TDC 062: Is the AI Bubble Bursting?? (The Shocking MIT Study, Key Observations from a Billionaire Summit)

TDC 062: Is the AI Bubble Bursting? My Keynote Insights + MIT's Shocking 95% Failure Rate

While 65% of all VC flows into AI, the real opportunity lies in the barbell approach.

---

Episode Summary  

In this episode of The Digital Contrarian, host Ryan Levesque dives into AI bubble dynamics and enterprise failure rates. 

You'll learn why 95% of AI projects fail, how billionaire investors view current market concentration, and discover the barbell strategy for hedging against bubble conditions while maintaining human competitive advantage. 

---

Question of the Day 🗣️  

How are you balancing AI automation with maintaining the human elements that make your business unique?

---

Key Take-aways  

- 95% of enterprise AI projects are failing according to new MIT research 
- 65% of all venture capital now flows into AI—the highest sector concentration in history 
- Geographic distance kills AI effectiveness in most business applications 
- The barbell approach protects your human core while deploying strategic AI assists 
- Lead with humanity first, then layer in AI where ROI is obvious 

---

Timestamped Outline ⏱️  

00:00 – Is the AI bubble bursting? 
00:48 – Newport Global Summit keynote insights 
02:15 – The shocking MIT study: 95% AI failure rate 
03:40 – Sam Altman's market manipulation timing 
05:35 – Venture capital hyper-concentration warning 
06:35 – Data center overinvestment analysis 
07:25 – Why geographic distance kills AI effectiveness 
08:30 – The barbell approach: Human core + AI assists 
09:15 – 4-step practical implementation framework 
11:30 – 30-day sprint you can start Monday 
13:15 – The big takeaway: Lead with humanity 

---

Links & Resources 🔗  

- The Digital Contrarian newsletter → https://thedigitalcontrarian.com 

---

Connect & CTA 🎯 
👉 Enjoyed this? Subscribe & leave a review on Apple Podcasts.  

🎁 Join 100 000+ digital entrepreneurs who get Ryan's "Strategic Insights for Digital Entrepreneurs Who Think Differently" every weekend:<https://ryanlevesque.net/join-the-digital-contrarian/>

---

Credits  
Host: Ryan Levesque
© 2025 RL & Associates LLC. All rights reserved.

Transcript

Is the AI bubble bursting? The shocking MIT study? And key observations from a Gilded Age billionaire summit? This past week, I was in Newport, Rhode Island, where I delivered the opening keynote at the Newport Global Summit, a billionaire investor event hosted at the Newport Art Museum in the room that you see here. Now, one of the highlights of this event was getting to connect with Steve Forbes, and to hear how my talk had an impact on his thinking. Now, in case you're not familiar, Newport, Rhode Island is the site referenced in HBO's hit TV show, The Gilded Age, and has been a billionaire playground since the late 19th century, with many third, fourth, and fifth generation Gilded Age families still calling Newport home with summer cottages like the one that you see here. 

Now, it was ironic because in my keynote, I referenced that like in The Gilded Age, which is circa 1870 to 1900, the vast majority of people in the United States and much of the world today are experiencing the greatest concentration of wealth in their lifetime, a trend that's been accelerated at least partly with the rapid ascent of AI. But the question is this, in spite of this rapid AI ascent, is it possible that the AI bubble is already upon us, and that it may be ready to burst? Well, brand new evidence suggests that the answer is perhaps yes. In fact, two private presentations at the Newport Global Summit, along with public data that's just been released this week, should lead us to at least question whether we may be closer to the bubble bursting than some may think. 

Let me explain. The MIT Gen AI study that sent stocks reeling. Now, in case you missed it, earlier this week, MIT released a report through their NANDA program, revealing that 95% of generative AI pilots at companies are failing right now.

And in case you haven't read this 26 page report, here are a few highlights. Based on interviews with 150 leaders, a survey of 350 employees, and analysis of 300 public AI deployments, only 5% of AI deployments at the enterprise level have achieved rapid revenue acceleration. The overwhelming majority of in-house AI projects are failing. 

