R > G. If perpetual growth equals or exceeds the discount rate, the denominator collapses toward zero and valuations explode—at least on paper. Scholes’ point wasn’t that the Gordon model is wrong; it was that the entire comfort of equity valuation depends on a gap that we treat as obvious, stable, and permanent. And he wasn’t sure it is.
That conversation took place in 1999, at the height of the dot‑com bubble. The market’s enthusiasm wasn’t only about new companies—it was about a new story: the internet would rewire the economy so profoundly that productivity and profits would accelerate structurally, pushing G closer to R. If technology could lift long‑run growth, then maybe traditional valuation anchors deserved to be reset.
The next decade didn’t resolve the question; it complicated it. In the 2010s, sovereign bond yields in several countries fell to (and in some cases below) zero, challenging conventional intuitions about discounting and the time value of money. Meanwhile, G proved stubbornly hard to pin down—buffeted by financial crises, policy shocks, globalization’s reversals, and geopolitical disruption. Both parameters that sit quietly in the denominator—R and G—looked less like constants and more like moving targets.
Fast‑forward to today, and we’re in a familiar moment of speculative intensity—this time centered on artificial intelligence. The hyperscalers—Amazon, Google, Microsoft, Meta, and others—are committing staggering capital to data centers, chips, power, and networking to fuel the AI build‑out. Investors are debating winners, moats, and timelines, but the most fundamental question often goes unasked:
Could AI systematically raise corporate cash flows while also compressing systemic risk—pushing G up and pulling R down at the same time?
AI and G: a plausible lift to growth
Start with G. AI is not just automation; it is a force multiplier for cognition, coordination, and experimentation. Across industries, AI can compress cycle times in product development, reduce error rates in operations, raise utilization of expensive assets, and accelerate discovery—especially when paired with abundant compute and proprietary data.
If adoption becomes broad and deep, it’s reasonable to expect a structural boost to earnings growth: not merely a one‑time level shift, but a sustained improvement in the rate at which firms convert inputs into valuable output. In other words, AI could make the “growth” term more durable than prior tech waves—less about one sector’s disruption and more about a general uplift in how the economy learns and optimizes.
But even here, the translation from productivity to equity cash flows is not automatic. Competition can pass gains to customers, labor can capture gains through wages, governments can capture gains through taxation, and dilution can spread gains across new capital. AI can raise productivity without guaranteeing that existing shareholders get the full benefit. Still, as a first approximation, AI creates a credible pathway to higher G.
AI and R: the under-discussed channel
The more interesting and less discussed possibility sits inside R.
Discount rates are not just about interest rates; they are about uncertainty. Part of what investors demand in expected return is compensation for risk: macro volatility, business fragility, credit stress, tail events, and plain old ignorance. If AI meaningfully reduces uncertainty by improving forecasting, detection, control, and response then it could compress the equity risk premium.
Consider a few mechanisms:
Better prediction and earlier intervention. AI systems can detect supply chain stress, fraud, operational failures, and credit deterioration earlier than traditional monitoring, reducing loss severity and tightening feedback loops.
Improved risk selection and pricing. In lending, insurance, and logistics, superior models can reduce defaults and mispricing, at least for those with access to the best data and compute.
Faster adaptation. Organizations that can re‑optimize quickly—pricing, inventory, capital allocation—may become more resilient to shocks, lowering perceived cash‑flow risk.
If the distribution of outcomes narrows, with fewer catastrophic misses, faster recovery, and lower volatility then investors may rationally accept a lower required return. Combine that with any regime in which risk‑free rates remain structurally moderate, and R could drift down.
That said, there is a credible counterpoint: AI may reduce some risks while creating others. Model monocultures can amplify systemic fragility. Cyber and adversarial threats become more potent. Concentration risk can rise as compute, data, and talent pool into fewer hands. Geopolitical competition around chips, energy, and strategic supply chains can raise tail risk. So the question is not “does AI reduce risk?” but “what is the net effect on the risk investors must be paid to bear?”
When the denominator gets dangerous
The truly destabilizing scenario is not “AI boosts growth.” Markets can handle growth. The destabilizing scenario is G rises while R falls, compressing the spread (R − G) across the market.
