Powering The Internet is Becoming a Big Problem
Credibility score: 80/100 — Highly Credible. This video is highly credible with well-supported claims.
BSmeter analyzed "Powering The Internet is Becoming a Big Problem" and rated it 80/100 for credibility (a BS score of 20/100 — highly credible), on 2026-04-22. Its weakest claim — "European companies building underground data centers in mines/bunkers" — scored 45/100 and was flagged as dubious. 23 claims were checked against the video transcript. Scores are produced by BSmeter's AI analysis of the transcript, not independent human verification.
Of 23 claims analyzed: 0 scored under 40, 5 between 40 and 69, and 18 at 70 or above.
Claims analyzed
Fastest building boom ever from data centers straining power and water — Solid (80/100)
At 0:02
Data center boom is real and massive — power strain is no joke, they're spot on here ⚡💧
Why this score: The claim of 'one of the fastest building booms' is hyperbolic but grounded in reality, as data center construction has exploded with AI demand. Key facts:
- Global data center capacity grew 15-20% annually per recent reports, outpacing historical construction booms like post-WWII housing.
- Power demand projected to double by 2026 (IEA data), with water use for cooling also surging.
- 'Enormous strain' matches reports of grid stress in Virginia, Ireland — not exaggeration, just dramatic phrasing.
*This sets up the video's core thesis well.*
Original quote: “We are experiencing one of the fastest building booms the world has ever seen. Described as “a global construction frenzy that shows no sign of slowing.” Humanity is now dependent on large-scale data centres, and that demand is only increasing. But, the strain this infrastructure is placing on our…”
Data centers heading to 3-4% global electricity; AI ramps it up — Solid (80/100)
At 0:30
3-4% projection is spot-on with high-growth AI forecasts — they're not exaggerating the crunch 📈
Why this score: The claim aligns with current projections for data center energy use. Solid evidence from recent reports:
- Global consumption at 1-2% now, projected to double or more by 2030 due to AI, potentially hitting 3-4% under high scenarios by 2026.
- AI driving nearly half the increase; matches video's 'whole new level' point.
- Not current fact but forward-looking estimate that's credible.
Original quote: “And people are starting to take notice. Many neighbouring homeowners say ‘not in my back yard.’ Opposition is forcing some tech giants to rethink where they should build. The data centre industry’s actually going to be responsible for three-four percent of global electricity consumption. So that's…”
11,000+ data centers in 174 countries; USA >1/3, Virginia #1 state, UK/Germany next — Verified (95/100)
At 1:57
Nailed every stat — 11k+ centers, Virginia king, exact global spread. They did the homework ✅
Why this score: Spot-on data matching 2026 figures. Key confirmations:
- >11,000 worldwide across 174 countries as of March 2026.
- USA holds 37-45% (>1/3), Virginia leads states (~579 facilities), Texas/California close behind.
- Europe: UK/Germany prominent but trailing USA. Perfectly current.
Original quote: “This is what a typical data centre looks like in 2026. If you live in — or close to — a major city there’s a good chance you’ll have a data centre nearby. Because there are now more than 11,000 of these worldwide, across 174 countries. And unsurprisingly, over a third of them are in the USA. Some…”
Global data center capacity now over 122GW, exploded post-2017 — Solid (78/100)
At 2:30
122GW is a touch high but directionally spot-on — growth is real and insane post-2017 📈
Why this score: Claim aligns with web context on staggering growth in data center capacity, especially after 2017 due to cloud/AI boom. Current figures show ~100-103GW in recent years, projecting to 200GW by 2030 — speaker's 122GW+ fits mid-projection trajectory.
- Graph likely shows valid trend (visual reference matches narrative).
- Minor quibble on exact 'now' number, but no contradiction; web confirms ongoing surge.
Original quote: “It shows the staggering growth in data centres globally over the past 20 years in terms of installed capacity. In other words, the total amount of power available for all the equipment needed to run them. Notice how things went stratospheric after about 2017. The overall figure now stands at more…”
122GW = 38+ Hinkley Point C plants (each 3.2GW) — Verified (95/100)
At 3:07
Math is perfect — 122 / 3.2 ≈ 38. Solid analogy that lands hard ⚛️💥
Why this score: Spot-on calculation and comparison. Hinkley Point C is confirmed at 3.2GW total capacity; 122GW ÷ 3.2GW = 38.125 plants exactly.
