Data Centers: 11,000 Facilities Amid Financial Doubts
Reports suggest 11,000 global data centers are built for demand 50 times higher than actual usage.
English · Original discussion in Spanish · Published
Data Centers: 11,000 Facilities Amid Financial Doubts
A quiet figure circulates: 11,000 operational data centers worldwide, according to public maps tracking these facilities. Most people don't know they exist, haven't seen them, and couldn't explain their purpose. Yet, this infrastructure powers everything from ChatGPT to the latest mobile assistant. The recent debate isn't about whether AI works, but who pays the electricity bill, who funds construction, and what hides behind a deployment far ahead of real demand.
What They Are and Why No One Talks About Them
A data center is an industrial facility with thousands of processors running at full capacity, cooled day and night, connected to the grid like a heavy factory. A common comparison is Cantabria: when one was proposed in that Spanish region, its estimated energy use matched the entire area's consumption. The project stalled. That's the pattern: announced, financed, then many fail when real numbers are crunched.
Three suspicions recur. First, capital concentration: multi-billion projects only a few tech giants can afford. Second, environmental impact: massive electricity use, cooling water, and noise pollution. Third, true purpose: speculation ranges from facial recognition to social scoring, though no public proof exists of large-scale implementation.
Circular Financing: Building for 50x Demand
Here is the unsettling data. Technical forums calculate real chat inference: ChatGPT processes ~2.5 billion daily queries, each costing ~0.34 Wh. This equals ~35 MW continuous power. Adding Gemini, Claude, Grok, and enterprise APIs, global chat inference totals 350 MW to 1 GW.
Conversely, Amazon, Google, Microsoft, and Meta plan capacity for 50 times that demand. There are no clients. No use. It's a race to accumulate compute before competitors do, hoping someone will need it later. Like 90s malls built en masse, now struggling. Here, the asset isn't a storefront, but a silicon-filled building consuming as much as a mid-sized city.
Energy Problem: Renewables Don't Solve Peaks
Data centers need 24/7 base load power, not wind turbines that spin intermittently. This requires backup: gas combined-cycle plants, diesel generators, or ideally nuclear. Spain's current energy policy doesn't offer competitive stability, so projects move to Portugal or Segarro. Meanwhile, China peine its largest coal plant ever to secure supply for industry and AI, without the international environmental outcry seen in Europe over similar gas projects.
The paradox is known: decarbonization is demanded, yet facilities cannot run on intermittent sources alone. Result: Oracle runs data centers on gas, and others do the same quietly. Cheap, stable electricity is the real bottleneck, not chips.
Espionage, Regulation, and Political Reaction
The US administration drafted a ban on Chinese optical transceivers to protect data centers from espionage. This is the visible face of a tech war fought in wiring, not models. Simultaneously, a US citizen movement against these facilities has emerged, which debates suggest transcends left-right divides: perhaps the only point where broad sectors agree. Reaction also targets proliferating urban surveillance cameras lacking public debate.
Industry response, per debate accounts, is direct: offering communities $10,000 checks, funded police departments, and lower trash bills. Message clear: if you don't want a data center, we'll buy you off. If you still refuse, the project moves elsewhere.
AI That Doesn't Replace Anyone (Yet)
The labor apocalypse narrative is cracking. Practically, testimonies show companies integrating AI find they need human employees back to fix AI errors. Auto-generated docs are unreadable, AI-written tests miss real bugs, and migrated code piles up hundreds of utility files no one understands. AI doesn't replace workers; it complements, monitors, and forces them to clean up messes.
This creates financial discomfort. If AI doesn't replace staff, companies can't free margins to pay AI licenses. Debts remain. Revenue lags. The sector looks like a bubble needing buyers for the narrative before deflating.
Simulated Hacks and Narrative War
Recent months saw cross-accusations about supposed hacks on three AI firms being exaggerated or simulated to gain regulatory privileges and slow competition. Theory circulating: executives exploit antiestéticar of "rebellious AI" to protect a trillion-dollar bubble. No independent confirmation, but the pattern—announcing existential risks near earnings reviews—invites skepticism.
Add "benchmaxxing": models optimized to crush standardized tests, not do useful work. Gemini 4, per complaints, is the latest example. Lingering question: Does AI have initiative, or is it just an advanced search engine with a conversational interface?
With these elements, the question isn't if AI changes the world. It's if the business model supporting 11,000 data centers survives the first adverse economic cycle. When capital withdraws, buildings remain, consuming light, processors on, with no one on the other end.
A quiet figure circulates: 11,000 operational data centers worldwide, according to public maps tracking these facilities. Most people don't know they exist, haven't seen them, and couldn't explain their purpose. Yet, this infrastructure powers everything from ChatGPT to the latest mobile assistant. The recent debate isn't about whether AI works, but who pays the electricity bill, who funds construction, and what hides behind a deployment far ahead of real demand.
