The Ledger
Every statement that passed quotation and attribution checks. Mix any filter with any other: certainty 1/5, debate potential 5/5, or both at once.
why aren't all 44 resolved? a statement only gets an assessment when the public record can support or contradict it. opinions and what-ifs never can, and 3 checkable ones are still open, waiting for their date. predictions held up or didn't; assertions are supported or contradicted. on every card: ▮▮▮▮▮ certainty · ▮▮▮▮▮ debate potential. speakers are clickable
Feldman: New data centers will consume more power than past 50 years
“What we're talking about now are data centers that are in the next several years going to use more power than the previous 50 years on earth took.”
Cerebras CEO discloses $25 billion order backlog
“And so, you know, we have a twenty five billion dollar backlog.”
Feldman: Cerebras chip gains will exceed 2x over next 18 months
“And my view is in the next 18 months will be way over two X.”
Feldman: AI gives humanity a shot at ending cancer deaths
“There's a shot that our children, none of them, nor their people they love will die of cancer.”
Feldman: GPU builders have near-zero chance of outperforming NVIDIA
“So if you build a GPU, the odds that you're better than NVIDIA in our view are approximately zero.”
Feldman: Cerebras is 15 to 18 times faster than a GPU for OpenAI
“So when OpenAI uses us, we're 15 or 18 times faster than a GPU.”
Feldman: Historically more money is made post-IPO than pre-IPO
“I think historically more money is made after IPO than before. I think every single study shows that there is more money to be made both in percentage and in, in what we care about, which is absolute.”
Feldman: Small modular reactors probably won't power data centers for 3-4 years
“I think it's both the obviously the right thing to do and probably not the source of data centers for the next three or four years.”
Feldman: Space-based data centers are 8 to 10 years away
“I think it's in the eight to 10 year category, not in the three to five.”
Feldman: GPUs face a major inference bottleneck due to slow memory speed
“The GPU, for example, has a lot of capacity of memory, but it's really slow. All right, and that's a huge bottleneck in inference. It's why they can't be fast.”
Feldman: China has pushed ahead in open-source AI models
“I think on the other side, I think they have pushed ahead in the open model category. The open source model is an area where they've pushed ahead.”
Feldman: China leads US in expanding electrical grid capacity for AI
“I think because they're a top down economy, they were able to Make decisions like we're going to bring a huge amount of power onto our grid. We're going to modernize our grid. So they were able to bring on huge amounts of power, and that's something that we're…”
Feldman: Middle management layoffs stem from SaaS efficiency, not AI
“What I think has happened is this is the delayed impact of good SaaS tools. That what's happened is your ability to extend your reach as a leader and as a manager to stay abreast of what's happening. Your scope is much, much bigger. And so the role of middle m…”
Feldman: Extreme AI demand means no silicon compute goes unused
“Look, I think there is so much demand right now that, that, There is no silicon that will go unused.”
Feldman: Government red teaming of powerful AI technologies is reasonable
“With a powerful new technology, it certainly doesn't seem unreasonable to say, hey guys let's at least do some red teaming at the government, so we know our defenses can block this.”
Feldman: AI has passed every historical definition of AGI
“I mean, if you think about, oh, there was a Turing test, blew it away. I mean, you think about that, that any period of time, sort of 1015, 20, 3040, 50 years ago, we, any definition we would have previously put forward. We've blown past it.”
Feldman: Cerebras Wafer Scale Engine is 56x larger than Nvidia B200
“And so this is, say, 56 times larger than a B 200. And it's a four trillion transistor part.”
Feldman: Sam Altman and Ilya Sutskever accurately predicted current AI in 2015
“We talked to them in 2015, and what they were saying is now. It was unbelievable how right they've been.”
Feldman: Nuclear power is vastly more efficient over 20-30 years than alternatives
“Nuclear has, ah, over 20 or 30 years is vastly more efficient than any other power we know how to create.”
Feldman: Very few enterprises have meaningfully adopted AI workflows yet
“I think we are still really early in the demand for AI compute, and I think if you think about, you know, what portion of enterprises have really adopted AI in a meaningful way that has changed their workflow, it's tiny.”
