20VC — Bloom Energy CEO KR Sridhar: Why We Aren't in an AI Capex Bubble

📋 한국어 요약 보기 →

"I'm a technologist — used to be a rocket scientist. When I started the company, if you said PNL or EBITDA, I wouldn't even know what that meant. We built the first five Bloom boxes ourselves with engineers — they went to Google and other places, working very well. Then we handed off to manufacturing and built the next 10, the next 20 — they were all going out in the field and failing spectacularly. I stopped production and called my board: I need help."

"Andy Grove came — he became a great friend and mentor. Manufacturing gurus. We packed three-ring binders full of data. Executive team on one side, advisors on the other. We're going to stay up all night. We sit down, and before I finish the first sentence Andy says: 'Can you stop for a minute. Can everybody else in the room leave — other than KR.'"

"It shocks me. I sit back: 'I'm trying to build a great device beneficial to people. I don't know what's wrong with me.' He says: 'You're such a smart guy. You built this technology that's absolutely amazing. If you walk the floors and talk to the people out there, they're going to tell you why it's not working because they don't understand what they're building. None of us know your product better than you. If you're smart enough to design it, you'll be smart enough to figure it out. Don't waste your time listening to everybody else and me. Go figure it out.' That was the best piece of advice I ever got."

What KR takes from that: not just "walk the shop floor" but empathy. "When you practice empathy by walking on the floor talking to the people building your product, they will tell you things that make the difference between a good company and a great company. You do the same thing with customers — you don't tell them why your product is great. You find out their pain and what problems they want you to solve. That's empathy."

"When I toast these days at executive dinners, I always remember: we're toasting this great champagne, but we're able to do this because there are technicians on the floor right now. Nobody is policing them — but their pride, their dedication, they're building an amazing product. Let's toast to them."

📋 한국어 요약 보기 →

Harry: are we in an AI infrastructure bubble?

"Anything going at that speed will have bumps. We don't know of anything else in human history of this scale and magnitude. But is the normal course after the course correction still going to be this phenomenal course? I absolutely believe so — because for the first time in human history we are manufacturing intelligence. When was the last time any person in any civilization said we have too much intelligence, let's stop?"

Harry: when intelligence becomes ubiquitous and abundant, what becomes valuable? "Wisdom." "AI is not going to give you wisdom. AI is not going to give you happiness. AI is not going to give you empathy. A bot giving you counseling is not the same as you and me sitting here talking, looking at each other, understanding facial expressions. Humans are social animals."

📋 한국어 요약 보기 →

"The single biggest bottleneck in introducing a new concept into society is the friction associated with new ideas. Very large infrastructure has been designed and regulated to move slowly — for good reasons in the past. We are entering an age where that becomes an impediment. I'm particularly sensitive being here on the European continent — you get A+ in friction." (Harry: "We had to get a permit to come up the stairs.")

Gavin Baker has argued that permitting actually helps by throttling supply to grow in lockstep. KR disagrees:

"China has amazing model development and a regulatory environment that's honestly not inhibiting them at all. If everything else were equal, what you're saying would be true. But the creative prowess, the innovative entrepreneurial spirit you see in the United States is still a major trump card. Human spirit will overcome whatever other obstacles. So I would say don't short the US. Don't short Silicon Valley."

On wealth concentration & "MAG7"/"MANGO": "I don't think governments taking equity is the right model. The core issue is — no new technology that's come along ultimately created less jobs. I don't think AI is going to be anything different. But the transition generation suffers — there's collateral damage. We told everybody 'learn how to code, or you don't have a future.' Then suddenly AI says coders have no future. There should be a construct — for political figures to figure out — to take care of those people."

"The top 20 most important companies — that list changes every 10 years in the US. Microsoft used to be considered a monopoly, no longer. Google was considered a monopoly, no longer. The only thing regulation can do is stop that innovative culture. If government took major equity and provided unfair access, by definition they're stopping startups with better ideas."

