The Real Estate Espresso Podcast

AI capability costs drop 67-80% as enterprise adoption rises to 26.8%

7 min Listen to the original Episode page
AI AdoptionEnterprise TechnologyCost DeflationMarket Segmentation

Summary

Victor Menasce discusses AI market segmentation, noting that while frontier labs race for features, enterprise buyers prioritize cost efficiency. He cites U.S. Census data showing 26.8% of U.S. businesses used AI as of October 2026, with higher rates in large enterprises (55%) and tech sectors (48%). The speaker explains that equivalent AI capability is now 67% to 80% cheaper year-over-year. This cost deflation is outpacing volume growth, causing revenue growth for pure-play model providers to decelerate.

Episode sections

  1. 0:01 – 1:38

    Market Segmentation and AI Adoption Rates

    Victor Menasce discusses how mature markets segment, citing real estate, automotive, and computers as examples. He introduces AI market segmentation, noting that while frontier labs race for features, enterprise buyers are seeking cost efficiency. He cites U.S. Census data showing 26.8% of U.S. businesses use AI as of October 2026, with higher adoption in large enterprises (55%) and tech sectors (48%).

  2. 1:38 – 7:03

    Cost Deflation and Gwinn's Paradox

    The speaker explains that while AI adoption and token consumption are rising, revenue growth for pure-play model providers is decelerating due to price deflation outpacing volume growth. He states that equivalent AI capability is now 67% to 80% cheaper year-over-year.

Do your own research

  • Verify U.S. Census Bureau data on AI adoption rates for October 2026.
  • Check financial reports of major AI model providers for revenue growth trends.
  • Analyze pricing trends for AI inference and training tokens from cloud providers.
  • Review enterprise procurement strategies for cost efficiency versus feature parity.
  • Assess the impact of price deflation on pure-play AI company margins.
  • Compare adoption rates between large enterprises and small businesses.
  • Investigate the definition of 'equivalent capability' used in cost comparisons.
Full transcriptexpand
0:01
Victor Manache: Welcome to the Real Estate Espresso Podcast, your morning shot of what's new in the world of real estate investing. I'm your host, Victor Manache. When any market matures, it inevitably it carves up into segments. We see this every single day in real estate. We see it in automotive, we see it in computers, we see it in hospitality, all kinds of different businesses. Now at the top
0:21
Victor Manache: of the market, suppliers compete in a race for amenities and features. Well we're starting to see the same thing happening in the realm of artificial intelligence. Recent government census data provides the most accurate picture of broad business adoption across North America this year. That avoids the skew of some of the enterprise only surveys.
0:40
Unidentified speaker (UNKNOWN): In
0:40
Victor Manache: the U.S., as of October of this year, data from the Census Bureau shows that 26.8% percent of all U.S. businesses are actively using artificial intelligence. In Canada, according to statistics, Canada 19.2%. But that was in Q2, that was a few months earlier. Adoption rates vary heavily depending on the size and nature of the business. Large
1:00
Victor Manache: enterprises in the US, with businesses having at least 250 or more employees, AI adoption sits at 55%. For small businesses with fewer than 20 employees, reported usage rates are below 30%. The highest usage, not surprisingly, is in the information and tech sectors, with overall usage approaching 48%, followed closely by education and
1:19
Victor Manache: professional services. Now that may sound like a low percentage, but it's actually much faster than the adoption rate we experienced in the mid-1990s with the internet. Well, right now, the exact same market segmentation
1:31
Unidentified speaker (UNKNOWN): that
1:31
Victor Manache: we see in automotive and many other industries is unfolding at breakneck speed in artificial intelligence. And the financial implications
1:38
Unidentified speaker (UNKNOWN): are not going to be a very important thing
1:39
Victor Manache: carry major lessons for both technology investors and real estate operators. At the top of the AI market, Frontier Labs are locked into a fierce feature race, spending huge sums of money to push their reasoning benchmarks marginally higher. Yet the biggest customers, the enterprise buyers, are increasingly shopping with a sharper pencil. Rather than paying top dollar
1:59
Victor Manache: for the newest frontier model for every single task, many businesses are routing workloads to lower cost AI suppliers, and they're often deploying open weight models on dedicated hardware. So here's an interesting paradox. Overall, adoption of AI is surging, and consumption of AI tokens is climbing rapidly, especially as companies deploy more complex
2:19
Victor Manache: workflows involving AI agents. But the revenue growth rate for pure play AI model providers is in fact starting to decelerate. And that's because the unit price deflation is outpacing the volume growth. In fact, for an equivalent workload, say a year ago, that same AI capability is anywhere from 67 to 80% cheaper year
2:39
Victor Manache: over year compared with this time last year. Economics has a model to describe this. It's called the G Von's paradox. The idea that falling prices unlock so much new demand that total spending increases.
2:49
Unidentified speaker (UNKNOWN): But
2:49
Victor Manache: when raw intelligence come up, commoditizes faster than the demand can absorb those price cuts, the foundational model providers hit a deflationary wall. Most business tasks like extracting data from documents, summarizing reports, managing your calendar,
3:03
Unidentified speaker (UNKNOWN): or
3:03
Victor Manache: handling Tier 1 support simply does not require the most sophisticated models. There was a time when only the state-of-the-art model could perform these tasks, but today a smaller, cheaper model crossing crosses that good enough quality threshold for many given workflows. Customers simply stop paying premium prices for marginal gains at the top
3:23
Victor Manache: of the market. So let's look at what this means for the broader business case, especially where technology intersects with commercial real estate.
3:29
Unidentified speaker (UNKNOWN): The
3:30
Victor Manache: hyperscalers, the Frontier Labs have committed hundreds of billions of dollars in front loaded capital expenditures for their data centers, for their electric substations, their cooling infrastructure and the GPUs, that entire build-out was predicated on the assumption of exponential revenue growth from massive centralized training clusters and high margin
3:49
Victor Manache: queries. If enterprise workloads increasingly migrate to smaller models on dedicated hardware, the payback timeline for those massive capital investments stretches out in time. For real estate investors, underwriting data center developments. or power adjacent industrial land, that capital expenditure to revenue mismatch is a critical factor to
4:09
Victor Manache: watch. Demand is gradually shifting from massive remote training campuses towards an actual co-location of the hardware next to the end users. So does this mean that the artificial intelligence business case is broken? Well, not really. But profitability is not appearing from the ecosystem. It's migrating to other layers. with tighter operational control. As
4:29
Victor Manache: raw token generation turns into a low margin commodity, the value is shifting towards chips that are optimized for answering questions rather than training. The value shifts to software integration with business systems and most importantly, the enterprise buyers themselves. In fact, slowing revenue growth amongst model vendors is paradoxically bullish for the real
4:49
Victor Manache: world business adoption. Now, over the last couple of years, unpredictable usage bills and linear her token pricing have forced many organizations to stall their rollout of AI before reaching full production. Today, some of the more mature buyers are reaching an ROI inflection point by choosing the right AI architecture. Instead of one expensive one
5:09
Victor Manache: size fits all model for everything, companies are using dynamic model routers that is sending the their particular query to the right tool. In some cases, high volume queries go to an inexpensive specialized model, and then they reserve those costlier frontier models strictly for complex cases that need the extra power. That single decision
5:29
Victor Manache: point, that shift routinely cuts inference costs anywhere from 50% fifty percent to 90% ninety percent without sacrificing quality of the output at all. And if you're ready to bite the bullet, and run some of these open weight models on dedicated hardware that you own, you're converting variable usage based fees into predictable fixed infrastructure costs. Once the
5:49
Victor Manache: hardware is in place, you can have high volume processing that you own at near zero marginal cost. And at the same time you're keeping sensitive data locked behind your own company's firewall, insulating operations from concerns about security and data leaks. And for those of us in real estate development, for property management and private equity, that is
6:09
Victor Manache: tremendous news. Now, whether you're underwriting acquisitions, auditing thousands of commercial leases, tracking construction draw schedules, or managing tenant communications across a multi state portfolio, the cost of intelligence is plummeting. You no longer need to hand over your proprietary data or sacrifice your operating money. margins to a high priced AI
6:28
Victor Manache: vendor just to automate a complex task. Instead you can deploy dedicated models tailored specifically to your asset class. Ultimately the macro business case for AI is transitioning from a supply side gold rush to a demand side productivity play. In this next phase of the cycle, durable profits will accrue to the businesses that combine
6:48
Victor Manache: low low-cost cost commodity intelligence with proprietary data and deeply integrated operations. As you think about that, have an awesome rest of your day. Go make some great things happen. We'll talk to you again tomorrow.
This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.