Feldman: Memory market will take 18 months to digest surge, keeping prices high
“And it will take us about 18 months to digest. The prices will stay high.”
Feldman: Recursive AI systems compound early leads into winner-take-all outcomes
“Systems that are recursive, right, that build on themselves at rapid rates, have a huge winner-take-all feel, right, right? By getting ahead, you get further ahead, your iteration speed accelerates, And even small differences at the beginning are magnified ver…”
Feldman: Huawei clobbered the U.S. in developing world telecommunications
“They clobbered us in Africa. They clobbered us in the developing world. They absolutely ran the table.”
Feldman: Cerebras is navigating local regulations across 14 data center locations
“I think one of the things that kills a company like us is trying to grow extremely quickly. Is having to deal with different regulations in each of 14 localities we're trying to put data centers.”
Feldman: Enterprises are shifting from unconstrained token access to cost-optimized AI usage
“I think at first people opened up and said, everybody, as much tokens as you want. And in enterprises, there's no open loop. We don't give sort of any resource unconstrained to people. And now we're jumping on and saying, whoa, all right, these guys should hav…”
Feldman: Modern AI models no longer require precise prompt whispering
“And increasingly, it's understanding what your intent was. And if you have a chance to play with Fable or Five Six from OpenAI, increasingly, what, you don't have to get the prompt just right. You don't have to be a prompt whisperer.”
Feldman: Cerebras wafer-scale chip cost half a billion dollars to develop
“When one costs half a billion to make,”
Feldman: A massive data leak is inevitable in the AI era
“I think we can also know that there will be a massive data leak.”
Feldman: AI world models will provide behavioral insights by watching humans
“I think that's some of the things the world models are gonna bring us as they begin to watch human behavior.”
Feldman: Intel and AMD captured zero market share in mobile chip compute
“Share changed when cell phone compute emerged, and Intel and AMD who had fabs and the best architects got zero share, and it all moved to ARM.”
Feldman: Cerebras spent $500M to make its first chip
“Well, the first one cost us half a billion to make.”
Feldman: OpenAI is a Cerebras customer, but Anthropic is not
“And Claude is not a, Anthropic is not a customer, but we recently announced OpenAI.”
Feldman: Slow AI model latency causes immediate user churn
“While you're waiting for Claude, or you're waiting for ChatGPT, you get a coffee, or you poke around somewhere else, and you've lost the customer. That's the cost of being slow, is the customer has gone somewhere else.”
Feldman: Cerebras announced 750-megawatt deal with OpenAI
“What we announced was, 750 megawatts, is what was announced.”
Feldman: Hydro is world's cheapest power source, followed by natural gas
“So the cheapest power in the world is, is Hydro. Without question. After that is natural gas.”
Feldman: Cerebras was among the first production AI systems using water cooling
“We were some of the first production AI systems to use water.”
Feldman: Funding local infrastructure would have prevented data center pushback
“If instead you'd have gone in and said look, great. We, we're going to be good citizens. We're going to be big taxpayers here. Let's build more schools. We can build a school for you. It's a rounding error in the cost of this facility. We're going to make a bu…”
Feldman: Santa Clara houses six of the world's top ten chip teams
“Within a few square miles of Santa Clara you had Intel, you had AMD, you had NVIDIA, you have our team, you have ARM, you have one of ARM's great teams. You have maybe six. You have six of the world's great 10 chip teams.”
Feldman: Cerebras runs Qwen, Kimi, and custom models for GSK and G42
“We run today, we run GLM, we run Kimi, we run the Quen set of models, and we run OpenAI's models, the closed source ones, we run Models for, say, GlaxoSmithKline, which they wrote and developed. We run models for our partner in the UAE G-forty-two, and MBZ-UAI…”
Feldman: On-premise Cerebras systems cost $1M to $1.5M
“On premise, they're about a million, million and a half, depending on how you have it configured.”
Feldman: Cognition uses Cerebras to power its AI coding engine
“So we have customers, ah, like Cognition, who use us to power their coding engine.”