annashay 
Hardware delays, rising electricity prices, and labor inequities show the hidden toll of AI’s surge
The global Artificial Intelligence (AI) race has entered a paradoxical phase where trillion-dollar valuations and euphoric investor sentiment sit side by side with power shortages, rising inequality, and existential warnings from leading scientists. While the AI revolution promises transformative productivity and automation, the scale and speed of its infrastructure buildout are straining the world’s hardware, energy, and ethical limits.
The financial bubble no one wants to call a bubble
McKinsey estimates that AI-related data centre construction will require an astonishing $6.7 trillion in capital expenditure by 2030. Yet, according to Bain & Company, AI companies would need $2 trillion in annual revenue by that year just to justify that spending. In comparison, total AI revenue in 2025 stood at roughly $50 billion, which is a fraction of what’s being invested.
To put that into perspective, the industry is now investing seven to eight times more than its current income. A ratio far exceeding the 4x multiple seen during the early-2000s fibre-optic bubble.
Market analysts compare the AI rush to the dot-com era’s overbuild, warning that extraordinary capital inflows are concentrated in a sector whose unit economics remain unproven.
OpenAI, for instance, is targeting $13 billion in revenue but is projected to post losses of $8.5 billion in 2025, while a recent MIT report found that 95% of companies using generative AI have achieved zero return on investment.
Put bluntly: the world is investing many times more into AI infrastructure than AI today earns. You can think of it like building a vast highway network for cars that don’t yet exist in sufficient numbers. Expensive roads with too few paying drivers.
Even OpenAI’s CEO Sam Altman conceded that the frenzy might be unsustainable. ‘It might all be a bubble,’ Altman said. ‘Bubbles happen when smart people get overexcited about a kernel of truth.’
When people in the industry warn ‘this might all be a bubble,’ they’re not using scare language, they’re pointing to a classic mismatch. Big up-front spending with uncertain future customers.
Circular money loops and hidden debt
The financial engineering behind AI’s growth is equally dizzying. Analysts describe the sector’s funding model as “circular financing”. Picture a loop: chipmakers invest in AI firms, the firms use those chips, and the chipmakers then count the resulting purchases as revenue. That cycle can mask real risk.
The term “Ouroboros” (a snake eating its own tail) has been used to describe this loop. Nvidia, for example, pledged up to $100 billion in investment to OpenAI, contingent on OpenAI buying millions of its chips. The company also invested in Coreweave, a cloud operator that buys its hardware, and even guaranteed payments for any idle computing capacity if Coreweave couldn’t find enough clients.
Economists have likened this arrangement to an extension cord plugged into itself, where the numbers keep spinning even if true end-user demand is uncertain.
To make things appear financially cleaner, companies are routing debt through special purpose vehicles (SPVs) which are separate legal entities that keep liabilities off their main balance sheets. While this accounting technique is legal, critics warn it obscures how much debt the industry is actually carrying.
Deutsche Bank analysts say bluntly: ‘If it wasn’t for AI spending right now, we’d be in a recession.’ The implication, as Altman himself acknowledged, is that governments may eventually act as the ‘insurer of last resort’ if the sector falters.
The hardware crunch: when demand eats the supply chain
The AI boom isn’t just digital, it’s intensely physical. Building and running vast data centres requires mountains of hardware, from chips and storage drives to memory and cooling systems.
Severe shortages: Enterprise-grade hard drives now face delivery delays of up to two years, forcing hyperscalers (hyperscalers are simply very large companies that run massive server farms: think Amazon Web Services, Microsoft Azure) to switch to QLC NAND-based solid-state drives. Those too are booked through 2026, according to industry sources.
Skyrocketing prices: DRAM memory kits have doubled in price, while NAND manufacturers such as SanDisk have raised prices by 50%. Manufacturers are diverting production to AI customers “willing to pay the big bucks,” squeezing out smaller firms and consumers.
Changing focus: Nvidia’s shift from gaming GPUs to $40,000 AI accelerators underscores how lucrative and exclusive the market has become.
QLC NAND refers to a type of flash storage that packs more data into each memory cell (higher capacity) but is generally less durable than other types like choosing a bigger, cheaper suitcase that wears out faster
The high cost of running intelligence
AI is also running headlong into an energy bottleneck. OpenAI’s planned capacity of 23 gigawatts which is equivalent to 23 nuclear power plants gives a sense of scale. A gigawatt is a billion watts. Translating that to everyday life, it’s roughly the output needed to power hundreds of thousands of homes.
Microsoft CEO Satya Nadella has acknowledged that the bottleneck is no longer chips but ‘a lack of power to accommodate all of those GPUs.’ These chips are the computational engines that train AI models, consuming immense electricity and generating heat that demands constant cooling.
That thirst for energy is cascading through local economies. US electricity prices have nearly doubled over the past three years, and data centres are consuming ‘tremendous amounts of water’ for cooling, putting them in direct competition with communities and agriculture.
In one striking example, an XAI data centre in South Memphis reportedly operated on gas turbines without emission controls, leading to higher asthma-related hospitalizations nearby.
The human and ethical cost
Beneath the high-tech veneer lies an uncomfortable truth about who benefits and who doesn’t. AI’s wealth is concentrated among a handful of companies and executives, while the invisible labour that makes it all possible earns a fraction.
While Silicon Valley engineers make six-figure salaries, data workers in places like Kenya or Eastern Europe can earn as little as $2 an hour tagging or cleaning datasets, work that is crucial for training AI models.
Then there’s the question of who owns the data. AI firms like OpenAI and others have been accused of using copyrighted books, songs, and art without permission. Some European officials have even floated temporary GDPR exemptions to allow AI training, a move that privacy advocates call a dangerous precedent.
AI and the human question
Beneath it all, a pressing question remains: what if the technology outpaces human control?
Nobel laureate Max Tegmark, often called the “Godfather of AI safety”, warned that the next species AI could wipe out is “us.” A survey of AI researchers found they assign an average 16% chance of human extinction due to AI. Odds that are worse than playing Russian roulette with a revolver.
Yet not everyone subscribes to the doomsday view. Meta CEO Mark Zuckerberg has argued that slowing AI development could be riskier, saying the greater danger is that others will build unchecked if any single nation or company deliberately stalls progress.
BlackRock’s Richard Rieder offers a financial counterpoint, noting that big hyperscalers like Microsoft and Google have “robust cash flows” and trade at reasonable multiples around 22 to 26 times earnings. In his view, these are profitable, well-capitalized firms, not a repeat of 1999.
Still, many experts caution that the AI boom’s foundation rests on continuous exponential growth and if that trajectory falters, the entire system could wobble.
As some researchers put it, the AI revolution feels like standing on a rocket: the outcome could be orbit or catastrophe, and the difference depends on how prudently the ride is managed.