More from The Real Estate Espresso Podcast

The Real Estate Espresso Podcast

Live From Calgary - Investing In An Upside Down Market

Episode discussion: Live From Calgary: Investing In An Upside Down Market

Speakers’ claims: Victor Menasce argues that the traditional real estate value-add playbook is broken due to excessive new supply and tenant 'lease surfing.' He advocates for 'value engineering' to control project economics by challenging design assumptions and navigating financing rules like CMHC MLI Select. Menasce suggests acquiring distressed assets at deep discounts and targeting niche gaps such as active adult housing. He concludes that while current oversupply exists, low construction starts may lead to a shortage in two years.

Real EstateValue EngineeringCMHC Financing
20 min0 assets10 claims to verify
The Real Estate Espresso Podcast

New Rules Affecting Smart Locks

Episode discussion: New state laws mandate non-app alternatives and data privacy for rental smart locks

Speakers’ claims: Host Victor Manash discusses new legislation in New York City, Oregon, and Washington State affecting smart lock usage in rentals. He highlights the Tenant Data Privacy Act in NYC, Oregon's House Bill 3378, and Washington's Engross Substitute Senate Bill 5937. The episode details three key compliance rules in Washington: providing non-app alternatives, delivering privacy policies, and adhering to data minimization. Manash advises against app-only Bluetooth deadbolts and recommends multi-credential hardware with scrutinized vendor privacy policies.

Real Estate LegislationSmart Home TechnologyData Privacy
5 min0 assets7 claims to verify
All insights