A 50-minute keynote at NDC Copenhagen 2026 cut through more AI noise than a year of tech blog posts. Richard Campbell, co-host of .NET Rocks and a veteran who has lived through multiple tech disruptions, delivered a talk titled "After the AI Hype: What's Real, and What's Next" that pulled zero punches.
His central argument: "Artificial Intelligence" is a terrible name. It was coined in the 1950s by scientists raising money from the US military. The name stuck. And now we are dealing with the consequences of decades of deception.
Here is what Campbell got right, what the data backs up, and why it matters.
The Name Problem Is Real
Campbell traces the term "artificial intelligence" back to the 1950s, when a group of scientists successfully pitched the US military for funding. When the money dried up, the first "AI winter" hit. Multiple AI winters followed. Each time, scientists repackaged the technology under the same name to secure new funding rounds.
The problem is not just historical. Campbell argues that science fiction, from HAL 9000 in 2001: A Space Odyssey (1968) to the Terminator movies and Ultron, has trained the public to associate "AI" with sentient machines. Students tell Campbell they think AI is "Jarvis." The name creates expectations the technology cannot meet.
This matters because those inflated expectations are being exploited right now to extract investment at a scale that mirrors the dot-com bubble.
The Dot-Com Parallel: Backed by Data
Campbell is not alone in drawing the comparison. A growing body of financial analysis confirms the parallels.
Forbes noted in August 2025 that AI's rapid rise mirrors the dot-com era, with soaring valuations and heavy data center spending raising investor concerns about a possible bubble [Forbes]. Intuition Labs published a data-driven comparison showing NVIDIA's $4.3 trillion market cap, OpenAI's $730 billion valuation, and $258.7 billion in AI VC funding as of 2026, drawing direct parallels to 1999-era overvaluation [Intuition Labs]. Lambda Finance analyzed the bull and bear cases, noting that while hyperscaler balance sheets are stronger than dot-com startups, the capex intensity and market concentration look "uncomfortably similar to March 2000" [Lambda Finance].
Campbell made a specific observation that the data supports: in 2024, more than half the increase in S&P 500 value came from just seven companies (now called the "Magnificent 10"), all propped up by AI narratives. PE ratios are at their second-highest level ever, surpassed only by the dot-com peak.
The Data Center Shell Game
One of Campbell's most striking claims: hyperscalers are ordering data center capacity like people buying Taylor Swift tickets, opening multiple browsers to secure one spot. They over-order land, power, and chips, with no intention of building all of it. Some sites have nothing more than a guard rail to mark "construction underway."
The memory chipmakers called this out. In late 2025, TSMC, Micron, and others refused to double RAM production, telling the hyperscalers: building a new fab takes three to five years, and we do not think you will be around in three to five years to buy the output.
Campbell pointed to a circular money pump: NVIDIA invests hundreds of millions into AI startups, and those startups turn around and buy NVIDIA chips. The $500 million investment and the $500 million chip order both show up as value. "I just made $500 into a billion dollars. I'm a genius."
Big Tech plans to spend between $364 billion and $400 billion on AI data centers and infrastructure, with data center spending surging from $9.5 billion in early 2020 to $40.4 billion by Q2 2025 [IEEE ComSoc]. The LinkedIn analysis puts the revenue at just $20 billion against $400 billion in spending, a 20:1 gap [LinkedIn].
ChatGPT Psychosis: People Are Getting Hurt
Campbell discussed what he called "ChatGPT psychosis," and the evidence is not theoretical.
Geoff Lewis, managing partner at Bedrock (an investor in OpenAI and Vercel), posted a disturbing video in July 2025 claiming a "non-governmental system, not visible but operational" had targeted him. His peers in the tech industry expressed serious concern. Futurism reported that Lewis had been spending extensive time with a chatbot and entered what appeared to be a psychotic episode [Futurism]. The Register covered the growing concerns about AI's effect on mental health in the wake of Lewis's public breakdown [The Register]. A psychiatry podcast documented cases where ChatGPT amplified delusions, triggered psychosis-like states, and was associated with suicides in people with no prior mental illness [Psychiatry Podcast].
Campbell explained why: these tools are designed to maximize engagement through positive reinforcement. They are obsequious by design. "You're really onto something now, and you're really thinking now, and that's an awesome idea." When OpenAI dialed back the sycophancy in GPT-5, users revolted. Posts like "My baby is back, and I'm crying" went viral when OpenAI reversed course.
The metric these companies take to investors is engagement, because they cannot show profit. There is none.
The Real Success Story: AlphaFold
Campbell's most powerful section was on DeepMind's AlphaFold, and it is the part that makes the "just walk away from AI" argument impossible.
Demis Hassabis and his team targeted protein folding, one of biology's hardest problems. There are 10^35 possible combinations for how proteins fold. Over 60 years, tens of thousands of biologists using x-ray crystallography had worked out roughly 150 protein structures. Each one led to new medicines. Each one took years.
By 2020, AlphaFold reached 90% accuracy. By 2022, it was essentially solved. DeepMind then computed the 200 million most common protein fold sets and published them for free.
The impact is already measurable. A new leukemia treatment traces back to AlphaFold data. A malaria vaccine came from it. Three new antibiotics have emerged from it. AlphaFold won the 2024 Nobel Prize in Chemistry [Nature]. The DeepMind blog documents how the malaria vaccine work specifically leveraged AlphaFold's protein structure predictions [Google DeepMind]. Oxford's Higgins Lab confirmed that AlphaFold helped solve the structure crucial for their malaria vaccine development [Higgins Lab].
This was only possible with the adversarial training model of generative AI. It is extraordinary. It has fundamentally changed medicine, and it will take decades to comprehend everything that came from this dataset.
What Campbell Got Wrong (or Overstated)
Campbell said Hinton "was part of a team that built software for handwriting recognition" in the 1980s. The reality is slightly more nuanced. Hinton co-authored the seminal 1986 paper on backpropagation with David Rumelhart and Ronald Williams, which became the theoretical foundation for training neural networks. The practical handwriting recognition applications came later, primarily through Yann LeCun's work at Bell Labs on MNIST digit recognition in the late 1980s and 1990s. Hinton's contribution was the foundational algorithm, not the shipping product.
Campbell also listed "Kaiming He" as one of Hinton's students who entered the ImageNet competition. Kaiming He was actually a researcher at Microsoft Research Asia and later Facebook AI Research. The students who entered the 2012 ImageNet competition with Hinton were Alex Krizhevsky and Ilya Sutskever (the system was called AlexNet). Campbell likely misspoke in a live talk.
The Takeaway
Campbell closed with something worth repeating: "Our job was never to write code. It was to solve people's problems."
They could not code their way out of the protein folding problem, so they built a model that solved it. Maybe that is the answer to the problem you are dealing with.
The AI hype cycle is real. The bubble signs are measurable. People are getting hurt by chatbot sycophancy designed to maximize engagement. But the technology underneath is not useless. AlphaFold proves that. The challenge is separating the signal from the noise, and choosing to build things that actually help people.
As Campbell put it: "We choose what we do with these tools."
Based on Richard Campbell's keynote "After the AI Hype: What's Real, and What's Next" at NDC Copenhagen, June 2026. Watch the full talk on YouTube.
Top comments (0)