From AI Frenzy to AI Fatigue
AI bubble skepticism is rising as enterprise AI fatigue sets in, with large organizations questioning whether expensive frontier models, escalating infrastructure spending and vague productivity promises can deliver reliable returns on investment instead of draining budgets and exposing valuable data without clear, measurable business outcomes. That mood shift was laid bare when Palantir CEO Alex Karp used a televised interview to accuse leading AI labs of irresponsible overselling, saying enterprises now expect to “waste [their] time with tokens, get no value, and watch vendors walk away with their IP and alpha.” His language reflects more than personal frustration; it captures a growing belief that the AI market oversaturation of glossy demos and incremental feature updates has outpaced real-world impact. The backlash against “tokenmaxxing” — pouring money into AI usage without hard metrics — shows buyers are no longer content to be the experimental playground for models that promise transformation but deliver little beyond cost and compliance headaches.
When Wall Street and Hyperscalers Start Worrying
Skepticism is no longer confined to a few contrarian executives; the financial system’s nerve center is sounding the alarm. The Bank for International Settlements warned in a late June report that an overheating AI bubble could burst and hit the global economy, comparing today’s capex frenzy to past manias where capital far exceeded what the industry could earn back. Hyperscalers are pushing that dynamic to extremes. Forecasts suggest Amazon plans north of USD 200 billion (approx. RM920 billion), Microsoft around USD 190 billion (approx. RM874 billion), Google USD 180 billion (approx. RM828 billion) and Meta USD 140 billion (approx. RM644 billion) in AI build-outs in a single year. One major cloud vendor with heavy AI exposure has already shed more than 40 percent of its share value in a month and detailed how badly it could be hurt if those AI bets fail to pay off. This is the purest expression of AI ROI concerns: even companies selling the picks and shovels are nervous the gold rush might not last.
Open-Source AI Competition Is Undercutting the Premium Story
While investors fret about overinvestment, a separate force is quietly smashing the old pricing logic: open-source AI competition from Chinese labs. Models such as DeepSeek, Qwen, Kimi and GLM now flood the market with free downloads that are cheap to run and increasingly hard to dismiss as second-rate. For three years, the dominant story was simple: the best AI lives behind a paywall, and enterprises will pay a premium for frontier performance. That premise is under serious strain when developers report paying about USD 10 (approx. RM46) for an hour of coding on a premium model versus less than USD 0.50 (approx. RM2.30) for roughly comparable output on DeepSeek. This is not a minor discount; it is a textbook disruption gap. These models already make up close to a third of usage on major aggregators, with Qwen overtaking Llama as the most downloaded open-source family in history, and even deep-pocketed platforms are probing whether cheaper open-source stacks can power parts of their own AI products. The oversaturated AI market now has a credible low-cost tier that can force every premium vendor to justify its price difference in concrete business value, not marketing slides.
Enterprises Want Control, Not Magic
Underneath the headlines, ordinary users are feeling the strain of the hype cycle. Bank analysts warn that hyperscaler capex is driving up component demand, making memory harder to obtain and even pushing up laptop prices for consumers because RAM supplies are squeezed. At the same time, many corporate buyers discover they are “paying for tokens that create no value,” as Karp puts it. Their everyday needs are not sci-fi: they want systems to summarise documents, translate reports, search internal knowledge bases, classify data, draft customer responses, assist with compliance, help with routine coding and automate repetitive workflows. In that world, the frontier model arms race matters less than predictable costs, data control and the ability to switch providers. No enterprise CFO wants to bet the company on a vendor that cannot pay its own bills and must continually raise new cash to survive. The shift from AI adoption frenzy to critical evaluation of actual business value is a rational defense against a market where technical opacity and aggressive pricing collide with unproven returns.
The Post-Hype Playbook for AI
The uncomfortable truth for AI vendors is that trust, not model size, is now the scarce resource. Executives are more vocal about fears that their institutional “alpha” could be commoditized or siphoned away, echoing warnings that entire industries risk seeing their knowledge turned into generic outputs sold back to them. At the same time, mega-projects like the multi-hundred-billion-dollar Stargate initiative — in which one cloud provider committed USD 300 billion (approx. RM1.38 trillion) of a planned USD 500 billion (approx. RM2.3 trillion) data center build-out — show how dependent some AI labs are on external financing to keep their ambitions alive. That fragility makes enterprises even more determined to seek AI sovereignty, open models, predictable pricing and clear opt-outs rather than handing over core data to closed systems. What comes next will not be an end to AI, but an end to uncritical AI buying: fewer pilots sold on hype, more projects judged on whether they cut real costs, protect IP and solve the unglamorous problems that actually keep businesses running.