📋 한국어 요약 보기 →

"Are hyperscalers secretly becoming energy companies?" — "We're manufacturing intelligence. The input is electricity and data. The only real cost input is electricity. So electricity becomes the key factor. Since Edison, 150 years later, we're going to create a new construct for electricity — digital electricity, as opposed to mechanical-age electricity. More efficient, cleaner, more reliable. And when these large companies that have the resources allow that to happen at scale, it trickles down — Sub-Saharan Africa, Bangladesh — just like an airbag comes in an expensive car first, then every car within a few years."

On scaling: "Tens of gigawatts — we'll be single-digit percentages of supply chains that already exist for similar devices. When Steve Jobs showed his first smartphone, nobody said 'will you build 5,000?' It's in the millions. That's the supply chain we'll rely on. The company was built to light up the planet — that was the mission."

"It takes 12 to 18 months minimum to build a 2GW data center — copper, cooling, tradespeople all stressed. We can stand up that power faster than they can build the data center. Our constraints: Can they build it in time? Can we get permits? And gas supply needs to come in."

"What's the long-term vision? Wherever you live there's wind, solar, geothermal — you use that when nature gives it to you, bottle it as hydrogen on site, then recycle the hydrogen through something like Bloom for reliable abundant power. Self-reliant communities — Iceland or Sub-Saharan Africa — not depending on some logistic lifeline."

"In the immediate term, America is endowed with unbelievable natural gas — the cleanest, smallest carbon footprint molecule available. Along with Australia, Canada, UAE, Qatar — the free world — we should provide that gas to every country so they don't have to burn coal. Good should not be the enemy of best. Let's not let people who don't have our best interest supply energy to us. Let's bring free fuel from a free world to make more of the world free."

Harry: how criminal is it that we still buy so much energy from Russia? "Exactly. It is illogical. It's not rational and there's a better way." Energy sovereignty: "After food, it's energy sovereignty, period. Big wars have been fought for water. Big wars have been fought for food. Big wars are being fought for energy — and we can avoid that."

On power at the edge: "In a digital world, power outages are unacceptable — it's like not having air. Imagine surgery by robot relying on data and not having power. Driverless transportation in chaos — you cannot rely on a large data center far away with latency. A pole and wire running hundreds of miles capable of disruption by mother nature, cyber or physical attack cannot be the only lifeline. Power has to be at the edge."

"Why do people concentrate in cities? Access. Anywhere there was a waterway, rivers, coast, railroad intersections — that's where access was created by people who controlled it. If you break that using distributed [power] — true democracy. Any technology coming to the edge democratizes it. The only reason parents leave their idyllic village settings and move into poultry conditions in cities is so their kids have better access. Why not bring the access right where they live?"

📋 한국어 요약 보기 →

"Before AI came along we had 5–6 data center customers over 10–12 years providing reliable power — 4MW, 10MW, 15MW. Our very first data center application was 2013 with eBay/PayPal — PayPal's entire business is that data center, all transactions happen there. They didn't have power in Utah, didn't want to move location, so they relied on us."

"A couple years ago Oracle had a data center in Utah whose power systems were delayed. They came to us: 'Can you give us a temporary solution for a year?' Absolutely. We signed a contract — 90 days from signing we'd deliver 50+ MW. We delivered in 55 days instead of 90. They were thrilled. So that formed the basis: why look for other solutions when we have a great one?"

Three attributes that make Bloom ideal for data centers:

(1) Availability. "A 500MW turbine's availability over 20 years is in the low 90s — like a jet engine or car, it has to stop for maintenance. What happens during the other 10%? You can't turn down the data center. In the past you relied on the grid as backup — now there's no grid that's a gigawatt in size to back you up. So you have to put two turbines instead of one. With Bloom we're tiny modular power builders — 50 kilowatts each. You can hot-swap any one. Very similar to data center architecture — server blades and racks where one going down doesn't take the data center down."

(2) Solid-state ramping. "We can ramp up and down in milliseconds, which is what AI loads need. AI loads are like the brain — huge activity, pulse, lots of energy, then calms down. Same thing in an AI data center with GPUs. A turbine — like a car — even the fastest takes a few seconds 0 to 60 because it's mechanical. We're solid state. All the batteries they needed to provide this ramp — they don't need with Bloom."

