Kristie Huang

27 followers · 20 following · 3647 views

on the atlas — 83

highlights — 684

A hedged OTC desk for compute. We buy GPU capacity forward, hedge it, and deliver one flat fixed rate for the term.
https://itomarkets.com
The system is a 105-mile-long piping network interconnected grid so a customer can get steam from any one of our locations at any time
How the New York City Steam System Works - Untapped New York
Some of New York City’s most famous buildings are steam powered, including the Empire State Building, the Chrysler Building, Grand Central Terminal, the United Nations, Rockefeller Center and more.
How the New York City Steam System Works - Untapped New York
To sidestep lengthy queues for connecting to energy grids, at least three of the seven sites will make use of on-site natural gas plants. To address public concerns about water usage, at least six sites will use closed-loop liquid cooling systems, which do not evaporate water.4 These decisions will likely save the project time but raise the cost of the facilities.
OpenAI Stargate: where the US sites stand
The critical distinction in relying on allies today, compared to the offshoring of the late 20th century, is about capability flows. In the 1980s, America surrendered the technological and workforce capabilities around shipbuilding, allowing them to flow towards other countries. Today’s proposed partnerships invert that flow. American workers travel to South Korea to absorb labor capabilities that then return home; Korean capital and equipment flow into American shipyards to rebuild industrial infrastructure on US soil.
Shipyards, Graveyards
In World War II, the US simply outbuilt our enemies. William S. Knudsen described the strategy as “[smothering] the enemy in an avalanche of production, the likes of which [they] had never seen, nor dreamed possible.” That won’t work this time because, today, China occupies the dominant manufacturing position the US held in 1942. Where the US produced around 40% of global manufacturing output in 1942, China represents 35% today.
Shipyards, Graveyards
To address this vulnerability, we must resurrect our national shipbuilding industry. But any attempt to do so must reckon with the chicken-and-egg problems inherent in bootstrapping a dying industry back into global competitiveness.
Shipyards, Graveyards
The dependence is reckless, but the economic justification is straightforward: Regulators only require these ships to be US-flagged and crewed; otherwise, the ships would cost roughly $200 million, instead of their current sticker price of $50 million.
Shipyards, Graveyards
the reality is that 80% of trade volume moves by sea. Without domestic shipbuilding capacity, the flow of global trade, including energy, raw materials, and consumer goods, is ultimately at the mercy of those who do control it.
Shipyards, Graveyards
Particularly in the 1980s, the US came to the conclusion that manufacturing and shipbuilding were commodity processes better suited for lower-cost labor markets in Asia. We’ve seen the same dynamic play out in semiconductors, batteries, mining, and pharmaceuticals. But shipbuilding represents a distinctly gnarly systems problem tangled between defense funding, policy, and a fading American manufacturing core.
Shipyards, Graveyards
While compute can also be bought on demand, take-or-pay is the primary pricing method because it creates revenue certainty. As noted above, that revenue, rather than data center hardware, is what makes neocloud CapEx financeable.
Why 20% of Neoclouds Won’t Survive The AI Boom
The AI infrastructure build-out by neoclouds is primarily funded by debt, not equity. Compared to equity investors, credit committees prioritize capital preservation over high returns. While debt for data centers and new power projects can be underwritten at 15-year lifespans, GPUs are too short-lived to satisfy credit requirements; lenders heavily discount GPUs as collateral due to depreciation risk and instead focus on the offtake contracts as collateral.
Why 20% of Neoclouds Won’t Survive The AI Boom
Combined with limited differentiation between models, this makes AI company pricing very sensitive to the cost of compute. Those companies that secure the cheapest reliable compute will have the highest margins when model quality is effectively commoditized.
Why 20% of Neoclouds Won’t Survive The AI Boom
who's been building his company for a few years. what became most striking is that he is now in a beautiful flow with his life. his days are devotionally repetitive: jiu jitsu at 7am, coffee, then work until 6pm. Spend time with his wife, read for a bit, then sleep. He tells me that his days are structured and predictable because that is how he does his best work. he says that he is no longer distracted, it is very easy to focus for long periods of time
shani 🌱 (sf) on X: "had dinner with an old roommate who's been building his company for a few years. what became most striking is that he is now in a beautiful flow with his life. his days are devotionally repetitive: jiu jitsu at 7am, coffee, then work until 6pm. Spend time with his wife, read for" / X
Linear is the shared product system that turns context into execution. It holds feedback, intent, decisions, plans, and code, shapes that context into work, and helps humans and agents carry it all the way to production.