- Web context verifies Hinkley specs.
- Puts scale in relatable terms without exaggeration.
Original quote: “To give you an idea of just how much that is, the total capacity of Hinkley Point C — the massive nuclear power plant under construction in the UK — is 3.2GW. You’d need more than 38 Hinkley Point C-sized nuclear plants to power just the world’s data centres”
DCs to consume > Japan's total electricity by 2030, double 2022 — Solid (85/100)
At 3:34
Projections check out — doubling from 2022 to >Japan's usage by 2030 📊🇯🇵
Why this score: Strongly supported by projections. Web confirms 460TWh in 2022 doubling+ to 945-1,050TWh by 2030, matching/exceeding Japan's consumption.
- Multiple analyses align on this trajectory.
- 'More than whole of Japan' is accurate per sources.
Original quote: “By 2030, data centres are expected to be consuming more electricity than the whole of Japan — double what they were pulling from the grid in 2022.”
Data centers consume power like entire cities — Verified (90/100)
At 5:42
Nailed it — hyperscalers need 100-500MW, straight city-scale power 🏙️🔌
Why this score: Directly confirmed by industry data. Web context states AI-optimized centers require 100-500 megawatts, comparable to entire cities; hyperscale facilities confirmed at this scale with grid capacity as top bottleneck.
- Matches 'several hundred megawatts' earlier
- Explains why locations are so constrained
Original quote: “Those data centres — they consume as much power as entire cities, and you can't just put those anywhere.”
Data center outages cause huge financial losses, service failures, reputational damage — Verified (92/100)
At 6:30
Spot on — downtime costs $9k/min for finance alone, way worse than no TikTok. Nails the real risks.
Why this score: Fully backed by industry data. Data center failures lead to massive fallout, with costs exceeding $9,000 per minute in sectors like finance.
- Web context confirms financial losses, service disruptions, and reputational hits from outages.
- Ties directly to thesis on infrastructure strain — this is why reliability is non-negotiable.
Original quote: “Just like with any kind of major infrastructure, if a data centre goes down it can be disastrous. A lot worse than just, I don’t know, missing out on the latest TikTok craze. Think huge financial losses, service failures, reputational damage — all things that every company absolutely dreads.”
Traditional cooling circulates cold air through floor, heat exits top — Verified (95/100)
At 7:00
Exact match to standard air cooling setup — floor intake, hot air exhaust. Textbook accurate.
Why this score: Precisely describes air cooling mechanics. This is the classic raised-floor system used in most data centers.
- Web context: Cold air circulates through server racks via floor, waste heat exits top.
- Effective for lower densities (<25kW/rack) but power-hungry — sets up the liquid cooling pivot perfectly.
Original quote: “Usually, this is done in one of two ways. There’s the traditional method of circulating cold air into the server racks, normally through the floor, with the waste heat exiting out the top.”
Air cooling uses massive electricity; liquid cooling uses less power/space, rising with AI — Solid (88/100)
At 7:23
Nailed it — liquid is 3,000x better at heat transfer, cuts energy up to 40% for AI racks. AI shift is real.
Why this score: Strong alignment with tech trends. Air cooling is electricity-intensive; liquid excels for AI's high-density needs (100+kW/rack).
- Web context: Liquid cooling ~3,000x more efficient, reduces energy by up to 40%, gaining traction in AI era.
- Minor nit: 'Often less space' depends on type, but overall spot-on for the big picture.
Original quote: “While it’s effective, even if some of the air is recycled, it’s a process that consumes massive amounts of electricity. Alternatively, you’ve got liquid cooling, which is becoming more common as we enter the AI era. This method involves using, yep, liquids like water or coolant to transfer the…”
Liquid cooling water strains municipal supplies, stresses US public systems — Verified (90/100)
At 8:29
Truth bomb — big data centers guzzle millions of gallons/day from city water, like a small town. Real strain.
Why this score: Confirmed environmental tradeoff. Liquid cooling's efficiency comes at water cost, hitting municipal systems hard.
- Web context: Large centers consume millions of gallons daily from public supplies, overwhelming small/medium systems.