What They Are and Why No One Talks About Them
A data center is an industrial facility with thousands of processors running at full capacity, cooled day and night, connected to the grid like a heavy factory. A common comparison is Cantabria: when one was proposed in that Spanish region, its estimated energy use matched the entire area's consumption. The project stalled. That's the pattern: announced, financed, then many fail when real numbers are crunched.
Three suspicions recur. First, capital concentration: multi-billion projects only a few tech giants can afford. Second, environmental impact: massive electricity use, cooling water, and noise pollution. Third, true purpose: speculation ranges from facial recognition to social scoring, though no public proof exists of large-scale implementation.
Circular Financing: Building for 50x Demand
Here is the unsettling data. Technical forums calculate real chat inference: ChatGPT processes ~2.5 billion daily queries, each costing ~0.34 Wh. This equals ~35 MW continuous power. Adding Gemini, Claude, Grok, and enterprise APIs, global chat inference totals 350 MW to 1 GW.
Conversely, Amazon, Google, Microsoft, and Meta plan capacity for 50 times that demand. There are no clients. No use. It's a race to accumulate compute before competitors do, hoping someone will need it later. Like 90s malls built en masse, now struggling. Here, the asset isn't a storefront, but a silicon-filled building consuming as much as a mid-sized city.
Energy Problem: Renewables Don't Solve Peaks
Data centers need 24/7 base load power, not wind turbines that spin intermittently. This requires backup: gas combined-cycle plants, diesel generators, or ideally nuclear. Spain's current energy policy doesn't offer competitive stability, so projects move to Portugal or Segarro. Meanwhile, China peine its largest coal plant ever to secure supply for industry and AI, without the international environmental outcry seen in Europe over similar gas projects.
The paradox is known: decarbonization is demanded, yet facilities cannot run on intermittent sources alone. Result: Oracle runs data centers on gas, and others do the same quietly. Cheap, stable electricity is the real bottleneck, not chips.
Espionage, Regulation, and Political Reaction
The US administration drafted a ban on Chinese optical transceivers to protect data centers from espionage. This is the visible face of a tech war fought in wiring, not models. Simultaneously, a US citizen movement against these facilities has emerged, which debates suggest transcends left-right divides: perhaps the only point where broad sectors agree. Reaction also targets proliferating urban surveillance cameras lacking public debate.
Industry response, per debate accounts, is direct: offering communities $10,000 checks, funded police departments, and lower trash bills. Message clear: if you don't want a data center, we'll buy you off. If you still refuse, the project moves elsewhere.
AI That Doesn't Replace Anyone (Yet)
The labor apocalypse narrative is cracking. Practically, testimonies show companies integrating AI find they need human employees back to fix AI errors. Auto-generated docs are unreadable, AI-written tests miss real bugs, and migrated code piles up hundreds of utility files no one understands. AI doesn't replace workers; it complements, monitors, and forces them to clean up messes.
This creates financial discomfort. If AI doesn't replace staff, companies can't free margins to pay AI licenses. Debts remain. Revenue lags. The sector looks like a bubble needing buyers for the narrative before deflating.
Simulated Hacks and Narrative War
Recent months saw cross-accusations about supposed hacks on three AI firms being exaggerated or simulated to gain regulatory privileges and slow competition. Theory circulating: executives exploit antiestéticar of "rebellious AI" to protect a trillion-dollar bubble. No independent confirmation, but the pattern—announcing existential risks near earnings reviews—invites skepticism.
Add "benchmaxxing": models optimized to crush standardized tests, not do useful work. Gemini 4, per complaints, is the latest example. Lingering question: Does AI have initiative, or is it just an advanced search engine with a conversational interface?
With these elements, the question isn't if AI changes the world. It's if the business model supporting 11,000 data centers survives the first adverse economic cycle. When capital withdraws, buildings remain, consuming light, processors on, with no one on the other end.
Also available in: Deutsch
Summary of a discussion on Burbuja.info - Foro de economía, actualidad y política., translated from Spanish and reviewed before publication. Read the full discussion (278 replies).
More summaries
-
China hikes tariffs to 125%, weaponizes yuan in trade war
China raises tariffs to 125% and devalues the yuan as a trade weapon. The debate: does it impoverish its population or shield its economy?
-
Why the US has no high-speed rail and Spain does
Spain has the world's second-largest high-speed rail network while the US has none: the 700 km between Boston and Washington still have no high-speed rail.
-
Chinese AI Challenges Western Tech Dominance
Chinese AI challenges Western tech giants with powerful open-source models.
-
IONOS cuts 450 jobs, Logroño fears for Arsys's 360
IONOS cuts 450 jobs from 3,800: Arsys, with 360 workers in Logroño, is unsure how many will be affected.
-
Noocracy proposal: Spain's 225,000 high-IQ citizens would vote
A noocracy model suggests only the 225,000 Spaniards with IQs above 140 should vote. It features rotation and claims to prevent corruption.
-
Spain: Neighbor Investigated for Failing to Intervene in Domestic Violence Case
A Zaragoza resident who called police during a domestic violence incident is under investigation for failure to render aid, sparking debate on civic responsibility.