That’s when valuation frameworks become hypersensitive. Small changes in assumptions create enormous changes in implied value. This is the world where price moves feel detached from fundamentals because the denominator itself is unstable.
And then there is the most provocative possibility: AI as a self‑improving system. If, over time, models materially accelerate the pace of scientific and engineering progress—improving chips, algorithms, robotics, materials, and energy systems—then growth could become less bounded by human attention and human labor. In that world, the constraint shifts toward inputs like energy, compute, and physical capital and even those constraints could loosen with breakthroughs enabled by AI‑accelerated R&D.
A related twist is that future “users” of value‑creating systems may not be humans at all. We can imagine ecosystems where autonomous agents transact, negotiate, design, and purchase services on behalf of firms or individuals, generating measurable economic activity without a one‑to‑one relationship to human time and attention. If such agent economies scale, the linkage between population growth and economic growth could weaken further.
None of this is a prediction. It’s an exploration of a tail scenario where growth becomes structurally more “exponential” than our current intuition allows and where the old comfort of R > G feels less like a law of nature and more like an empirical observation from a pre‑AI era.
The sobering view
Skeptics will (correctly) point to history. The dot‑com era promised a permanent step‑change in productivity, and while the internet did transform the world, the path was uneven, the profits were concentrated, and exuberant valuations outran realized cash flows. AI could hit similar friction: regulation, litigation, public backlash, safety constraints, IP conflict, energy bottlenecks, and technical plateaus.
And even if AI lifts aggregate growth, it may not lift equity cash flows proportionally. A world of fast productivity growth can still produce mediocre equity returns if competition is brutal, capital requirements are high, or bargaining power shifts away from shareholders.
Still, Scholes’ remark from 1999 lands differently now. In a post‑AI world, the relationship between R and G is not just an academic curiosity; it’s the hinge on which valuation regimes swing. If AI pushes cash flows up and risk premia down at scale, and for long enough, then the foundational assumption that stabilizes equity valuation becomes less secure.
So the question remains, newly sharpened by today’s moment:
Is R still bigger than G?
The answer, whatever it is, may end up being the most important number in finance.
@elonmusk Not true. You are smarter than this. Radiative heat transfer is vastly less efficient than convective
For a 700W H100 GPU, you'd need roughly 1 square meter of radiator panel at 77°C (350K) to dissipate that heat. And that's just for ONE GPU - a typical data center has thousands
Excited to announce the closing of our Series B financing to bring 10 Gigawatts of Power to Utah! Sending deep grattitude to leadership from Trident Ridge, and new participation from Pelion Ventures! Check out the interview here:
https://t.co/BhxIVUtg9Y
I'm excited to share this interview about the groundbreaking Delta Gigasite, our effort to create the world's most powerful AI computing campus. #10gig https://t.co/yBjgTwwQjQ
SMR nukes are not economic. Just to cover the cost of capital at 5% rate of return, you'd need to have $0.30 per kWh, before labor and fuel. https://t.co/zhz3ySqGB5
Unlock the code of consciousness: How a stats revolution is reshaping our understanding of reality:
https://t.co/ph1EuwwOIm
(longer version at: https://t.co/3tEL3Fhueu )
@elonmusk@Space_Station Auction instead of de-orbit. Buyer pushes it to lunar orbit. 1 Starship or 2 Heavies payload carries enough fuel to make the orbital transfer. Generates $$$ and puts useful mass in a holding pattern for subsequent moon base. Eliminates polluting the air with vaporized metal bonus
New Attorney General on #Bitcoin : Congressman Matt Gaetz Introduces Bill to Allow Federal Income Tax Payments in Bitcoin - Bitcoin Magazine - Bitcoin News, Articles and Expert Insights https://t.co/G9FpitDu2L
@elonmusk@fema True. Volunteers on the ground are scrambling to move aid supplies from distribution centers set up at schools over to churches in hopes it will stop FEMA from confiscating the aid packages.
My 2 oldest kids went to college in New England, but my 2 youngest went to the South. This article capture the megatrend change: https://t.co/pcOICRo57W
I refuse to be a thespian in the Speaker’s failure theater.
The 6 month continuing resolution with the SAVE Act attached is an insult to Americans’ intelligence.
The CR doesn’t cut spending, and the shiny object attached to it will be dropped like a hot potato before passage.