Why the Prophets of AI Doom Got It Wrong (For Now)
There is a truism in life: nothing lasts forever. This must eventually become true for the current generative AI technology wave as well. Eventually, gravity will rule the day—when some key promise in AI, announced by a globally famous technology CEO, falls flat, prompting skeptics to exclaim: “I told you so!”
However, certain observers have already started proclaiming that “AI Winter 2025” is on its way—citing hype fatigue, ballooning capital expenditures, and the apparent “scaling walls” of model architectures. However, gigantic forces are converging to ensure that the so-called AI Winter of 2025 looks more like a sweltering summer in the AI tropics.
Table of contents
The Mysterious Payback Model
The recent wave of doubt began when analysts added up the capital being poured into frontier AI models. Annual investments have crossed $100 billion and could skyrocket to $350 billion—with the first gigawatt-scale data centers already on the drawing board. The numbers simply don’t add up if you rely on traditional software business models.
Critics therefore ask: Where is the payback? The market sees massive R&D outlays from the likes of OpenAI, Microsoft, Facebook, and Google, but many are skeptical these companies can recoup those investments in a timely manner. Fueling the skepticism is the alleged “wall” that marquee AI players (e.g., OpenAI, Anthropic) appear to be hitting with scaling laws. Add in the widely publicized issues of “hallucinations” and fabrications, and it’s easy for doubters to predict an AI downturn.
Yet that dire view has been forced to the sidelines more recently, thanks to two phenomena:
- Artificial General Intelligence (AGI) systems that continue to make surprising strides.
- The burgeoning world of Virtual Employees (VEs) that promise an entirely new revenue and adoption model.
AGI Systems Become Real
Major AI lab researchers have long considered how to address the limitations of large language models (LLMs)—and they’re moving on from a purely “scale-it-up” philosophy to more creative architectures. Enter the so-called “slow thinking” or step-by-step AGI systems. These approaches blend LLMs with other algorithms that “chunk” tasks into smaller pieces, applying iterative reasoning until a solution emerges.
OpenAI’s new system, nicknamed “o3,” is one of the poster children of this approach. It purportedly solves math benchmarks at a level only achievable by a small number of humans, building on earlier iterative-thinking models like “o1.” Early users—research scientists among them—claim it has provided unique insights and critiques of their own work that they had not considered before.
If these developments are any indication, the frontier model makers are nowhere near slowing down. Indeed, like the leaps we witnessed in early 2023, we could see another exponential surge in AI’s capabilities in 2025, thanks to these new AGI techniques.
Virtual Employees Take Center Stage
Perhaps the more immediate commercial breakthrough is the proliferation of AI agents or Virtual Employees (VEs). These are software-based “workers” that perform entire job roles—like sales development reps (SDRs) or customer support agents (CSRs)—rather than just assisting with tasks. Powered by next-generation AGI systems, VEs can:
- Coordinate complex workflows using step-by-step reasoning.
- Accumulate knowledge from shared databases that are automatically curated as they work.
- Interface seamlessly with enterprise software—think Salesforce, cloud-based ERPs, and marketing automation platforms.
If you listen to Marc Benioff and other IT luminaries, they’re already talking about offering digital labor to run the very software businesses are buying. This shift in the conversation is telling. Once top technology execs seriously explore offering entire “virtual departments” to their enterprise customers, a new economic engine emerges—one that turns all that R&D investment into a subscription-based or usage-based service.
Why AI Winter Is Far Off
Here are some of the reasons why the AI Winter of 2025 won’t happen.
Enterprise Demand for Efficiency
Labor costs, talent shortages, and performance pressures drive businesses toward AI-driven automation. As soon as VEs show solid ROI—especially for highly repetitive or standardized tasks—adoption could scale much faster than historical tech hype cycles predict.
Exponential vs. Architectural Breakthroughs
Even if raw “scaling laws” for large language models appear to plateau, more “architectural” innovations (like slow thinking, multi-agent systems, or domain-specific AI frameworks) could unlock fresh leaps in functionality and reliability.
New Revenue Models
Instead of purely licensing AI software, tech vendors are shifting to usage-based or “digital labor” models, which may prove far more lucrative than standard software sales. These new models help justify the massive datacenter and R&D investments because they scale with usage—not just upfront license deals.
No Single Winter
Past AI winters often followed a pattern: hype, unmet promise, disillusionment. But we’re in a period where AI is already being used in mission-critical workflows—call centers, SDLC tools, content generation, code generation. The question is not whether it will be used, but how pervasive it will become.
Welcome to the Summer of AI
The notion that AI is about to hit a wall might have been more credible if we hadn’t already seen real-world traction: from ubiquitous chatbots to the slow-thinking AGI breakthroughs and the emerging Virtual Employees. Yes, nothing lasts forever—but it’s equally true that markets and technologies can evolve in ways that far outpace old assumptions.
When we consider AGI’s steady progress and the lucrative potential of VE-driven enterprise models, the grand “AI Winter of 2025” starts to sound less like a blizzard and more like a day at the beach. Instead, we may just be entering a season of accelerated innovation, fueled by the very forces some analysts once dismissed as hype.
