The Fourth AI Prediction That Hasn't Held Up: Cheap AI Workers—and Where I'm Investing
By Ian Ippolito · Last updated October 1, 2026 · Code of Ethics
AI has emerged as a transformative technology, but many bosses expected it to be an ultra-cheap way to replace employees. Instead, some big companies have run through a year's AI budget in a few months. And the buildings and electricity needed to run AI are getting harder to come by. That scarcity creates potential opportunities for investors, and so does the work AI still can't do. Here's what I've found in each area since laying out my AI investing strategy earlier this year -- including what has and hasn't passed muster and where I'm still looking.
(Usual disclaimer: I'm just an investor expressing my personal opinion and am not an attorney, accountant nor your financial advisor. Consult your own financial professionals before making any financial decisions. Code of Ethics: To remove conflicts of interest that are rife on other sites, I/we do not accept ANY money from outside sponsors or platforms for ANYTHING. This includes but is not limited to: no money for postings, nor reviews, nor advertising, nor affiliate leads etc.
Nor do I/we negotiate special terms for ourselves in the club above what we negotiate for the benefit of members. Info may contain errors so use at your own risk. See Code of Ethics for more info.)
In my last article, we discussed how AI is an extremely transformative technology -- but several predictions about it have failed to materialize. Now here's a fourth prediction about it, that also hasn't panned out -- and the implications for investors.
Prediction 4: AI Would Be a Cheap Way to Replace Employees
What the Bosses Expected
Companies cut staff or stopped hiring because they expected AI to do the same work for less.
In 2025, Salesforce CEO Marc Benioff cut 4,000 customer support roles. He explained why on The Logan Bartlett Show podcast, as CNBC's Scott Budman reported:
"I've reduced it from 9,000 heads to about 5,000, because I need less heads," Benioff said while discussing the impact of AI on Salesforce operations.
At IBM, CEO Arvind Krishna told Bloomberg News in 2023 that AI could take over thousands of back-office jobs. Al Jazeera reported:
“These non-customer-facing roles amount to roughly 26,000 workers,” Krishna said in the interview published on Tuesday. “I could easily see 30 percent of that getting replaced by AI and automation over a five-year period.”
And many other bosses said something similar.
What Happened Instead
In my last article, I described how many bosses are now doing an about-face. A majority of leaders who made AI-related cuts told Orgvue it was the wrong decision. Robert Half reports that 32% of U.S. hiring managers who eliminated positions after implementing AI later brought those or similar roles back. Gartner expects up to 30% of roles displaced by AI to be rehired by 2029, often at a higher cost.
Quality was part of it, as Klarna's CEO admitted. And we talked about that in the last article. But the cost turned out to be another surprise.
The Bills They Got
Many of the biggest companies ran through a year's AI budget in a few months.
Uber and ServiceNow were among them. Jyoti Mann reported in The Information on June 12:
Uber and ServiceNow used up their entire annual budgets for Anthropic’s tools in just the first few months of 2026.
And surveys show how common the overruns are. CFO Dive's Alexei Alexis reported on July 22, 2026:
Nearly seven in 10 U.S. companies (68%) say at least some of their artificial intelligence initiatives ran over budget in the past year, with one-third (33%) reporting that overruns occurred mostly or always, AI security firm WitnessAI said in a report released Wednesday.
And the surprises are forcing hard decisions. CFO Dive reported on August 11:
Sixty-two percent of organizations surveyed said an unexpected AI cost materially altered a business decision over the past year. Among those organizations, 40% required board-level escalation, 33% implemented emergency spending freezes and 25% delayed or canceled an AI initiative, according to Mavvrik’s 2026 State of AI Cost Governance Report released last month in partnership with Benchmarkit, a software-as-a-service performance metrics firm.
And many companies have already started limiting use. The Information reported on June 16:
The latest is AT&T, which has begun limiting some employees’ access to Microsoft’s Github Copilot, according to a person at the company.
It added:
Uber and Walmart similarly capped employee usage of AI coding tools, Bloomberg recently reported.