In fact, consumer tools like ChatGPT offer a massive boom to individuals, but they don't have the same effect at the enterprise level because they don't learn from or adapt to complex corporate workflows. The failure rate of enterprise AI projects is actually understated based on organizations unwillingness to report their failures. In other words, when you dig beneath the surface and take time to measure profitability, organizations at the enterprise level are not seeing the payoff that Gen AI promises to deliver, at least not yet. 

In addition, there are several Gen AI myths from the report that I think are particularly interesting to note, especially around AI job replacement and what's really holding back AI effectiveness. Check it out. Now, the release of this report coincided with Sam Altman's most recent market manipulation media tour, pushing a narrative that an AI bubble is forming, strangely on the heels of his record $500 billion private company valuation, surpassing that of SpaceX, making OpenAI the most valuable private company in history. 

And it was the combination of the MIT Manda report and Altman's public AI bubble assertions that sent AI and AI adjacent stocks like Palantir Technologies and NVIDIA reeling earlier this week. Stocks have since recovered on the Fed's signal of a possible rate cut in September, but the fact remains. We are very likely experiencing AI bubble conditions, and this extends far beyond just a few public company stocks, which brings us to the shocking data on data centers. 

There were two key observations from the talks at the Newport Global Summit that really stopped me in my tracks. Number one, venture capital hyper concentration. Right now, 65% of all venture capital investment is going into AI projects. 

This is the greatest concentration of venture capital into any one sector of the economy that we've ever seen in history. And the last time we saw anything close to this was the 1999 to 2000 dot com boom right before the crash. Observation number two, data center overinvestment.

Right now, we're seeing massive investment going into data center infrastructure. In fact, check it out. In the United States, the value of data centers under construction is on track to surpass the value of office construction as early as this year, meaning at a physical infrastructure level, we are literally replacing space design for human workers with space design for our digital replacements at a pace unlike anything we've ever seen. 

But here's the insight I found most interesting from the Newport Global Summit. Most of this investment is going into data centers built far away from population centers and instead close to power generation centers. For example, locations like Northwest Texas and North Dakota. 

These locations make sense for data centers built specifically for training LLM models because real estate and power generation is relatively cheap in these geographies. But what these locations are not good for is supporting ongoing back and forth AI model interaction with end users. The geographic distance makes latency high and interaction slow. 

So while these high distance data centers make sense for the initial training of models needed today, they're not great for the longer term need to support back and forth AI interaction with users, which is what we will really be needing in the coming years. Meaning, just like what we've seen with vacant shopping malls and empty office space dotting the country, there's a chance that we may be seeing a massive overinvestment into soon to be obsolete AI real estate and infrastructure. Turns out the FOMO and pressure to be not left behind in the AI bonanza is strong. 

Whether you're a billion dollar hedge fund manager looking to deploy capital or a solopreneur or small business owner simply trying to keep up with the AI curve and figure out how and where to implement AI in your business. Which brings us to the question, what does this all mean for you and me? Well, I believe the short answer is to hedge. And based on everything that we've explored together in this issue and past issues of the Digital Contrarian, I recommend taking a barbell approach. 

Specifically, on one end, a human crafted and artisanal core in your business that nobody else can copy. I'm talking about your voice, your relationships, your in-person experiences, the unscalable touches that define your category of one business. On the other end, highly targeted AI assists that remove the drudgery and free your calendar to automate repeatable Monday non-client interfacing tasks and amplify what already works behind the scenes. 

What you don't want to do is this. You don't want to toss the baby out with the bathwater and let AI everything bulldoze what makes your business special. The magic that makes your people attracted to you and your brand. 

I'm seeing so many people mess this up with more and more wishing that there was a giant control Z undo button for much of what they've done with AI generated content over the last 12 months. Now, I've put together a hypothetical pragmatic plan that you can potentially execute in the next 30 to 90 days, beginning with step one, draw the line, human core versus AI assist. Number one, define your no AI core. 

So this is where you can draft a one page doc. These are the things that are human only in your business. Things like strategy, source of truth, content, brand voice, sales calls, client interaction, and anything that requires heart, humor, humanity. 

So push this and declare this to your team. And if you're feeling really bold to your customers, then number two, define your AI assist perimeter. So document the repetitive, low judgment work that AI will help with things like summaries, first pass research, formatting, repurposing, tagging, scheduling, doing transcripts, SOP drafts, and so on and so forth. 

