←
David Tearse147 posts
David Tearse avatar

David Tearse

@7earse

Co-Founder and CEO at Karman Industries

◷ Joined September 2018▧ 8 Photos and videos
Elon Musk avatar
⟳ David Tearse retweeted
Elon Musk@elonmusk·Nov 3

And electrical equipment, cooling, networking and batteries

Euwyn@euwynNov 3

Elon Musk says the AI boom will hit a WALL by the end of 2026. He says the whole industry is about to learn a hard lesson in hardware. Chip output is exploding, but electricity is nearly flat everywhere outside China. So the two lines cross this year. He says by the end of the year, we will be making more AI chips than we can actually turn on. Millions of them pile up with nowhere to plug in. In his words, the bottleneck for the next year is not chips, it is raw power. Whoever can turn on the most chips fastest wins the race. In the end, the AI race comes down to one thing: Raw electricity. — Elon Musk (.@elonmusk) on Dwarkesh Patel's (.@dwarkesh_sp) podcast

1:09
SemiAnalysis avatar
⟳ David Tearse retweeted
SemiAnalysis@SemiAnalysis_·Sep 12

Another theme that emerged from our Q2 tour of datacenter industrials is that modular capacity is sold out, with demand growing faster than the pace of capacity bring-up. Modular is gaining share because it addresses the two biggest constraints on datacenter construction: time and field labor. Power and cooling systems can be assembled in factories while the shell goes up on site, compressing delivery timelines by up to 40% and reducing on-site man-hours by up to ~70% for some parts of the build. We broke down the different approaches—and the schedule, labor and cost savings—in our recent newsletter (1/4)🧵 newsletter.semianalysis.com/…

David Tearse avatar
David Tearse@7earse·Sep 2

More data centers = more heat = more Karman HPUs needed 🫡

To bring large AI data centers online requires building massive power plants, transformers, liquid cooling loops & chillers, as well as incredibly complex networking (especially for training clusters)

Elon Musk avatar
⟳ David Tearse retweeted
Elon Musk@elonmusk·Aug 29

Consensus estimate is that ~15GW of AI compute produced in 2027 cannot be turned on in 2027. This is harder than just finding power, as you also need to build out all the transformers, wiring, liquid-cooling, (massive) chillers & complex networking. grok.com/share/bGVnYWN5_4f06…

Jonathan Lacoste avatar
⟳ David Tearse retweeted
Jonathan Lacoste@lacostejonathan·Jul 21

Today, Space VC becomes Ambition Capital Five years ago, I launched a venture firm to little fanfare. Few in venture had heard of us, and even fewer believed the thesis

0:55
Shanu Mathew avatar
⟳ David Tearse retweeted
Shanu Mathew@ShanuMathew93·Jul 19

CoreWeave CTO Peter Salanki on where AI infrastructure economics are heading: >~50% of CoreWeave compute is already inference. Reasoning, reinforcement learning and distributed key-value caching increasingly require training-grade networking. >A 100,000-GPU Grace Blackwell cluster consumes ~250 MW and has ~16M connection points. At that scale, failure is inevitable. Goodput and completed work matter more than headline uptime. >Losing 10% of cycles to failures, retries or poor orchestration raises effective token costs by ~10%. Cost per token is a full-system output, not a chip benchmark. >Older GPUs can retain value through workload specialization: newest chips for prefill and frontier compute, older GPUs for decode, evaluations and smaller-model inference. >Vera Rubin could cut inference costs by up to 90%, improve time to first token and increase density. >Salanki “doesn’t lose sleep over power.” He sees enough transmission-level power in the US; the actual 2026 constraints are substations, transformers, cooling equipment and the skilled trades needed to turn grid capacity into an operating data center. >CoreWeave is staying NVIDIA-only because of what they see from customers. Better focus and execution today, with obvious supplier concentration and bargaining-power risk.

Jonathan Lacoste avatar
⟳ David Tearse retweeted
Jonathan Lacoste@lacostejonathan·Jul 13

Investors are chasing behind-the-meter power Cooling manufacturing capacity is the single biggest bottleneck for AI buildout over the next 3 years

This tweet is unavailable

Jonathan Lacoste avatar
⟳ David Tearse retweeted
Jonathan Lacoste@lacostejonathan·Apr 16

AI workloads generate unprecedented heat, forcing new architecture of data center thermal systems Every GW of new compute brings massive incremental cooling demand Karman built the HPU to solve exactly this. Proud to have backed team from day zero datacenterdynamics.com/en/an…

Post attachment
David Tearse avatar
David Tearse@7earse·Jan 15

Big day for Karman introducing the Heat Processing Unit (HPU): a new category of thermal infrastructure for giga-scale AI. As data centers push toward scale, heat has become the real bottleneck. The HPU is designed to reduce footprint, cut energy use & run with zero water.

Jonathan Lacoste avatar
⟳ David Tearse retweeted
Jonathan Lacoste@lacostejonathan·9 Sep 2025

Reminder: I am actively investing $1M in founders raising their first round across: ⚡️energy generation & storage 🤖robotics & physical AI 🛰️defense & space tech 📶next-gen comms infrastructure 🏭manufacturing & industrials DM if you're starting a company & raising soon 🙌

Jonathan Lacoste avatar
⟳ David Tearse retweeted
Jonathan Lacoste@lacostejonathan·26 Aug 2025

Is venture broken today? No. But early-stage VCs need more courage. In deep tech, the “weird,” the non-obvious, the misunderstood…those are the companies that break out. Think: – Desalination before @VitalLyfe – Fishing automation before @ShinkeiSystems (@IanRountree) – Industrial thermal energy before @7earse started Karman As @caitlinbolnick1 says, it’s about “the weird & wonderful.” Enough VC group-think chasing only AI & defense trends. It feels safer for IC memos & LP calls, but it’s worse for returns. Go where others won’t. Back the founders building in overlooked markets. Build durable companies.

David Tearse avatar
David Tearse@7earse·11 Jun 2025

👀 @Karman_ind

After popular demand, we broke down 'The Current State of Global Chip Supply, Power Constraints, & Competition' Special thank you to Dylan Patel Founder of SemiAnalysis, Talal Al Kaissi, Group Chief Global Affairs of G42, Packy McCormick of Not Boring, and Gil Verdon of Extropic, for input, editing, & research. Full piece linked below @dylan522p @SemiAnalysis_ @TalalAlKaissi @packyM @beffjezos

Post attachmentPost attachmentPost attachmentPost attachment