Underclassic rss archive buyer's guide
Permanent copy of issue 87. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
SK hynix reported record third-quarter revenue of 104.6 trillion won and an operating margin of 80 percent on Wednesday, and said memory for AI will stay in short supply through 2030. A day earlier Oracle and Nvidia agreed to expand Oracle's Vera Rubin clusters to multi-gigawatt scale, Oracle said it would become the first cloud outside Google to rent Google's TPUs, and AMD said Meta is testing its Helios rack systems. North Carolina's governor and attorney general asked regulators to make data centre owners help pay for new power plants, while Apple's new $1,999 MacBook Pro with the M6 chip promises faster language models on the laptop itself.
Date: 2026-10-28
Updated: 2026-10-28 06:30
Status: LIVE
Broadcast
The world's second-largest memory maker now earns 80 cents of operating profit on every dollar of sales, and it has already promised most of the next two years to its biggest customers. That is the plainest measure yet of who is buying the future of AI: the clusters that Oracle, Meta and their peers are building at gigawatt scale. North Carolina is asking the obvious question of who pays for the power plants those clusters need. At the other end, a $1,999 laptop that runs language models on its own is not a rival to a gigawatt campus, but it is the reason the local side of the ledger still exists.
Watch
Stories
SK hynix reported third-quarter revenue of 104.6 trillion won and operating profit of 83.3 trillion won on Wednesday, an operating margin of 80 percent, as sales of HBM4 memory for AI accelerators began in earnest. The company said it has long-term agreements covering more than half of its 2027 and 2028 production and expects demand to outrun supply through 2030.
-- When the memory that feeds AI chips is booked two years ahead by the largest buyers, smaller labs and desk-side builders pay more and wait longer for the same parts.
2026-10-28 / SK hynix Newsroom / chip / central / #sk-hynix-posts-record-profit-and-says-ai-memory-
Oracle and Nvidia said on Tuesday that Oracle Cloud Infrastructure will expand its Vera Rubin NVL72 clusters to multi-gigawatt scale, and that the systems and Nvidia's Vera CPUs are now generally available on OCI. Oracle also said it will become the first cloud provider outside Google to offer Google's tensor processing units.
-- Renting rival chips under one roof gives cloud customers more choice, but it measures the build-out in gigawatts, which is power that someone must generate and pay for.
2026-10-27 / NVIDIA Newsroom / chip, energy / central / #oracle-expands-nvidia-deal-to-multi-gigawatt-sca
AMD said on Tuesday that Meta has installed initial Helios rack-scale systems for testing ahead of a broader deployment, and that Meta will deploy a custom Instinct GPU built for its workloads. AMD also said it provides day-one support for Meta's open Muse Spark Open models on Instinct GPUs and for Muse Glimmer on Ryzen AI PCs and Radeon graphics cards.
-- Meta is buying racks at data centre scale while handing out weights that run on a single PC, so the same company is feeding both ends of the market.
2026-10-27 / AMD Newsroom / chip, efficiency / mixed / #amd-says-meta-is-testing-its-helios-racks-and-ba
North Carolina Gov. Josh Stein and Attorney General Jeff Jackson asked the state Utilities Commission on Tuesday to require data centre developers to pay for new generation or bring their own clean energy. Stein said data centres make up about 80 percent of projected load growth and that their added power could cost the state about $50 billion by 2050.
-- If regulators move the cost of new plants from household bills to the data centres that need them, the true price of central computing rises and the sums behind cheap API tokens change.
2026-10-27 / NC Newsline / energy / central / #north-carolina-leaders-ask-regulators-to-make-da
Apple introduced a 14-inch MacBook Pro with its new M6 chip on Tuesday, saying it runs on-device AI workloads such as language-model prompt processing up to 25 percent faster than M5. The laptop starts at $1,999 with 16GB of unified memory, configurable to 32GB, with pre-orders from October 28 and deliveries from November 6.
-- A mainstream laptop that runs useful models without a meter keeps the local option within reach of students and small offices, though 32GB still caps how large a model it can hold.
2026-10-27 / Apple Newsroom / efficiency / local / #apple-puts-its-m6-chip-in-a-1-999-macbook-pro-bu
Axis
local [#########################-----] central
Score: 83
clusters still winning
unchanged
SK hynix has sold more than half of its 2027 and 2028 output under long-term contracts, and Oracle is adding gigawatts of Nvidia racks and Google chips, so the money and the memory are still flowing to the largest clusters. Apple's M6 laptop and AMD's support for Meta's open models on Ryzen PCs keep the desk in the race, and North Carolina's push to bill data centres for new plants could raise the cost of central computing; the score holds at 83.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
98 | central | SK hynix earned 83.3 trillion won of operating profit on 104.6 trillion won of revenue in the third quarter and has long-term agreements covering more than half of its 2027 and 2028 capacity; Oracle will scale Nvidia Vera Rubin NVL72 systems to multi-gigawatt size and add Google TPUs. | ||
| ENERGY |
|
97 | central | North Carolina Gov. Josh Stein says data centres account for about 80 percent of projected load growth and that their added power could cost the state about $50 billion by 2050; he and Attorney General Jeff Jackson want developers to pay for new plants or bring their own clean energy. | ||
| EFFICIENCY |
|
74 | mixed | Apple says the M6 in the new 14-inch MacBook Pro runs on-device AI workloads such as language-model prompt processing up to 25 percent faster than M5, with 16GB of unified memory standard and up to 170GB/s of bandwidth; these are Apple's own figures. |
Line charts (0=local 100=central 2026-10-11 to 2026-10-28)
Market overlay (2026-09-29 to 2026-10-28)
Latest: BTC $81,342 / hashrate ~1072.36 EH/s / token markers avg $5.2583/M / axis 83. These lines are different units. We compare timing, not dollars-to-hashrate.
Not shown: GPU cloud $/hour, hub electricity, miner hashprice, or lab cost per token -- none of those are a clean public feed.
Power and list prices
BTC $81,342 / hashrate ~1072 EH/s. List-price gap (not a subsidy %): 123.1x list gap (blended) Mercury 3 -> Claude Fable 5.1 ($0.16/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)
Marker prices (USD per 1M tokens; verified 2026-10-22; blended = (3*input + 1*output) / 4; -- = open/self-host or no single public list)
| # | Model | Org | CC | Weights | Tier | In | Out | Blend |
| 1 | Claude Sonnet 5 | Anthropic | US | closed undisclosed |
workhorse | $2.00 | $10.00 | $4.00 |
| 2 | GPT-5.5 | OpenAI | US | closed undisclosed |
flagship | $5.00 | $30.00 | $11.25 |
| 3 | DeepSeek V4.1 Flash | DeepSeek | CN | open ~284B total / ~13B active (MoE) |
commodity | $0.30 | $1.20 | $0.53 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
Subscribe
$ echo you@email.com | underclassic --subscribe
One email per issue. No tracking pixels, no filler.
home | rss | page 2 | page 3 | archive | models | buyer's guide | subscribe | data | board | mcp | llms.txt