Give yourself a target of say X hours per week of your time that you want to reclaim so that you don't end up swapping work you're doing now for time that gets sucked up simply working with AI tools instead. Which brings us to step two, install a simple AI hedge stack. Make it small business friendly. 

So number one, pick one co-pilot, one automator, and one guardrail. So co-pilot, pick one large language model that's yours of choice, whether it's Claude or ChatGPT or Gemini or Grok or whatever, with your custom instructions and a small private knowledge base. Pick one automator. 

So a basic workflow tool like make.com or Gumloop or N8N for two to three high leverage chores in your business. And then a guardrail, a standing human in the loop that is a step before anything goes public, whether that's with your brand, with facts, with your tone, with ethics. And this is where the 10-80-10 rule that we've explored together in past conversations is very, very effective. 

And there are no exceptions to this rule. Then number two, pilot where ROI is obvious. So start with your back office and ops, not your brand voice. 

In fact, this is one of the biggest observations from the MIT study. If it doesn't save you time or money in the next 30 days, then kill it. Don't buy into the hype. 

Step three, keep the moat human. Remember, oxytocin over dopamine. This is where you want to ship one source of truth piece of content weekly. 

This is your long form in your voice, unsanitized, straight from your brain, straight from your hearts that you produce. Everything else that you do can be repurposed from this one source of truth piece of content. Number two, design one unscalable client ritual. 

This can be a handwritten note, a 10-minute loom, a surprise phone call, a local meetup, or a client roundtable. Trust is earned in these so-called wasteful moments. Remember, optimize for oxytocin over dopamine. 

And then number three, label it, don't fake it. When AI helps you out, disclose it. An approach of disclosure over deep fake builds trust with your audience. 

Your voice has a unique energy signature, and your audience can feel the difference when it comes from you versus when it comes from AI, even if it's an AI assist. Like, for example, with this section of this episode here today, I actually asked Chat GPT to help me brainstorm ways to operationalize some of the concepts that I've been speaking about, that I've been writing about with practical tactical steps. And that's why this section may feel a little different from the first part of this week's episode.

It will also feel different from the passage coming after this section, and you'll see what I mean by that in just a moment. But that takes us to step four, be ruthless about results, not hype. So step one, red team your AI spent. 

Once a month, what did AI actually save or make? Keep what works and cut what doesn't. Consider capping your AI OpEx to a percentage of your revenue until you've proven payback. And remember to factor the value of your time that you're personally spending into this ROI calculation. 

Don't let it be a black hole. And then number two, continue to own your data. Continue capturing zero party data ethically.

I'm talking things like surveys, quizzes, applications, and keep it portable so that AI can analyze it, but make sure that it doesn't own or control it. Even in 2025, an email list of customers is still more valuable than a social media audience of subscribers. Now, here's a simple 30-day sprint that you can start literally on Monday. 

So if you're looking for a simple way to start putting this all into action, here's a simple sample plan. So in week one, draft your no AI core and your AI assist perimeter. Pick one copilot, one automator. 

Week two, ship your first weekly source of truth piece of content. Remember, coming from your brain, from your heart, and one unscalable client ritual to your delivery. Week three, automate two low judgment chores and install that human review step using the 1080 framework. 

And then finally, week four, keep the wins and kill your AI darlings. And then iterate from there. Remember, you don't have to get it perfect. 

You just have to get it going. And the best time to get it going is right here, right now, today. Which brings me to the big takeaway in all this. 

The big takeaway that I want to leave you with is this. Do not be misled by all the AI hype and euphoria. Because despite what the LinkedIn echo chamber might lead you to believe, outside the world of AI-obsessed digital marketers, the shiny sheen of AI-generated everything is already wearing off in many other corners of the world. 

What people are craving right now is what's real. They want to buy from humans whose demonstrated values they align with. So yes, hedge with AI, absolutely. 

But remember to lead with your heart, your humor, and your humanity. And yes, your humility as well. Because at the end of the day, humans want nothing more than to connect with another human, a real human being with a real beating heart. 

And that is something AI cannot replicate. At least not yet. Okay, I'll leave you with that for today. 

Have a great rest of your week. Remember to hug the ones that you love. And until next week, I'll see you again soon. 

Take care.