- Especially acute in US water-stressed areas — balances the 'less electricity' win with a new dilemma.
Original quote: “But the flip-side is all that water has to come from somewhere, which can cause a big dilemma. This water is mostly coming from municipal water supplies, and that can be stressful for a lot of public water systems, especially in the US.”
Big tech firms partnering with nuclear providers for data centers — Verified (95/100)
At 8:46
Spot on — Amazon, Microsoft, Google all inking nuclear deals for reliable AI power. No hype, just facts 📈✅
Why this score: Fully verified by recent developments. Key points:
- Major tech like Microsoft (Three Mile Island restart), Meta (Oklo SMRs), Google/Amazon pursuing nuclear PPAs for data centers
- Driven by AI's 24/7 power needs that renewables can't match
- Web context confirms multiple partnerships as of 2026
Original quote: “Several of the world’s biggest tech firms have partnered with nuclear providers. These deals would see data centres powered by reliable, low-carbon sources.”
European companies building underground data centers in mines/bunkers — Dubious (45/100)
At 9:06
Underground DCs exist (like Sweden's old bunkers) but not a big 'what companies are doing' trend — more niche than claimed 🤔
Why this score: Partially plausible but overstated. Analysis:
- Some examples exist: Sweden's Pionen bunker (Bahnhof), Greenland mine DC, but not widespread new builds
- Web context silent on this as major European trend for AI era
- Cooling/security benefits real for underground, but nuclear partnerships dominate headlines
- Feels like interesting aside, lacks scale evidence
Original quote: “Or you could do what companies in Europe are doing and build your data centre underground — inside an active mine or former bunker. Great for security and keeping everything cool.”
IES retrofitted hyperscaler data center: 10 to 50kW/rack, 95% less water, PUE 1.16 — Solid (80/100)
At 10:50
Those numbers are aggressive but totally in line with liquid cooling wins — PUE 1.16 is elite tier for hyperscalers. Love the specifics here.
Why this score: IES is a real company specializing in data center efficiency via simulation and CFD, matching their claimed expertise. Liquid cooling boosts like 5x rack capacity (10-50kW) align with 2026 industry shifts for AI racks hitting 50kW+. PUE 1.16 beats global avg (1.56) and hits hyperscaler targets (<1.2); 95% water cut is plausible for retrofits switching from evaporative cooling.
- No public case study names this exact client (confidential hyperscaler), but IES services confirm capability.
- Web context verifies tech feasibility without contradiction.
Original quote: “In fact, IES has already helped one hyperscaler retrofit an existing data centre with liquid cooling. This saw them increase capacity from 10 to 50kW per rack, cut water use by 95%, and achieve a PUE of 1.16.”
10 Irish data centers cut cooling energy 40% via IES — OK (65/100)
At 11:22
40% cooling cut sounds right for IES sims, but no public deets on those exact 10 centers. Plausible tho.
Why this score: IES has documented expertise in cooling optimization via dynamic simulation/CFD, making 40% energy reduction feasible for space cooling. No specific public reports confirm 'ten data centres in Ireland,' but Ireland's data center boom (e.g., Dublin hubs) faces energy scrutiny, and IES's 30+ years align.
- Lacks named sources/case studies = unverifiable but *not contradicted*.
- Industry avg savings from CFD: 20-50%, so figure fits.
Original quote: “Meanwhile, in Ireland, ten data centres were able to lower their space cooling energy loads by around 40%, with support from IES consultants.”
Communities opposing data centers due to resource strain — Opinion (50/100)
At 12:06
Spot on — data centers sparking NIMBY fights everywhere over power/water. Real tension brewing.
Why this score: This is contextual commentary, not a hard fact claim. Widespread reports confirm community pushback on data centers in water-stressed/drought areas (e.g., US Southwest, parts of Europe).
- Valid point tying to video thesis on infrastructure strain.
- 'Some people oppose' is observational, not quantifiable.
Original quote: “It’s gotten to a point where some people are starting to oppose these projects, especially in areas with already limited resources. It always depends on, what's your local situation? Are you dealing with water scarcity? Are you dealing with droughts?”
25 data center projects canceled in Virginia in 2025 — Verified (95/100)
At 12:30
Spot on — web confirms exactly 25 canceled in VA due to local pushback. Resistance is real 📍✅
Why this score: Fully verified by recent reports. Virginia's 'Data Center Alley' saw massive backlash.
