PodcastThe Monetary Matters NetworkBen Pouladian

Why AMD Is the Most Overhyped Chip Stock | Ben Pouladian on Meta’s Muse, Anthropic IPO, & Semis

The discussion weighs the scale of AI compute demand against the practical limits of chip supply, data-center construction, and financing. Ben questions AMD’s valuation and execution while highlighting Nvidia’s integrated stack, the importance of inference, and risks around custom silicon. The speakers also point to AI adoption and revenue growth, while noting uncertainty in reported figures and the potential for hardware pricing power to fade as capacity expands.

SemiconductorsAI InfrastructureTechnology Stocks

Preview · 25 more points in the app

Read the Full Summary

Key Takeaways

Ben says local opposition has led some U.S. towns to halt approved data centers, while pointing to a Louisiana Meta project’s proposed $50,000 annual teacher bonuses and 300 training slots that drew 30,000 applicants as possible community benefits. He also argues that AI-related construction and U.S. manufacturing could create jobs and support consumer spending, including through suppliers and workers.

Ben links large compute commitments to continued demand for AI infrastructure: he cites Anthropic’s disclosed commitments of more than $500 billion over the next few years and says larger training runs have so far produced more capable models, though he expects the buildout eventually to plateau without giving a timeline. The speakers also cite rapid OpenAI and Anthropic revenue growth, but give unreconciled figures for their combined ARR, so the episode does not establish a single consistent total.

Ben sees inference as a key chip battleground because Nvidia has largely captured model training, and says companies and cloud providers are developing inference-focused chips. He argues Nvidia’s integrated hardware and software can bring optimization and security benefits, while Google’s ability to manage TPUs internally and Amazon’s reported chip-related operational difficulties illustrate the trade-offs of custom silicon; Nvidia GPUs also have more established financing pathways because they can be redeployed.

The episode describes uneven supply and pricing across semiconductors: the host cites TrendForce’s assessment that categories were tight to very tight except for balanced GPUs, and says Nvidia’s gross margin rose from 60% to 75% over five years. Ben cautions that hardware pricing power could fade as capacity expands, while also saying CPU growth may be constrained by the two fabs making the chips; the host characterizes Lam Research’s 18% one-year revenue-growth forecast as modest relative to the current quarter and a strong equipment cycle.

GEO-POLITICS

Data Center Pushback — Ben says some U.S. towns imposed moratoriums on approved data centers after local opposition. He cites a Louisiana Meta project as a possible source of community benefits: teachers were to receive $50,000 annual bonuses, and 30,000 people applied for 300 data-center fiber-technician training slots.

“I met with a few people from the government local governments and the state and they said that a bunch of bombed-out towns in the southern panhandle and parts of…”—

TECHNOLOGY

Muse Adoption — According to the host, Muse had about 3 million monthly active users. Ben says everyday users can put it to practical use and expects adoption to grow as it reaches more people; he also points to Meta’s glasses and necklace as physical AI products.

“That has totally reversed upon the release of Muse, which is a customer-facing application that can do lots of different things for people and has currently about 3 million monthly…”—

Token Economics — Ben argues that agentic workflows can drive substantial token costs because AI may run a task repeatedly, while CPU costs are less significant than commonly assumed. Converting a large share of Meta’s installed user base to Muse could require substantial compute, he says. He proposes measuring production cost per billion tokens per megawatt.

“The CPU cost is not as much as people think it is because they use like a AMD virtual server and it's provisioned in a way where people aren't constantly…”—

AMD Benchmark Dispute — Ben says AMD’s event materials listed CPUs he could not find in the company’s product specifications. He says no one responded to his request for an apples-to-apples benchmark comparison and calls the marketing unfair.

“When I was at the AMD everywhere event, they had a bunch of slides, and they listed a bunch of CPUs that when I looked back, didn't exist in their…”—

ECONOMY

U.S. Manufacturing — Ben argues that expanding U.S. manufacturing—including TSMC and Intel plants in Arizona and Corning fiber plants on the East Coast—could create jobs and help the U.S. lead in AI. He links profitable AI growth to addressing the country’s debt and cites 10-year yields of 5.2% as a concern.