Issue tracking is dead – Linear
PROFESSIONAL NEWS ALERT: As we have said a couple months ago, NVIDIA GPU rental prices are rising rapidly again and capacity is being sold out. Between mid 2024 to Q3 2025, it was an customers market for compute where customers were able to neogigate low prices & favourable terms with Neoclouds. But due the surge of demand of agentic coding & increase in DRAM pricing, Customers no longer have as much negotiating power.
SemiAnalysis on X: "PROFESSIONAL NEWS ALERT: As we have said a couple months ago, NVIDIA GPU rental prices are rising rapidly again and capacity is being sold out. Between mid 2024 to Q3 2025, it was an customers market for compute where customers were able to neogigate low prices & favourable terms https://t.co/scV55IcbPI" / X
If anything, this shift is teaching me what it is like to be human again—not human in the romantic, AI-can-never-replace-us sense, but human in the uncomfortable sense, the part where you have to let go of the thing you were in order to become the thing you might be.
Aditya Agarwal on X: "When Your Life’s Work Becomes Free and Abundant" / X
The doomers say we’re all going to be replaced, and the boosters say everything will be fine. Neither camp captures what it actually feels like to live through this. The truth is messier. You can hold wonder and grief in the same hand, mourn a version of yourself while sprinting toward a new one.
Aditya Agarwal on X: "When Your Life’s Work Becomes Free and Abundant" / X
1. The Identity Collapse I have never imagined, that would be a term like that. In fact, in your 20s, your identity is 'becoming'. You're becoming better You're becoming richer You're becoming more skilled. But in your late 20s, something much dangerous happens. I've already become something, you've already become something. But now, both you and I must decided: Is thit it? If your entire identiy was built around growth, hustle and achievements - what happens when you reach a milestone and still feel empty? That's when the crack appears, not because you failed, but becasue success didn't feel …
andrew.moh on X: "Facing Late-20s Hidden Depression: How to rebuild intentionally" / X
Whether they grow online or IRL, fortressed microcommunities are easier targets for commercial entities: the head of A24 Labs hinted at this, saying how, with AI, “[y]ou can maybe make something that’s valuable for a thousand people, because the cost is cheap enough that it doesn’t have to reach ten million people” (The New Yorker).
Three creative tides to care about now - by Steph Yen
I get itchy when a candidate uses an LLM to think through how the project should be completed. It's not enough to screenshot the Figma, upload it to Claude Code, and ask it to solve the problem. You know how to architect. Show me!
How We Hire Engineers When AI Writes Our Code | Tolans.com
In a world where implementation is getting easier and easier, what I'm looking for is judgement. You won't have enough time to create a polished product. With this in mind, I want to see what sacrifices you're willing to make. And when you can't get everything looking or feeling just right—what drives you crazy?
How We Hire Engineers When AI Writes Our Code | Tolans.com
If I give you a product spec, can you build it quickly? Can you tell me where the holes are? Can you reason through trade-offs? Do you know when something's not ready to ship? Do you know when something is ready to be reviewed and maintained by another person?
How We Hire Engineers When AI Writes Our Code | Tolans.com
In traditional Software Engineer system design interviews, candidates will be asked to describe the architecture of a distributed system, usually involving web servers, API gateways, load balancers, caches, databases, microservices, message queues, streams, etc. For Front End Engineers, system design interviews are slightly different – there's more emphasis on what goes on in the client and API design between the client and the server, as opposed to what goes on in the back end.
Front End System Design Playbook | GreatFrontEnd
But it takes a brave and resolute spirit not to be disconcerted in times of difficulty or ruffled and thrown off one's feet, as the saying is, but to keep one's presence of mind and one's self-possession and not to swerve from the path of reason.
Chip Futures — Felix Stocker
The point of all this is to say that 1989 and 2001 were particularly good times to be an innovator in commodity futures. Not only was the CFTC relatively toothless, the internet was still the Wild West. Post Dodd-Frank, these things are much more tightly regulated.
Chip Futures — Felix Stocker
The chip market is idiosyncratic; typically, futures markets for agricultural commodities have more difficulty convincing consumers than producers.
Chip Futures — Felix Stocker
Customer relationships: producers don’t want to sell a commodity, they want to sell a personalised product to a customer with whom they have a long-term relationship. Differentiating on things other than price keeps prices higher.
Chip Futures — Felix Stocker
The second statement makes sense; it makes sense for oil wells and diamond mines to reduce production of a finite resource in order to prop up prices, but that’s not the case for semiconductors.
Chip Futures — Felix Stocker
In other words, he was happy to ink bilateral forwards contracts; he didn’t want futures.
Chip Futures — Felix Stocker
But in the case of DRAM futures, the challenge was to get the manufacturers on board.
