Jun 18, 2026 · 1h 0m · latent-space
Why AI Labs With Unlimited GPUs Still Fail — Anjney Midha, AMP
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Anjney Midha, CEO of AMP, examines why AI labs fail from cultural misalignment despite limitless capital, outlining a roadmap for horizontal compute grids, scientist-led governance, and applying frontier technology to fundamental scientific breakthroughs.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 5.8 Guest teaching 5.7 Guest disagreement 3.3 The hosts pushing back 3.3
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 36:28 Anjney forcefully dismisses VC assumptions about scientist founders
Anjney strongly rejects the notion that academics cannot be CEOs, contrasting superficial venture opinions with the intense leadership demonstrated by top published scientists.
Hardest push from the hosts ▶ 7:19 Shawn challenges the grid model versus full-stack integrationShawn directly questions why a third-party compute grid would be more aligned than vertically integrated labs like xAI and OpenAI.
Biggest teaching moment ▶ 1:21 Anjney outlines strict node and MFU utilization standardsAnjney educates on industrial infrastructure standards, pointing out that anything under 95% node utilization is treated as an outage in elite engineering organizations.
The host holds their own ▶ 13:28 Shawn cites David Luan and Google's internal marketplace failureShawn demonstrates deep technical context by explaining how internal credit prioritization at Google historically prevented central commitment to big models like GPT.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Infrastructure Realities: Cluster Utilization, Waste, and Community Impact | 6 | 5 | 4 | 3 | Shawn chimes in with knowledge regarding community pushback and power grid constraints. Anjney explains cluster utilization benchmarks and breaks down the systemic inefficiencies and regulatory risks facing modern AI data centers. | |
| The Compute Grid: AMP's Independent System Operator Model | 7 | 6 | 3 | 6 | Shawn pushes back by pointing out that full-stack integration like xAI and OpenAI aligns incentives, and later cites Google's internal credit failures leading to missing GPT. Anjney details the Independent System Operator model and horizontal compute grid mechanics. | |
| Unlocking Trapped Research and Scaling Compute Capacity | 6 | 5 | 3 | 2 | Shawn notes European researchers complaining about DeepMind's publication scraps. Anjney explains the adverse selection created by corporate research embargoes and details AMP's multi-gigawatt compute targets. | |
| AI in Healthcare: End-of-Life Prediction and Cultural Perspectives on Death | 4 | 7 | 2 | 1 | Anjney leads an extensive monologue discussing his bioinformatics background, end-of-life care predictions, malpractice dynamics, and cultural perspectives on death. Shawn mostly listens and shares his Buddhist upbringing. | |
| The Discipline of Output Maxing and Compute Market Protocols | 6 | 5 | 2 | 3 | Shawn references SF Compute's work creating compute futures contracts. Anjney outlines output maxing, alignment trade-offs across API layers, and compute market liquidity dynamics. | |
| Hardware Co-Design and Navigating Silicon Trust Boundaries | 6 | 6 | 3 | 4 | Shawn questions whether non-NVIDIA chips like MatX harm standardization. Anjney clarifies that MatX adheres to NVIDIA's open reference architecture IO footprint and explains hardware co-design trust boundaries. | |
| The Scientist-CEO: Academic Excellence as High-Performance Leadership | 5 | 6 | 5 | 3 | Shawn mentions Anastasios's background, and Anjney strongly attacks the VC trope that researchers cannot be great CEOs, citing Dario Amodei and top academic rigor as peak high-performance training. | |
| Authentic Connection, Mentorship, and Rejecting Superficial Narratives | 6 | 5 | 4 | 4 | Shawn and Anjney discuss industry narratives versus first-principles engineering. Anjney rejects media narratives of winning versus losing, arguing true researchers care about specific bottlenecks rather than generic categories. | |
| Cultural Resilience, First Principles, and Anthropic's Preparedness | 6 | 6 | 4 | 4 | Shawn questions the hypothesis that Anthropic's coding lead was just a lucky dice roll and cites burn rate comparisons. Anjney counters with Anthropic's multi-year preparation, mission-aligned P0 focus, and the fragility of culture. |
Statements from this episode (23)
Most AI clusters fail to hit Google's 96% node utilization standard
“My co-founder, Seb came from he built the Borg export GQM scheduler at Google, and there, I think, 95% was considered an outage, so 96% node utilization is, should be standard, and most single-time clusters are not running at that”
Midha: Best-in-class AI cluster MFU is between 60% and 70%
“And then MFU utilization should be, I would say, the best in class today, somewhere between 60 and 70%.”
