Skip to content

AI Is Getting Much Cheaper

The Real Estate Espresso Podcast

7m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode of the Real Estate Espresso Podcast explores the rapid maturation and market segmentation of artificial intelligence, drawing direct economic parallels to real estate and other industries. The host examines government census data on business adoption, noting that while usage is surging—reaching 26.8 percent of U.S. businesses and 55 percent for large enterprises—unit prices for AI capabilities have fallen dramatically by 67 to 80 percent year over year. The discussion argues that as raw intelligence commoditizes and companies route workloads to smaller, cheaper models, foundational model providers face a deflationary wall. For real estate and technology investors, this shift impacts data center capital expenditures, power-adjacent industrial land underwriting, and enterprise profitability as companies transition to low-cost commodity intelligence.

  • Monitor government census data alongside enterprise surveys to get an accurate picture of broad business technology adoption. 0:26
  • Recognize that unit price deflation in AI capabilities can outpace volume growth and slow revenue for foundational model providers. 2:26
  • Use dynamic model routers to send high-volume queries to inexpensive specialized models and reserve costly models for complex cases. 5:06
  • Convert variable usage-based fees into predictable fixed infrastructure costs by running open-weight models on dedicated hardware. 5:35

Highlights

  • 0:42

    US Business AI Adoption Ratedata

    Census Bureau data reveals that 26.8 percent of all U.S. businesses actively use artificial intelligence, with large enterprises at 55 percent.

  • 1:45

    Frontier Labs Feature Raceinsight

    Top-tier AI labs spend massive sums pushing reasoning benchmarks marginally higher, while enterprise buyers shop with a sharper pencil.

  • 2:30

    AI Capability Cost Deflationdata

    Equivalent AI workloads are 67 to 80 percent cheaper year over year as unit price deflation outpaces volume growth.

  • 3:37

    Data Center Capital Expenditure Riskinsight

    Massive capital investments in data centers and power infrastructure face stretched payback timelines if enterprises migrate to smaller models.

  • 5:12

    Dynamic Model Routing Savingsstrategy

    Routing high-volume queries to inexpensive models and complex queries to frontier models cuts inference costs by 50 to 90 percent.

  • 5:40

    Dedicated Hardware Cost Controlstrategy

    Running open-weight models on owned hardware turns variable usage fees into fixed costs while protecting sensitive data behind a firewall.

Quotes

  • “26.8% of all U.S. businesses are actively using artificial intelligence.”
    Unnamed speaker · 0:42
  • “For small businesses with fewer than 20 employees, reported usage rates are below 30%.”
    Unnamed speaker · 1:03
  • “That single decision point, that shift routinely cuts inference costs anywhere from 50% to 90% without sacrificing quality of the output at all.”
    Unnamed speaker · 5:23
  • “Ultimately, the macro business case for AI is transitioning from a supply-side gold rush to a demand-side productivity play.”
    Unnamed speaker · 6:31

About this episode

The cost to achieve a fixed level of AI capability has been collapsing by 67% to more than 80% year-over-year. When any market matures, it inevitably segments. At the top of the AI market, frontier labs are locked in a fierce feature race, spending staggering sums to push reasoning benchmarks marginally higher. Yet enterprise buyers are increasingly shopping with a sharper pencil. Rather than paying top dollar for the newest frontier model on every task, businesses are routing workloads to lower-cost token suppliers or deploying open-weight models on dedicated hardware. There was a time when only the state of the art model would perform these tasks. But today, a smaller, cheaper model crosses the "good enough" quality threshold for a given workflow. customers stop paying premium prices for marginal gains at the top of the market. Companies are making very granular decisions to choose the appropriate tool for each job. The savings are substantial. ---------------- **Real Estate Espresso Podcast:** Spotify: [The Real Estate Espresso Podcast](https://open.spotify.com/show/3GvtwRmTq4r3es8cbw8jW0?si=c75ea506a6694ef1) iTunes: [The Real Estate Espresso Podcast](https://podcasts.apple.com/ca/podcast/the-real-estate-espresso-podcast/id1340482613) Website: [www.victorjm.com http://www.victorjm.com/](http://www.victorjm.com http://www.victorjm.com/) LinkedIn: [Victor Menasce](http://www.linkedin.com/in/vmenasce) YouTube: [The Real Estate Espresso Podcast](http://www.youtube.com/@victorjmenasce6734) Facebook: [www.facebook.com/realestateespresso](http://www.facebook.com/realestateespresso) http://www.facebook.com/realestateespresso%5D(http://www.facebook.com/realestateespresso) Email: [[email protected] mailto:[email protected]](mailto:[email protected] mailto:[email protected]) **Y Street Capital:** Website: [www.ystreetcapital.com http://www.ystreetcapital.com/](http://www.ystreetcapital.com http://www.ystreetcapital.com/) Facebook: [www.facebook.com/YStreetCapital](https://www.facebook.com/YStreetCapital) http://www.facebook.com/YStreetCapital%5D(https://www.facebook.com/YStreetCapital) Instagram: [@ystreetcapital](http://www.instagram.com/ystreetcapital)

Transcript

0:00
7:03

1,155 words · 1 speaker

Topics

  • AI
  • Frontier Labs
  • Canada
  • open-weight models
  • Census Bureau
  • North America
  • dynamic model routers

Page updated · transcript and summary are machine-generated