Chip Futures — Felix Stocker
Rockefeller’s Standard Oil established a near monopoly on pipelines, refining, and retail distribution; and so the wildcatters and production companies who took on the risk of striking oil in the first place were firmly in the position of price-takers. If you’ve got a monopsony, prices can be whatever you want them to be - who needs futures?
Chip Futures — Felix Stocker
If one side of the market is too consolidated, then it’s very difficult to build a futures market:
Chip Futures — Felix Stocker
But to make them work, you need people on both sides who want to hedge out price risk, plus speculators/market makers. Ideally, you want a fragmented set of buyers and sellers.
Chip Futures — Felix Stocker
“The problem is that DRAMs are an unstable commodity that undergoes a wide variety of fundamental product changes over time. We believe the inherent instability of the offering renders forward contracts impractical”
Chip Futures — Felix Stocker
As a result, it makes sense that consumers of microchips would be more willing to engage in a futures market than manufacturers.
Chip Futures — Felix Stocker
Someone involved in the 1989 effort told me that it was the manufacturers, not Compaq and Dell, who didn’t want to admit that their products were interchangeable. That makes more sense - Dell would really like to commoditise their inputs, in order to have suppliers compete on price.
Chip Futures — Felix Stocker
Size isn’t everything; other attributes matter too. As analyst Sherry Garber put it in 2001, “It's impossible to come up with a so-called commodity DRAM with fit, form, and function that serves all memory applications”; in 2003, she said that “you can have a standard density, a standard type, and a standard configuration, but you can have infinite combinations of those”. The SGX contract is actually for a “256Mb Double Data Rate Synchronous DRAM, 32M x 8, PC266 and above, 66 pins TSOP2, taped & reel.”, with a specific SKU for Samsung, Micron and Hynix. At this point, we’re not trading a commod…
Chip Futures — Felix Stocker
The focus on AI/ML comes from the large amounts of GPU power and fast processing required to run these programs.
Report: Modal Business Breakdown & Founding Story | Contrary Research
Together, Bernhardsson and Bubna had two goals when developing: first, faster computing, and second, using a Python SDK to allow easy application development that could turn normal functions into serverless functions.
Report: Modal Business Breakdown & Founding Story | Contrary Research
Pereyra offered this advice for other founders working in AI: “Don’t build for the current capabilities of models today—build for where the models are going to be. Tackle more complex versions of problems so that when better versions of the models come out, they aren’t solved as a side effect.”
Customizing models for legal professionals | OpenAI
Harvey and OpenAI worked together to add the depth of context needed, first starting with case law from Delaware, and then expanding to include all of U.S. case law. They added the equivalent of 10 billion tokens worth of data to power the custom-trained case law model.
Customizing models for legal professionals | OpenAI
The core concept is that in any arena of life, at any stage, you will always sense that there is some in-group you are not quite in—a slightly-higher-status, intangible upper tier that you feel just on the cusp of.
want what you have - Isabel Unraveled
The people building this technology are simultaneously more excited and more frightened than anyone else on the planet. They believe it's too powerful to stop and too important to abandon. Whether that's wisdom or rationalization, I don't know.
Matt Shumer on X: "Something Big Is Happening" / X
I'm writing this because I think the single biggest advantage you can have right now is simply being early. Early to understand it. Early to use it. Early to adapt.
Matt Shumer on X: "Something Big Is Happening" / X
I think the honest answer is that nothing that can be done on a computer is safe in the medium term. If your job happens on a screen (if the core of what you do is reading, writing, analyzing, deciding, communicating through a keyboard) then AI is coming for significant parts of it. The timeline isn't "someday." It's already started. Eventually, robots will handle physical work too. They're not quite there yet. But "not quite there yet" in AI terms has a way of becoming "here" faster than anyone expects.
Matt Shumer on X: "Something Big Is Happening" / X
A lot of people find comfort in the idea that certain things are safe. That AI can handle the grunt work but can't replace human judgment, creativity, strategic thinking, empathy. I used to say this too. I'm not sure I believe it anymore.
Matt Shumer on X: "Something Big Is Happening" / X
If you're really short on time, we recommend covering the Delivery Framework section, skimming the Key Technologies, and spending any remaining time studying the Core Concepts section.
System Design in a Hurry | Hello Interview System Design in a Hurry
At a high-level, while all candidates are expected to complete a full design satisfying the requirements, a mid-level engineer might cover the basics well but not into great depth, while a senior engineer will quickly work through the basics leaving time for them to show off the depth of their knowledge in deep dives.
System Design in a Hurry | Hello Interview System Design in a Hurry
The most important tactic: individually talk about it with many relevant people, before you finally pitch it to a decision-making group.
Engineering Buy-in