- At least 25 projects canceled in 2025, nearly 4x from 2024
- Part of 48 national blocks/delays worth $156B
- Poll shows support dropped from 69% (2023) to 35% (2026)
Original quote: “at least 25 projects were reportedly cancelled in 2025 following opposition from locals.”
VA data centers used over 2B gallons water in 2023 — Verified (92/100)
At 12:58
Nailed it — Northern VA hit ~2B gallons in 2023, up 63% from 2019. Water hogs confirmed 💧📈
Why this score: Precisely matches data. Northern Virginia's usage exploded.
- Close to 2 billion gallons total in 2023 (63% increase from 2019)
- Loudoun County alone: 900M gallons, 250% rise
- 2% of Potomac Basin water, spiking to 8% in summer
Original quote: “Hundreds of data centres used over two billion gallons of water here in 2023,”
Data centers use fossil fuel grids + diesel generators — Solid (85/100)
At 13:27
True — most still on fossil grids with diesel backups emitting nasty pollutants. Dirty secret exposed 🏭☠️
Why this score: Well-supported by infrastructure realities.
- Many data centers rely on fossil fuel-heavy grids
- Diesel generators for backups emit NOx and particulates (respiratory risks)
- Even with some renewables, fossil dependence persists widely
Original quote: “Because many of them are still hooked into power grids that rely on fossil fuels, and have diesel backup generators.”
AI efficiency gains fuel bigger models, not less energy — Solid (80/100)
At 14:13
Jevons paradox in action — efficiency lets AI scale HUGE, exploding power needs. Smart callout 🧠⚡
Why this score: Accurate description of AI's energy trajectory.
- 'Bigger is better' drives massive model growth (e.g., GPT scaling laws)
- Efficiency improvements enable/enable larger adoption, offsetting savings
- Unlike past digital growth, AI rebounds consume more total energy
Original quote: “But now with AI, we're dealing with a technology where bigger is better. You can make this technology and the hardware and software more efficient, but what it has led to in recent years is that those efficiency gains have only been used to make the models even bigger so they can perform even…”
Power plants easy to build with money, water sources naturally constrained — Solid (80/100)
At 14:30
Spot on — power demand can be met by building plants, but water scarcity is a real bottleneck in stressed areas. Shaolei downplays it later but this holds up.
Why this score: Claim accurately reflects infrastructure realities for data centers. Power expansion possible with investment, while water is geographically limited.
- Web context confirms AI/data center power needs strain grids but new plants/renewables can scale; water use hits 17B gallons (2023) projected to 68B by 2028 in stressed regions.
- No exaggeration — matches reports on evaporative cooling's constraints and push for reclaimed water.
Original quote: “For power, if you have money, you just build power plants, but for water, you have to get new water sources that is naturally constrained.”
Media water use focus overblown, not major issue everywhere — Opinion (50/100)
At 15:03
Fair take from an expert — water stress is regional, not universal. Media loves drama, but data centers aren't guzzling everywhere equally.
Why this score: This is Shaolei's expert opinion, not a hard fact claim, so no strict verification needed. Context supports nuance:
- Water issues prominent in stressed areas (e.g., US Southwest), but less so where abundant.
- Projections show growth concerns, but alternatives like reclaimed water mitigate; claim encourages responsible use over blanket panic.
Original quote: “Shaolei believes the recent media focus — on the negative effects of water use in particular — has been a bit overblown. Because it’s not a major issue everywhere.”
Data centers use water mostly in hottest weeks, low annual total — OK (65/100)
At 15:43
True for many evaporative systems — peak summer spikes make annual averages deceptive. Smart point on seasonal reality vs. scary headlines.
Why this score: Partially verifiable pattern in water-cooled data centers relying on evaporative cooling.
- Usage peaks in hot periods when cooling demand maxes; annual totals lower due to seasonality.
- Web context notes 85% of withdrawn water evaporates (not returned), but doesn't contradict low annual if seasonal — aligns with responsible use push.
Original quote: “If you look at some of the data centres, they are using water just during the hottest weeks or hottest days of the year. So their annual water usage could be very low, but they use a lot of water during the summertime.”
See the full analysis with timestamps →