“We don't even make TVs in America anymore. Right?…”—

MARKETS

GPU Financing — Ben argues Nvidia GPUs are easier to finance than custom chips because they are fungible and can be redeployed to another user. Lenders have established ways to finance Nvidia GPU clusters, he says, while custom startup chips lack the same financing pathways.

“The whole world can still use a GPU. It's still fungible. Today if Monetary Matters rents a cluster and says, "You know what? We don't want to use this anymore."…”—

Chip Supply Tightness — The host cites TrendForce’s assessment that semiconductor categories were tight to very tight, except GPUs, which were balanced. Ben suggests GPU supply may have held up because Nvidia controlled its own supply planning while supply-chain partners expanded more cautiously.

“And it's based out of Taipei. And they say that they're looking at every single stack of semiconductors…”—

Hardware Pricing Power — The host says Nvidia’s gross margin rose from 60% to 75% over five years amid tight supply and cites Seagate as another company with strong current pricing power. Ben cautions that commodity hardware benefiting from a squeeze cycle could lose pricing power once supply catches up. He says the data-center buildout follows 10–15 years of limited hardware investment and could spur capacity expansion, including potential Chinese supply.

“Like Nvidia's gross margins went from 60% to 75% over the past 5 years just because it's been so tight…”—

Intro

Ben describes Muse as a practical, accessible assistant that his family uses for everyday tasks, arguing that its usefulness and Meta’s broad user base could make AI adoption mainstream.

Why Meta's Muse is a GameChanger

The speakers argue that agentic AI drives far greater token consumption than CPU usage, making token-production costs, energy, and integrated system efficiency more consequential than a simple CPU-demand thesis.

CPUs are Overhyped (Why Token Bills Dwarf CPU Bills)

Ben favors Nvidia’s integrated platform over custom ASICs, emphasizing software, security, fungibility, financing, and total cost of ownership while questioning the economics and execution of AMD’s GPU ambitions.

Nvidia Dominates Custom ASICs ("Free Isn't Cheap Enough")

Ben says GPUs are not the only constrained part of the AI buildout: power, memory, storage, and other components matter, while supply-chain expansion and future commodity competition may erode pricing power.

GPU Supply Isn't Actually That Tight Compared To Everything Else

The speakers discuss local opposition, permitting delays, and scarce powered land as constraints on data-center construction, with Ben arguing that community benefits and domestic jobs are key to sustaining investment.

The Data Center Backlash in U.S.

Ben sees Anthropic’s large compute commitments as evidence of strong AI demand, while the speakers weigh fast-growing lab revenues against spending risks and the possibility that reported growth could slow.

Anthropic IPO Filing (S-1)

Ben expects scaling to continue for now but anticipates model segmentation, with different price and capability tiers judged by the cost of completing useful tasks rather than frontier performance alone.

The Limits of AI: Chess Comparison and Scaling Laws

Ben argues AI could accelerate biological research by analyzing molecular and experimental data, improving drug discovery and diagnostics, while acknowledging that clinical validation and successful treatments remain uncertain.

AI & BioPharma and Medical Breakthroughs

The speakers return to semiconductor investments: Ben views Meta as a potential AI beneficiary, considers Lam Research’s growth outlook, and describes Lattice as an interesting but richly valued niche supplier.

Lam & Lattice

Ben calls AMD overhyped given its valuation and unproven rack-scale execution, discusses Nvidia’s supplier commitments and economic spillovers, and closes by stressing the broader economic stakes of the AI buildout.

Related Episodes

Important Disclaimers

Not Investment Advice: For education only — not financial, tax or legal advice.

AI-Generated: Made by AI from source transcripts and may contain errors.

No Guarantees: Past performance does not predict future results; investing carries risk.

Our Commentary · Request Removal