Micha "mies" Hernandez van Leuffen @mies

Founder @fiberplane; autonomous software factories. Investor in teams solving hard problems @nphardvc. fiberplane.com distributed Joined October 2006

Tweets

5K
Followers

2K
Following

1K
Likes

6K

Micha "mies" Hernandez van Leuffen @mies

2 days ago

@robindchnt some kind of quantum entangled comms network that goes faster than the speed of light

0 0 1 70 0

View Details

Micha "mies" Hernandez van Leuffen @mies

3 days ago

@kellyjgreer Multi-harness world seems likely. Specifically across certain verticals

0 0 0 61 0

View Details

Micha "mies" Hernandez van Leuffen @mies

3 days ago

Everybody patched (but not too fast)???

0 0 3 196 0

View Details

Micha "mies" Hernandez van Leuffen @mies

4 days ago

@carmenli @josephcompute Not only nightlife; Club Berghain is open during the day as well ;)

0 0 0 173 0

View Details

Micha "mies" Hernandez van Leuffen @mies

a week ago

“Build or surrender: for now, the choice is still ours.” Good piece by @arthurmensch, and I'm with him on building (a lot!). But the thing is that Europe isn’t going to out-spend the US on GPUs. The reality is that we have less capital, more expensive power, and fewer data centers. So building looks different in Europe. Every GPU and every watt we have needs to do more work than it does anywhere else. If we get more work out of every watt, we don’t have to wait years for a grid connection to get there. We should still invest in power, obviously. But the reality is that new generation and grid capacity takes years, and we can’t afford to sit around and wait for them. So this is where I’d put the effort now, and this is also where startups can shine. Build different, do more with less 🇪🇺.

Arthur Mensch @arthurmensch

a week ago

Should Europe leave frontier AI to the Americans and the Chinese? The idea has gained ground in Europe. It is a fundamental mistake. Build or surrender: for now, the choice is still ours. Full article below. x.com/i/article/2110…

268 764 4K 437K 1K

1 0 9 470 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@blwiertz congrats and much need in Europe!🇪🇺

0 0 0 97 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

The timeline tunes fast here

0 0 5 205 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

Coinbase made training a specialist model one command in their harness. My sense is that serving is where the work goes next. If every company ends up with a handful of models like this, each one has its own latency target and its own scorer. A fraud model that has to answer in milliseconds and an overnight code-review model want different batching, and often different GPUs. This means that a company's fleet ends up running many models on mixed hardware, with perhaps a frontier API kept for the cases nothing else passes. Serving all of that well is an interesting problem space.

Brian Armstrong @brian_armstrong

2 weeks ago

We were able to fine tune an open weight model (Qwen3.5-9B) on proprietary internal data at @coinbase, and it out performed frontier models, at much lower cost and speed, for a specialized task (fraud prevention in this case). I think we'll see more of this pattern inside

617 795 10K 1.6M 7K

0 0 2 348 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@DadaJudith 100% agreed but also; x.com/mies/status/21…

Micha "mies" Hernandez van Leuffen @mies

3 weeks ago

x.com/i/article/2103…

3 3 18 2K 3

0 0 1 56 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

Look at those two lines together📈: the token index halved since May while B300 rent went up 38%. Whoever serves the model pays the GPU rent and gets paid per token. The only way through is more tokens out of every GPU-hour. In Europe, where adding GPUs takes years, that matters most...

Gavin Baker @GavinSBaker

2 weeks ago

Open-weight models that decrease margins at the model layer are positive for AI infra, exhibit 1001.2 Since the @silicondata “token index,” which captures both price and mix, peaked on May 31: B300 prices are up 38% B200 prices are up 10% Other GPU indices are up *much* more.

127 158 1K 347K 589

0 1 3 314 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@AndreRetterath yes :) x.com/mies/status/21…

Micha "mies" Hernandez van Leuffen @mies

3 weeks ago

x.com/i/article/2103…

3 3 18 2K 3

0 0 0 27 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

100% agreed on cache reads being the price that matters. The Slow Vale agent simulation runs 800+ agents at 4.4B tokens a day and gets over 90% cache hits, so most of the bill is cached reads. I would add one thing though; the cache lives in the serving stack. With open weights you can run the same model at several providers, and the one that routes you back to the GPU holding your cache sets your real price. That's where I'd expect the lock-in honestly.

1 0 0 70 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@Bell11A @carmenli yes, a standard hour makes the volume hedge possible. I think the harder part is that two identical H100 hours can serve very different amounts of tokens depending on how the model is run. Would you price the hour, or the work done in it?

1 0 0 16 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

$36B of neocloud backstops and $20B of DC leases to hand off to neoclouds. Nvidia now carries part of its customers’ demand risk. The question if that ever costs them anything depends on how busy those GPUs are.

SemiAnalysis @SemiAnalysis_

2 weeks ago

Nvidia, in its most recent 10Q, disclosed $530B of gross off-balance sheet guarantees across six different line items – a jump from $184B of obligations disclosed in the prior quarter. The main drivers were increases in Supply Commitments - primarily commitments to buy memory,

41 40 408 129K 195

2 0 4 458 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@guy_drt75 @nebiusai congrats on the move! 🇫🇷

0 0 1 26 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@ttunguz Margin probably comes down to serving now. The same model on the same GPUs can cost very different amounts per task, depending on how it’s served. Are your portfolio companies investing in inference tuning yet, or just eating the bill?

0 0 1 76 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

@GavinSBaker @arakharazian Agree. The low-utilization part is underrated. Lots of small buyers running open weights means more GPUs for the same amount of work. Good for Nvidia, and probably the biggest efficiency gap in AI today.

0 0 0 784 0

View Details

Micha "mies" Hernandez van Leuffen @mies

2 weeks ago

Agree on demand. The margin is the fun part though. Same weights, same GPUs, and the cost per finished task still swings a lot depending on the serving stack. Free weights turn it into an inference engineering problem imho.

Gavin Baker @GavinSBaker

2 weeks ago

Why open-weight models that compress margins at the model layer are net positive for AI infra demand, exhibit 1001. An open-weight token consumes *roughly* the same amount of compute as a frontier token for a similar size model.