AI hype is cooling, funding is getting weird, and 95% of projects still fail. Here’s why that might be the best news yet.
AI was supposed to change everything overnight. Instead, search interest is dipping, investors are side-eyeing their portfolios, and the internet is roasting AI-powered mattresses. Let’s unpack why the buzz is fading—and why that could actually be healthy.
When the AI Buzz Starts to Flatline
Remember when every headline screamed that AI would solve everything from climate change to your grocery list? That fever pitch peaked earlier this year, and now the charts are telling a different story. Macro strategist Spencer Hakimian dropped a sobering graph showing search interest and investor buzz cooling off fast. The same stocks that mooned on AI promises—think NVIDIA and friends—are wobbling as markets digest the difference between demo-day magic and real-world ROI.
Why does this matter? Because hype cycles don’t just move markets; they move policy, hiring, and even what gets built next. When enthusiasm outruns results, capital gets misallocated, workers get laid off, and regulators start sharpening their knives. The current dip isn’t necessarily a crash—it could be a healthy correction that separates signal from noise. But it’s also a moment when skeptics gain volume, asking whether we’ve been funding solutions or just fancier PowerPoints.
The debate splits along predictable lines. Optimists argue foundational models are still improving exponentially and that short-term volatility is the price of long-term transformation. Pessimists counter that most AI projects fail to deliver measurable value, burning cash and carbon in equal measure. Meanwhile, everyday users are left wondering if their next job interview will be with a chatbot or a human who survived the chatbot purge.
Mattresses, Jewelry, and Other Questionable AI Pitches
Gergely Orosz, the sharp-tongued author behind *The Pragmatic Engineer*, recently went viral for calling out “peak AI hype.” His thread highlighted eyebrow-raising funding rounds: a mattress startup raising millions to “fix sleep with AI,” another pitching AI-embedded jewelry. These aren’t moon-shot labs chasing general intelligence—they’re lifestyle brands slapping machine-learning stickers on ordinary products.
The pattern feels familiar if you lived through the dot-com or crypto booms. Investors, afraid of missing the next big thing, throw money at anything with “AI” in the pitch deck. Founders respond by retrofitting algorithms onto problems that never needed them. The result? A graveyard of over-engineered toasters and $500 smart water bottles.
But there’s a twist this time: the stakes are higher. Unlike the early web, AI consumes massive energy and data, amplifying both environmental and privacy concerns. When a smart mattress fails, it’s not just a bad product—it’s a data breach snoozing beside you every night. Critics argue this wave of gimmicky applications distracts from genuine challenges like algorithmic bias, labor displacement, and surveillance creep.
Still, defenders say experimentation is the price of progress. Every silly idea tests a boundary, and sometimes the boundary moves. The question is whether we’re iterating toward utility or just burning venture capital as performance art.
From Hype to Hard Questions: What Comes After the Gold Rush
So where does this leave us? Edward Ongweso Jr. summed it up bluntly: the AI bubble hasn’t burst yet, but it’s “hype, delusion, and PR” all the way down. Meanwhile, fresh MIT data shows 95% of AI projects still fail to deliver measurable value, echoing older Harvard studies that put the failure rate at 80%. Those numbers aren’t just academic—they translate into pink slips, shuttered startups, and skeptical boards.
Yet NVIDIA’s latest paper offers a counter-narrative. Instead of bigger Large Language Models, researchers propose lean Small Language Models (SLMs) powering agentic systems. The pitch: less energy, lower bias, faster deployment. It’s a pivot from brute-force scale to surgical precision, and it could deflate the current hype without popping the entire sector.
What happens next depends on which story wins. Scenario one: the SLM approach catches on, venture dollars flow toward focused, high-impact tools, and AI earns a reputation for solving real problems. Scenario two: the gimmicks keep coming, failures mount, and regulators step in with blunt instruments that slow everyone down.
For readers watching from the sidelines, the takeaway is simple—stay curious but skeptical. Ask what problem an AI product actually solves, who benefits, and what it costs beyond the sticker price. Because the next wave won’t be announced by a press release; it’ll be proven by paychecks, power bills, and privacy policies that finally make sense.
