AI Crosses Into Agentic Era — Compute Demand Just Jumped a Thousandfold

The Shift No One Is Pricing Correctly

Agentic AI isn't a feature update. It's a phase transition — and the compute infrastructure underneath global markets hasn't caught up.

Jordi Visser, former CIO at a multi-strategy fund and now head of AI macro at Nexus/22V Research, makes the case that late November 2024 marked the most consequential inflection point in recent financial history. Not because of a Fed meeting. Not because of a CPI print. Because the AI stack crossed from chatbot to agentic — and that crossing multiplied compute demand by a factor of a thousand.

The distinction matters more than most macro analysts are letting on.

---

Chatbot vs. Agentic: The Plumbing Just Changed

A chatbot waits for a human to ask a question. It answers. Done. The compute load is bounded by how many humans are typing.

An agentic system runs autonomously — browsing, executing, building, iterating — 24 hours a day, without a human in the loop. Scale that to billions of agents operating simultaneously, and the compute demand isn't ten times higher. It isn't a hundred times higher. Visser puts it at a thousandfold increase, and the infrastructure to support that doesn't exist yet.

"We have entered the agentic side, which means the amount of compute we need has gone up by a thousand times already from what we went through."

The supply-demand mismatch is already showing up in the data. DRAM prices are up 400-500% from September lows. Silver — a critical input in every semiconductor and drone component — is up 60% over the past six months. PMI readings have turned higher. These aren't coincidences. They're the early fingerprints of a compute buildout that's just getting started.

---

Key Details

• DRAM prices up 400-500% from September 2024 lows — direct signal of memory supply constraints hitting the agentic compute wave
• Silver up 60% over six months vs. gold up 20% — Visser's preferred commodity expression of the AI infrastructure thesis, citing silver's role in semiconductors and drone hardware
• Jensen Huang's GTC guidance projects Nvidia revenues approaching $1 trillion — roughly 2.5x current consensus estimates of ~$600 billion; the market isn't pricing the gap
• Micron (MU) trading at approximately 4x forward earnings — Visser's largest personal position, on the thesis that memory demand is structurally underpriced relative to agentic compute requirements
• CPI forecast: Visser expects year-over-year CPI to print above 4% within two months, citing an elevated import price inflation print as the leading signal — historically a headwind for S&P 500 multiples

---

The Mag 7 Split Nobody Is Making

The Mag 7 gets treated as a monolith. Visser argues it's actually two completely different businesses wearing the same index label.

Three hardware companies: Nvidia, Tesla, Apple. Four software companies: Google, Amazon, Meta, Microsoft. The software cohort faces an existential valuation problem — their seat-based, recurring revenue models were priced on the assumption that human headcount would keep growing. Agentic AI destroys that assumption. When agents replace the humans buying software seats, the discounted cash flow math breaks down. You can't model terminal value when you can't model the customer base.

The hardware companies face a different equation. Compute demand is structurally infinite relative to current supply. Nvidia's multiple is compressing while its revenue trajectory is accelerating — Visser reads that as a setup, not a warning.

"Nvidia's business is the key to this entire thing. There's no way that anything bad will happen until Nvidia is not a company that's going to be able to produce things."

---

The Labor Market Read

Visser isn't in the mass unemployment camp, and he's specific about why. The U.S. needs only 30,000–40,000 jobs per month to hold unemployment steady — a threshold that reflects demographic tightening and immigration constraints, not AI displacement. The unemployment rate sits roughly 80–90 basis points above its cycle low, and Visser sees that as attrition, not structural breakdown.

The real damage is psychological, not statistical. The corporate ladder is gone. Entry-level white-collar jobs — the ones that used to train the next generation of analysts, associates, and junior managers — are being compressed or eliminated. The unemployment rate doesn't capture that. The BLS doesn't have a metric for "career path destruction."

His read: healthcare, service, and trades workers face no displacement risk. Knowledge workers below mid-career face a structural renegotiation of what their labor is worth.

---

The Bottom Line

Watch the compute supply chain, not the AI narrative. DRAM pricing, silver demand, data center buildout velocity — these are the real-time indicators of whether the agentic transition is accelerating or stalling. If Visser's CPI call lands above 4% in the next two months, expect multiple compression across risk assets and a rotation window into the hardware layer.