Up to 20% of US data centers risk cancellation from community backlash
“Up to 20% of all data centers this year in the US, my understanding is are at risk... Of not getting the community support they need to get brought up.”
Midha: Regulators will investigate AI infrastructure operators that bypassed rules
“There will be audits. There will be investigations, and when the regulators come, I don't know when it's going to be. The folks who are moving fast and breaking things in the name of AI progress better be prepared.”
Midha: The 'NeoCloud' concept is mostly marketing speak
“I think this whole idea of NeoCloud being somehow this new category is a lot of marketing speak. There are really good, reliable, trusted data center providers in America who've been around 20 plus years.”
The most enduring electric grids pooled uncorrelated demand without owning assets
“In order, if you study, like, the history of grids, the most enduring ones were those that never owned their own assets. They were ones that had, ah, already often started with long-term anchors who are uncorrelated sources of demand”
Midha: AMP Is Pooling 1.3 Gigawatts of Compute Supply Over Four Years
“We pool demand, we pool supply from a number of partners we trust at about 1.3 gigawatt scale over four years.”
Google's internal compute credit market caused it to miss GPT
“This is a thing that has been tried internally within Google, and it led to Google missing GPT.”
Midha: AMP Foundry invested hundreds of millions into Anthropic this year
“We put a few hundred million dollars into Anthropic from our fund earlier this year.”
Google DeepMind's six-month internal embargo shelves commercially valuable research papers
“I mean, what's worse is the paper is actually not even being Published anymore because there's a six month embargo inside of DeepMind, right? Like we've heard about this where a paper comes out and then I think there's a six month embargo window where if anybo…”
Midha: Stanford Holds America's Second-Largest Longitudinal Patient Dataset Behind the VA
“Stanford is one of the only research facilities in America that has a longitudinal patient data set that's Larger at scale, I think it's at least twelve million patient lives. The only larger data set is the VA, the Veterans Affairs, you know, of America.”
AI diagnostic adoption is blocked because legal liability remains with doctors
“The problem remains then and now is regulatory because you actually can't shift the burden Of the wrong clinical diagnoses from the physician to the AI system.”
Midha: Anthropic's velocity came from standardizing on the transformer architecture
“Like, one of the reasons Anthropic has had extraordinary sort of velocity is because they picked the transform architecture and said, this is simple, let's double down on it, right? And now, luckily, there's enough investment going into space that we can affor…”
AMP's projected excess annual compute capacity vanished in six weeks
“Unfortunately, what's happened the last six weeks. Is, you know, we thought we'd have a bunch of excess capacity by the end of this year. It's all gone.”
Midha: MatX chips adopt NVIDIA reference architecture to plug into existing sites
“When they decided to pick the standard for their data center, they picked the NVIDIA reference architecture. So the Matex chips just plug in to any site that has an NVIDIA bring up planned. And you know.”
Midha: AI chip co-design requires visibility into next model architectures
“To do co-design well, you need visibility into the next model generation as soon as possible, because it takes two years to tape out. So if by the time I bring my chip to market, your model architecture's changed, I'm host.”
Anthropic will become a trillion-dollar company within four years of founding
“Have you met Dario? Dario's a scientist. He's gone from zero to like what will soon be a trillion dollar company in four years.”
Midha: Elite academic publishing prepares scientists for high-performance CEO leadership
“Being a great CEO actually requires a level of performance that scientists who have already published at the top of their field have accomplished. It is super hard to be a competitive scientist.”
Midha: Categorizing AI architectures as 'world models' lacks useful technical precision
“World models, don't get me wrong, are cool and everything, but you and I both know that that is a layer of abstraction that is sometimes not usefully precise enough.”
Anthropic achieved technical model takeoff during its October 2023 training run
“What happened is, Anthropic basically achieved takeoff in October of last year. That training run.”
Anthropic made coding its day-one priority as the mechanism to AGI
“And there, P zero from day one was coding. The reason the mechanism system there was, if we crack coding, Then we will crack AGI. You know, our mission is AGI. We want to get there safely. If we focus on coding, it's such a generally powerful capability that i…”
Midha: AI Labs Raising Too Much Money Too Fast Develop Fragile Cultures
“And I think teams who can raise too much money too fast, too early, who don't have to define what the P zero is, because that's the only thing when you have scarce resources, you gotta invest in. Those cultures end up being the most fragile and brittle, and th…”
Midha: Frontier AI Models Were Terrible at Analyzing Condensed Matter Physics Data
“We had started benchmarking frontier models on physics and science capabilities, and they were not very good. They were good at, like, doing things like summarization of papers, but if you said, hey, could you, like, analyze the scientific data coming out of a…”