Why the Price Tag Fooled People
AI is priced by the token. Tokens are the small pieces of text AI reads and writes, and the price of each one really did fall fast. OpenAI CEO Sam Altman wrote in Three Observations on February 9, 2025:
The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use. You can see this in the token cost from GPT-4 in early 2023 to GPT-4o in mid-2024, where the price per token dropped about 150x in that time period. Moore’s law changed the world at 2x every 18 months; this is unbelievably stronger.
But here's the key thing. A token isn't a unit of work. Newer AI models work through far more tokens to finish a task. And as Altman's own sentence says, lower prices lead to much more use. So a falling price per token can still come with a bigger bill for getting the work done.
Research firm Gartner now expects AI to cost more than the people it was supposed to replace. Its June 24, 2026 forecast for software development:
By 2028, AI coding costs will overtake the average developer’s salary due to rising large language model (LLM) token consumption and the shift to consumption-based licensing models, according to Gartner, Inc., a business and technology insights company.
Gartner analyst Nitish Tyagi:
“Software engineering leaders are increasingly concerned as token-driven AI spend becomes harder to justify, with budgets often being depleted earlier than expected.”
Gartner expects the same in customer service. As CX Dive put it in its headline, "Gartner challenges assumption that AI will be cheaper than human support." Gartner now forecasts that AI customer service will cost more per case than many offshore human agents. From its January 26, 2026 forecast:
By 2030, cost per resolution for generative AI (GenAI) will exceed $3, higher than many B2C offshore human agents, according to Gartner, Inc, a business and technology insights company.
Rising data center costs, a pivot from subsidized growth to profitability for AI vendors, and increasingly complex use cases that consume more tokens and require expensive talent, will lead to soaring AI costs for customer service organizations.
Gartner analyst Patrick Quinlan added:
“Full automation will be prohibitively expensive for most organizations; instead, leading organizations will use AI to drive customer engagement rather than to cut costs.”
The Physical Limits
Even if the price per token keeps falling, AI runs on things that don't get cheaper at software speed: land, buildings, permits and electricity. And those are getting harder to come by.
Communities are pushing back. Data Center Watch reported for the first quarter of 2026:
At least 75 data center projects worth approximately $130 billion were blocked or delayed in a single quarter — roughly matching the scale of all of 2025 in just three months.
Its second-quarter report added nearly $68 billion more. That's almost $200 billion blocked or delayed in six months.
The opposition is widespread. Gallup's March 2026 survey found:
Seven in 10 Americans oppose constructing data centers for artificial intelligence in their local area, including nearly half, 48%, who are strongly opposed.
And governments are adding conditions. Newsweek reported on Texas Governor Greg Abbott:
He said future projects in Texas will have to meet requirements related to water usage, electricity demand, consumer costs and local community approval before they can move forward.
Money alone doesn't get a data center approved.
Power is a huge bottleneck. ICF's 2025 forecast expects U.S. electricity demand to grow about 25% by 2030 and 78% by 2050, measured from 2023. That covers more than AI, including electric cars, heat pumps and manufacturing. So data centers are competing for power with everything else that's growing.
My view of the bottom line
My view is that AI will remain scarce and expensive for a long time, with power and infrastructure as a main bottleneck. Cheaper processing won't instantly build power plants, transmission lines or an approved data center. That's why I look at the things AI can't run without, and at the work it still can't do.
How I'm Investing: An Update to the Original Series
This section includes links to detailed info and due diligence on specific investments in the Private Investor Club. Existing members can access them by clicking on the links. If you are not a member you can join for free--after verifying that you're not a sponsor or a sponsor affiliate. Click on the link for more details. In Part 3, I laid out ways to invest in AI's growth and in businesses whose essential work is harder to replace. Here's what that search has produced so far.
A. Frontier Models and AI venture investments:
Those growing bills are revenue for the companies selling AI. Uber and ServiceNow, for example, ran through their annual budgets for Anthropic's tools.