(3) Lego-style scaling. "You cannot put a partial turbine and keep adding turbine blades as the data center ramps up. With Bloom — tell us what you need today, and as you're growing we keep adding Lego blocks."

Harry: how do I reconcile the world's biggest leaders telling me power is the biggest problem with you saying we can produce it at scale and edge? "It's a scale issue. You're seeing the beginning of the hockey stick. Electricity is a $5.5 trillion market; we did $2 billion last year. We have a long way to go. But power at the edge is going to be dominant. We're going to look back and say AI accelerated this. Two reasons: clean — out of sight out of mind tolerates dirty; at the edge you demand clean. And waste heat — wasted at large plants, used for heating/cooling at the edge."

📋 한국어 요약 보기 →

"Backlog is $20 billion, manufacturing capacity is 1GW today. End of this year — over 2GW publicly stated. Going forward — not just ramping from 2 to another big step function. Think of it like an analog dial — every month, every quarter, we keep adding capacity. Our factory is state-of-the-art production, state-of-the-art construction zone, and state-of-the-art innovation area all happening at the same time — we're innovating manufacturing processes thanks to AI."

"As a public company we haven't given forward guidance. But the promise: come put a firm deposit for power before you build your greenfield data center, we'll have that power for you. We will not be the bottleneck."

Customer concentration: "No. Signed up with multiple hyperscalers, multiple neoclouds. More importantly we have a robust commercial and industrial business — factories, Home Depots, Costcos, Walmarts. Two-thirds of Caltech campus is powered by us."

On Leopold Aschenbrenner: "First time I knew was when my 22-year-old daughter showed me all her friends were talking about it. I've met him since — very sharp, deep-thinking individual. The idea of distributed power coming to the edge is going to change city planning, access, the need for people to move from inaccessible area to metropolitan city just so the next generation has a better life. I don't think that's how 10 billion people should live."

What KR has most changed his mind on in the last 2–3 years: "Until this big switch happened, we were a lot more internally focused. Now we feel as a company we have a larger mandate than just building the company — also to speak about where the future can be: electricity for all, abundance for all. So I'm spending more time speaking about it."

On abundance vs zero-sum: "When you create abundance — like the intelligence abundance we're talking about — abundance changes the entire framing. You getting a larger piece of the pie is not coming at my expense. We just have to include all the people. Technology has been the best equalizer in the history of mankind — infant mortality, hunger, clean water, healthcare. There is probably no force in the world other than technology that has done that at such scale."

📋 한국어 요약 보기 →

Biggest lesson from John Doerr? — "To dream big."

One widely held belief about AI that's completely wrong? — "That it is going to diminish the number of jobs."

Most underrated person in the energy transition? — "The people who created horizontal drilling and figured out how to extract more gas and oil in a clean way as a bridge to the future — that entire sector is underrated."

Country best set up to be a power powerhouse in next decade? — "Still the US."

Is Europe slightly screwed? — "Europe is waking up faster than I would have expected because of AI. I'm optimistic hearing both from the EU and UK about the changes they want to make. The key — can they make changes fast enough?"

If you had unlimited money — what would you do that you're not doing today? — "Go to the rest of the world where they don't have upfront money and capital to buy our boxes — install them, collect later from the benefit."

What do you think a lot about but rarely see in the news? — "Creating more equal access. Equity is different — creating access is important. Giving everybody an opportunity. What they make of it is up to them."

Hardest time in 25 years? — "Post financial crisis. We had built a completely US supply chain — but the Big Three automotives failed, and they couldn't keep their promises to me. I had to within months find a Plan B for what was a strong Plan A six years in the making."

Kindest thing anyone's ever done for you? — "I was in China in a remote part. I had a bad accident — could have easily bled to death. Through my translator with few seconds left of consciousness I was like: get me a helicopter, get me to Hong Kong. The doctor — I vaguely remember his face — told the interpreter: ask him to shut up, I'm going to stitch him up, otherwise his body will be flown to Hong Kong. He stitched me up. That was a very kind thing — not listening to me."