Underclassic rss archive buyer's guide
Permanent copy of issue 88. Today's issue is always the front page.
Previous issue | This week | Next issue | archive
Alphabet spent $51.4 billion on capital projects in the third quarter and reported negative free cash flow of $9.1 billion on Wednesday, and its shares fell about 10 percent as it said spending will rise significantly again in 2027. Microsoft spent $45.2 billion in the same quarter but stayed cash-positive as Azure revenue grew 48 percent, and Samsung Electronics said on Thursday it earned a record 107.4 trillion won of operating profit as AI memory demand outruns supply. California's grid operator approved rules that let data centres connect early if they accept being curtailed, while Nvidia released an open speech model meant to run on desk machines such as DGX Spark.
Date: 2026-10-29
Updated: 2026-10-29 06:30
Status: LIVE
Broadcast
Alphabet is the most profitable advertising business ever built, and last quarter it still spent $9 billion more than it brought in, because data centres cost more than search earns. Microsoft still paid its own way, though its $18.6 billion of free cash flow was 27 percent less than a year earlier. Samsung and SK hynix, who sell them the memory, have never been more profitable. That is the shape of central AI this autumn: the builders strain, the suppliers prosper, and California is writing rules to switch the new loads off first when the grid runs short. Nvidia's open transcription model is a small thing by comparison, but it is the kind of work that can leave the cloud entirely.
Watch
Stories
Alphabet said on Wednesday that it spent $51.4 billion on capital projects in the third quarter, more than double a year earlier, leaving free cash flow at negative $9.1 billion. Its shares fell more than 10 percent after hours as the company kept 2026 guidance of $195 billion to $205 billion and said spending will increase significantly in 2027.
-- When the company that owns Google Search must borrow and sell stock to pay for data centres, the cost of central AI is running ahead of what even the richest business can earn from it.
2026-10-28 / CNBC / chip / central / #alphabet-s-free-cash-flow-turns-negative-as-ai-c
Microsoft reported first-quarter revenue of $92.3 billion on Wednesday, with Azure up 48 percent and contracted future revenue up 73 percent to $677 billion. It spent $45.2 billion on capital projects and finance leases, below forecasts because some third-party capacity arrived late, and still produced $18.6 billion of free cash flow.
-- A $677 billion backlog of signed cloud contracts means much of the next few years of AI work is already booked to rented data centres rather than to machines customers own.
2026-10-28 / Microsoft Investor Relations / chip / central / #microsoft-stays-cash-positive-as-azure-grows-48-
Samsung Electronics said on Thursday that it earned 107.4 trillion won of operating profit on revenue of 195.3 trillion won in the third quarter, the first time a technology company has topped 100 trillion won in a quarter. The company said it sharply increased sales of HBM4 memory and expects AI-driven demand in 2027 to far exceed industry supply.
-- With both big Korean memory makers now saying supply stays short, the price of the memory inside every AI computer, large or small, is set by the clusters bidding for it.
2026-10-29 / Samsung Global Newsroom / chip / central / #samsung-reports-record-107-trillion-won-profit-a
The California ISO board on Wednesday approved a proposal allowing large loads such as data centres to connect before network upgrades are complete, provided the operator can limit their use or disconnect them remotely when the grid is strained. The rules also include a framework to stop costs shifting to other customers and will be filed with federal regulators by November 16.
-- Faster hookups help data centres arrive sooner, while the right to switch them off first is how the grid operator means to keep household lights on when demand peaks.
2026-10-28 / California ISO / energy / central / #california-grid-operator-lets-data-centres-plug-
Nvidia published open weights on Wednesday for Nemotron 3.5 Transcribe, a model that transcribes 40 languages, labels up to eight speakers and handles recordings of up to an hour. The company says quantised versions are coming soon so it can run locally on its DGX Spark and Jetson AGX Thor machines, alongside a public speech dataset of more than 460,000 hours.
-- A transcription and translation model that a small office can download and run on its own hardware keeps recordings off rented servers and off the meter.
2026-10-28 / Hugging Face (Nvidia blog) / efficiency / local / #nvidia-releases-open-speech-model-and-says-quant
Axis
local [#########################-----] central
Score: 83
clusters still winning
unchanged
Alphabet and Microsoft together spent about $97 billion on capital projects in one quarter, and Samsung says AI demand for memory will far exceed supply in 2027, so the money and the parts are still flowing to the largest clusters. Alphabet's negative cash flow shows the strain of paying for it, and Nvidia's open transcription model for desk machines keeps the local side alive; the score holds at 83.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
98 | central | Alphabet spent $51.4 billion and Microsoft $45.2 billion, including finance leases, on capital projects in the September quarter; Microsoft says about three quarters went to short-lived assets, chiefly CPUs and GPUs. Samsung's chip division earned 106.5 trillion won of operating profit on 147.1 trillion won of revenue. | ||
| ENERGY |
|
97 | central | The California ISO board approved a proposal letting large loads such as data centres connect before grid upgrades are finished, provided the operator can curtail or remotely disconnect them when the system is strained; it goes to federal regulators by November 16. | ||
| EFFICIENCY |
|
74 | mixed | Nvidia released open weights for Nemotron 3.5 Transcribe, a 34-billion-parameter speech model with about 4 billion active parameters, and says quantised versions to run locally on DGX Spark and Jetson machines are coming soon; its accuracy rankings are the company's own claims. |
Line charts (0=local 100=central 2026-10-12 to 2026-10-29)
Market overlay (2026-09-30 to 2026-10-29)
Latest: BTC $80,062 / hashrate ~1091.08 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 $80,062 / hashrate ~1091 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