The most direct approach is investing in the companies building frontier AI models—the leading systems behind the tools people use. If their technology becomes more useful and widely adopted, those businesses have an opportunity to grow.
Quartus AI Fund II: This fund could be a match for others but was not for me. It targets mid stage companies which are a higher risk versus targeting late stage companies. On the other hand that also means higher reward if it pans out.
Hermes AI client: A successful looking company but again not a match for me because I was concerned that they did not have a defendable niche and looked likely to me to be displaced by the Frontier Labs.
Accerl 8: Similar issue. Not a match for me because I feel it's too adjacent to the frontier AI models (and too easy to get subsumed).
AI and Robotics fund: name withheld by request: I invested in this fund. I liked the late stage venture focus -- which can be lower risk due to more established businesses, --the established and profitable past track record and the specific portfolio selections. All investments have risks and it may or may not turn out ... and we'll see.
Note their name has been withheld by request in this public article, but is available to Private Investor Club members at the above link. If you are not a member you can join for free via that link--after verifying that you are not a sponsor or a sponsor affiliate.
Mistral AI: This is Europe’s sovereign AI lab and grew 20x in the last year with annualized revenue run-rate exploding from ~$20M in January 2025 to $400M+ in January 2026. It has the largest revenue base of any European developer of large language models and management expects to exceed $1B ARR by the end of 2026 -- which would represent more than 50x growth in two years. I invested through a fund.
Project Prometheus: Founded by Jeff Bezos, it's applying AI to A brand new space: the physical world of advanced engineering and manufacturing products (which is a $6.8 trillion dollar market). eam has scaled to roughly 150 employeeswith senior researchers recruited from OpenAI, DeepMind, Meta, and xAI and offices in San Francisco, London, and Zurich. I invested through a fund.
ZaiNar: Has working technology that's the first practical alternative to satellite GPS and works where it goes dark: indoors, underground, through walls -- and much more accurate (to under 10 cm at ranges up to 1.5 km). Has $450M+ in signed contracts and letters of intent with software-first, high-margin recurring economics across telecom, defense, healthcare, and logistics. I invested through a fund.
B. Infrastructure: The Buildings and Networks AI Runs On
When communities block or delay billions of dollars of new data centers, the ones already approved, powered and running become harder to replace.
AI needs physical places to run. Data centers house the computers, and communications networks connect those computers to customers. Investing in those buildings and networks is a way to participate in AI's expansion without having to pick which AI company wins.
The PwC/Urban Land Institute property outlook ranks data centers first among 27 property categories for investment and development prospects:
Looking at the ratings of all asset classes provided by the Emerging Trends survey respondents, data centers remain at the top for investment and development prospects. Data centers have held this first-place rank for three consecutive years and are the only subsector with both prospect scores above four, indicating sound investment and development conditions.
The funds I've found so far:
Accordant / Affinius: the data-center investment example, I introduced in the original series. Unfortunately, they never responded to the last set of questions so I dropped them from my personal list.
Palistar Digital Infrastructure Fund III: invests in communications to Did I accidentally publish wers, fiber connections and broadband, with some data centers. AI services depend on these networks to move information between users and the computers running the models, so this covers the connections as well as the buildings. I had concerns about their amalgamated track record and a high number of unrealized deals and decided to pass. But I could see others liking this investment and it certainly could do well.
KKR AI infrastructure fund: another investment in the infrastructure behind AI—the physical capacity needed to run AI, as use expands. Still gathering data on this one and arranging access ... but potentially interesting to me.
Aphorio Carter Mission-Critical Infrastructure Fund II: owns data-center properties leased to businesses. I currently intend to invest. They requested that their details not be posted here, so they are available to private investor club members at the link.
C. Energy: Feeding the Machines
In Part 2, I wrote that power was quickly becoming a bottleneck for AI. The forecasts above show it has arrived. AI consumes massive amounts of electricity, and the data centers running it need dependable power. Meeting that demand takes a mix of sources, including natural gas, renewables and nuclear power. The International Energy Agency explains:
Renewables and natural gas take the lead in meeting data centre electricity demand, but a range of sources are poised to contribute.
