RenMac Off-Script: AI's Reality Check*

PodcastRenMac Off-ScriptAra KharazianOct 9

The guest argues that AI adoption and spending remain concentrated, with limited evidence so far of broad productivity gains or benefits beyond software and tech. Competition is lowering frontier-model prices, while data center costs and local opposition may add pressure. Speakers also discuss inflation and Fed policy, mixed market breadth and credit weakness, and strong demand at recent Treasury auctions.

Artificial Intelligence AI Infrastructure Technology Stocks Inflation Monetary Policy Market Breadth

Preview · 22 more points in the app

Read the Full Summary

Key Takeaways

The guest says AI adoption is broad in one sense—about 56% to 58% of businesses pay for AI—but spending remains modest for the median firm at about $12 per employee per month, and adoption of tools outside OpenAI and Anthropic in areas such as customer service and sales is below 1% to 2%. The guest finds it difficult to envision the median firm spending thousands per employee, a level the guest says current valuations may require or assume.

AI spending and reported gains are concentrated: the top 1% of firms account for 80% of spend, while the guest’s analysis found faster headcount growth after adoption mainly among high-intensity adopters. The guest says revenue and headcount gains come from a highly selected group concentrated in software and tech, and that available evidence has not established similar gains spreading beyond tech and other white-collar firms.

The guest describes AI’s labor-market reach as limited so far: software engineers make up less than 3% of the U.S. labor force, and few common jobs would be meaningfully augmented by AI. Coding agents are commercially advanced but concentrated in software engineering, while the guest frames self-driving cars and image or video models as potential extensions of longer-running automation rather than a wholly new paradigm.

The guest attributes falling prices per million tokens mainly to competition between OpenAI and Anthropic, with providers also cutting public prices and offering contract discounts; about 95% of U.S. business AI spending goes through American models. The guest expects U.S. model companies to capture nearly all U.S. AI spending in the near future and ongoing product competition, while noting that self-hosted open-source models could push frontier prices lower, though few firms use self-hosted inference today.

GEO-POLITICS

AI Restrictions — One speaker expects the administration to avoid imposing more onerous AI restrictions, even if Democrats win control of the House and Senate. The speaker points to Jay Clayton's appointment as AI czar and the administration's concern about losing the race to China.

“My answer is no. I don't think that the administration is going to do anything to put more onerous restrictions on that…”—

TECHNOLOGY

AI's Labor Reach — The guest says AI can deliver clear returns for software firms, but software engineers make up less than 3% of the U.S. labor force, and few of the country's most common jobs would be meaningfully augmented by AI. Given that limited reach, the guest finds it difficult to envision the median firm moving from about $12 per employee per month to thousands in AI spending.

“And that's where I disagree with the discourse. I think AI has clear ROI gains and benefits for software firms and think about the job of software engineers specifically. 100%…”—

AI Spending Gap — The guest says the median firm spends about $12 per employee per month on AI and finds it difficult to envision that rising to thousands. The guest adds that current valuations may require or assume this level of spending.

“And so it's very hard to consider for me at least a situation in which you get the median firm that currently spends about $12 per employee per month spending…”—

AI Productivity — The guest says about 56% to 58% of businesses pay for AI in some form, but its use has not yet appeared in productivity statistics.

“One of the main metrics that we're tracking is just whether or not businesses are even using AI…”—

ECONOMY

Fed and Productivity — One speaker argues that promised productivity gains would not, in the speaker's opinion, deter the Fed from hiking if current data show strong demand and nominal growth. The speaker says the Fed can respond only to what it sees.

“You kind of have to I mean the Fed really can only fight one battle at a time and it can only respond to what it sees…”—

Productivity Growth — The guest argues that productivity improvements can improve products while also prompting price competition, rather than lifting every firm's margins. The guest says the accumulation of these changes is consistent with historical aggregate growth of 2% to 3% a year, not necessarily the much larger gains or losses some in tech expect.

“So you both kind of improve the product that also leads to the price competition. So you get quality improvements that may be same or similar prices…”—

MARKETS

Semiconductor Strength — A speaker says only about a third of semiconductor names are making relative-strength highs, while many remain weak. AMD and NVIDIA still look good, while SanDisk and SK Hynix remain uncertain.

“But look, you're still, I mean, the semi space in particular, you know, still only, I don't know, maybe a third of the names are, you know, making relative strength…”—

Equity Breadth — A speaker says stocks remain in an unequivocal uptrend and stresses that this is not an outright bearish call, despite weakness in credit and market breadth. With the market within 1% of a new high, 3.81% of names made 52-week highs and 13.4% made 52-week lows.

“First, I wouldn't say that we're bearish on stocks. There there are bearish elements to stocks. I think you want to be cautious, but that's not outright bearish…”—

Crypto and Equities — The speaker says crypto and equities are both in bullish trends, while emphasizing that assets need not move in lockstep. The view is based on market trends and momentum.

“So, you know, our 2¢ and the way that we think about the world is through trends and through momentum and what the market, the feedback of the market is…”—

Ramp AI Data

The hosts introduce Ramp economist Ara Karazian and his business-spending data, which offers a granular view of AI purchases and company adoption.

AI Adoption Concentration

Karazian says AI use is widespread but concentrated in a few firms and sectors, while advanced productivity tools remain uncommon outside tech.

AI Spending By Industry

AI spending is strongest in tech, finance, and technical services; adoption also reaches manufacturing and construction, though intensive use remains concentrated.

AI Hiring Effects

Ramp's matched-company analysis finds faster headcount growth among high-intensity AI adopters, but no clear evidence that AI is causing job losses.

Productivity And AI Returns

Karazian sees revenue and hiring gains among a selected group of tech-heavy firms, but says broad productivity benefits remain unproven and valuations may assume too much.

AI Beyond Software

Coding agents lead today's commercially useful AI applications, while voice and industrial tools have limited adoption; Karazian expects automation to extend into blue-collar work.

Frontier Model Price Competition

Karazian says AI spending is slowing among top spenders because OpenAI, Anthropic, and other frontier providers are competing aggressively on prices and contracts.

Market Breadth And AI

The hosts discuss uneven market performance, arguing that AI's concentrated benefits are prompting more selective investment across semiconductors and software.

Data Center Politics

The panel expects local resistance to data centers to spur companies to fund power and community benefits, while the administration remains focused on competing with China.

AI Inflation And Productivity

The speakers debate rising technology and data-center costs, while noting that AI's productivity effects may appear as better services and competitive gains rather than broad labor savings.

Crypto And Equity Trends

Asked how the team can be bullish on crypto without being bearish on stocks, DeGraff says trends differ by asset and warns that weak breadth and credit warrant caution.

Inflation Data And Fed

The panel previews CPI and PPI, highlighting potential price pressures in phones, subscriptions, and groceries that could challenge expectations for a temporary Fed pause.

Tracking AI Token Costs

Karazian is watching token volumes and spending to determine whether frontier-model price cuts or growing use of self-hosted open-source models explain falling AI costs.

AI Models As Commodities

Karazian expects model competition to keep improving quality and compressing prices, but argues that demand for the best frontier models will persist.

Credit And Treasury Auctions

DeGraff flags weak junk-credit performance and narrow equity-market strength, but says strong demand at the Treasury auction counters immediate fears of a US fiscal crisis.

US And Europe Rate Views

The hosts contrast US investors' view of higher rates as reflecting stronger nominal growth with Europe's treatment of rising rates as a shock, then thank their guest.