Isabel Juniewicz (@brightertmrrw) header
Isabel Juniewicz (@brightertmrrw) profile picture

the only way out is through. not sure where we're going but trying to make sure we get there. answering questions @epochairesearch

Joined November 2021
Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)Photo by Isabel Juniewicz (@brightertmrrw)
Mackenzie’s work on Megaspeed last year was one of the things that convinced me that the scale of smuggling was larger than previously estimated and was an input into my work on smuggling earlier this year epoch.ai/publications/chip-s…
A year ago, we asked Nvidia about a Shanghai data center with links to Alice Huang. They declined to answer our questions and insisted that when it came to Megaspeed, there was nothing to see We now have filings showing there are indeed banned Nvidia chips at that site Thread:
1
3
14
740
I’ve updated my median estimate even higher since then! But still broadly in the previously estimated range.
3
55
but this relies on the CEO not having outsized voter shares, doesn’t it? would dual class stock with the CEO having more super-voting shares produced the same outcome?
Earlier this week, many expressed incredulity that Huang could be ousted. Some arguments were valid, but other arguments suggested people don't always understand the differences between different kinds of institutions and jobs. Some had a kind of "it's his company, you can't just remove him" idea. Not true, shareholders can force the issue if they need to; this isn't a sole proprietorship. Some drew a comparison to Eric Swalwell, who made a similar offense. But an election mechanically confers some job security that a corporate job does not. Some invoked US criminal law standards, where one gets a chance to prove innocence in court. But shareholders and boards can oust executives for non-criminal reasons. This goes back to a good @xwanyex point from a while back; for many, there's just a vague "they" who do stuff. Really, though, the specific designs of institutions do drive outcomes, and Huang was subject to a pretty high level of accountability.
3
200
I think this might be the most positive I’ve felt about an election since 2008 - feeling optimistic about tonight and about 2028.
1
12
245
That's because Nvidia's design work is missing from US GDP statistics! It flips back to positive once you account for that. epoch.ai/publications/the-nv…
Replying to @JosephPolitano
So much of the value of AI infrastructure is embedded in imports rather than US production that the boom in AI investment has had much less impact on GDP than headline figures suggest, even flipping negative this quarter as import growth outstrips investment growth
3
14
703
for a while my 1-year-old would go to bed late (9 or 10) but sleep pretty well through the night - would wake up at 6ish to nurse and then would sleep in until 7:30 or so, which is when we'd like to get up.
2
9
410
now he's been falling asleep closer to the recommended time (7 or 8) but his sleep is terrible, wakes up more times in addition to concluding that he's done sleeping at 5am. how deranged is deliberately trying to keep up a toddler until 9pm?
4
5
307
it’s warning shot summer and i think now is a good time to slow down or stop. hopefully we can coordinate on this before we get deadly warning shots or something catastrophic. unilateral actions like this one can help kickstart coordination.
Replying to @hilbertspaess
The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible - but I hear the same people express fear privately. No other human activity poses this level of danger.
1
10
396
me: oh right, it’s September. look, a lot is going on and I have a baby.
1
5
119
also arbitrarily many ai things may yet happen between now and the equinox.
1
64
Imagine an airplane crashes under mysterious circumstances. Except in this world, there is no government oversight, and all investigations are done voluntarily by the airlines themselves. Nonetheless, the airline wants to reassure their customers, so they do an investigation. To investigate, the airline invites three of the world's most respected airplane researchers to look into it. Except by the time the researchers get there, the airplane has already been disassembled and melted down into little cubes. The researchers are instead given the logs of all the airplane directions and the transcripts of the conversations between the pilot and the copilot. Except one tenth of the log has been deleted. And the airline also tells the investigators that they can only look at the time between when the airplane first started losing altitude and when the airplane made contact with the ground and that all investigation of the time before the airplane started losing altitude is off limits. Also there are a lot of rumors in the airplane community of a few other airplane crashes under somewhat similar circumstances, but the airline says that these are also explicitly off limits. When the researchers arrive to review the logs, they find that the logs are 10,000 pages. But they only get six days to read them, and half the logs don't arrive until day 4. Now you understand the independent METR investigation.
20
128
14
1,168
45,407
fact check: false, this just doesn't get booked into GDP anywhere epoch.ai/publications/the-nv…
Replying to @SemiAnalysis_
The catch: roughly half the value of those AI imports is US value added — design and IP. Trade data only captures the freight value of the goods. The US-created value shows up as profit at US companies, correctly booked into GDP, just somewhere else. (3/6)
1
25
1,020
Isabel Juniewicz retweeted
We discovered that US GDP statistics miss most of the value Nvidia adds to the US economy. As a result, GDP growth has been understated by ~0.3 percentage points over the last year.
90
254
101
1,901
805,435
I would not have correctly predicted that babies can climb up slides before they can slide down them. or walk.
7
363
Isabel Juniewicz retweeted
The end of the self-funded AI buildout? Hyperscaler cash capex is growing much faster than cash inflows. On current trends, they will be unable to fully fund the AI infrastructure buildout with cash from operations by the end of this year.
12
67
26
424
98,222
both of the numbers used to estimate these returns are way too low - capital investments are growing 70 percent a year, and AI revenues are growing waaay faster than that (though from a lower base)
The AI numbers are starting to look very ugly. Even under "best case" assumptions, FT's own data shows Microsoft AI ROI at -9%, Google at -15%, Meta at -28%, Oracle at -35%. Only Amazon barely comes out positive. This is exactly why I keep comparing this to the dot-com era. Incredible technology does not automatically mean sustainable economics. The internet survived. Most internet companies didn't. Right now hyperscalers are spending trillions hoping future demand catches up to present capex. That's not certainty. That's a leveraged bet.
1
3
568
if these are your customers, you might reasonably expect to have revenues growing at more than 15 percent a year.
1
176
Isabel Juniewicz retweeted
Hyperscaler capital expenditures came in on trend in Q1 2026, continuing the trajectory that projects them spending $770 billion this year and over a trillion dollars in 2027.
Driven by investments in AI, hyperscaler capital expenditures have grown 70% per year since the release of GPT-4, nearing half a trillion dollars in total during 2025. If this trend continues, Alphabet, Amazon, Meta, Microsoft and Oracle will spend $770 billion on capex in 2026.
5
42
3
186
20,611
Isabel Juniewicz retweeted
Given that developing sufficiently advanced AI has potentially severe risks, whether we should intentionally slow down still depends a lot on the current level and pace of AI capabilities. But as of this year I would like to see a major slowdown.
Replying to @BethMayBarnes
(4) IMO, any “reasonable” civilization would clearly be taking things much more slowly and carefully with AI. The benefits of getting upsides of advanced AI a little faster are small compared to the risks of getting it irrecoverably wrong, and we could lower these risks by going slower
2
3
2
37
2,648
has anyone ever put "colony on Mars" as a vesting condition before? not sure if that or 100 terrawatts in space is more unusual (or more unlikely!)
1
119