So this investment thesis is to help supply that power and the fuel needed to produce AI. Here's an update on those funds:
Mewbourne: One of the top performing track records in oil and gas drilling syndications. Getting an allocation can be difficult. I was able to at least get a foot in the door for a potential future allocation. So I am keeping a close eye on them for the future.
Cantor Fitzgerald Energy Fund: a strategy buying partial interests in producing wells, without operating the wells itself. This one is still preliminary so waiting to see Exactly what they end up structuring this as, before deciding.
Montego Minerals: Also one of the top performing track records in this sector, but with a mineral and royalty strategy (which can be lower risk and safer than drilling). I like a lot of things about this fund including the strategy, the sponsor experience and the 1031 exchange which was much more tax beneficial than with a drilling fund. In the end though I didn't pull the trigger due to some documentation issues and higher than expected minimums. But I can see another investor really liking this fund and would not be surprised to see it do well.
Others: There are eight or nine others that are being tracked right now in the club. So far none are compelling enough to make me to pull the trigger. But other investors are liking them and investing. Also there are a lot of tailwinds right now with the Iran war driving up the price of oil. So I would not be too surprised to see this vintage year end up being an extremely good one across the industry.
D. AI-Resistant Businesses:
My last article discussed how AI still needs people to design, supervise and check its work. For work done in person, AI can help with the scheduling and paperwork, and someone still has to show up and do the job.
Another focus for me is owning AI resistant businesses (whose essential work is harder for AI to replace).
Small-bay industrial.
These properties house local service businesses such as plumbers, electricians and heating and air-conditioning contractors. AI can help them run their businesses, but a human still has to go out and do the physical work. And landlords tend to be able to raise rents aggressively because many can pass along the cost to their customers. For example if you have a water leak in your house, you will probably not haggle if the price is a little bit more than last year. One downside of this versus traditional industrial is that it requires a lot more hands on management because of the large number of businesses. So for me the experience of the operator in the specialty area, is key.
Denholtz and Cove Capital are small-bay examples in my research. It's early but I am Doing due diligence and considering both.
Outpatient medical real estate.
AI can help with diagnosis and administration, but patients still need places for examinations, procedures and treatment. Clarion Partners describes the shift toward care close to patients:
Greater emphasis on localized care, as patients prioritize convenience and proximity, is driving providers to move services closer to where patients live.
Cypress Exchange and Vital Capital are medical-property examples that I've found in my research. These were only recently identified so still early in the process for me.
Search funds.
These back entrepreneurs who buy and operate established small businesses. The AI-resistant opportunity is in businesses built around physical work and local services. The Search-fund Sponsors discussion covers several of these funds (and several that I chose to invest in).
Sports and live entertainment.
In a world where AI grows, people are expected to pay more for authentic, human experiences. That includes sports:
AlphaSummit GoalLine: I could see this being a match for others because of many of the well known sports teams in the portfolio. But it wasn't a match for me due to the VC sleeve (which I felt was not a core competency) and the two layers of profit-sharing fees. Others will probably look at this differently and I hope it does well.
Apollo Sports Fund: roughly 80% was lending and related financing, rather than the equity approach I personally was looking for. So it wasn't a match for me. But I could see others being attracted to it due to the long Apollo history in credit and what should be access to deals that others can't get.
Arctos / KKR: Confidentiality challenges prevented club access. So this one didn't workout.
So I haven't found a sports fund yet that works for me, but am continuing to look.
Discussing AI and More
Private Investor Club members have been discussing AI in depth. There’s also been detailed discussion on thousands of sponsors and deals (including due diligence and real-life investor experiences).
If you're already a member, click here to discuss further:
If you're not yet a member, then joining is free.
- To protect all members and keep the conversations useful and confidential, all applicants are required to verify that they're solely investors (and not investment sponsors, platforms or their affiliates).
