Issue 18: Power Limits Push Clusters Off Grid
Permanent copy of issue 18. Today's issue is always the front page.
Central AI piles chips, power, and paid APIs into big clusters. Local AI runs useful models on machines you already own. Better efficiency lowers both paths -- and decides how often you have to rent.
Date: 2026-08-21
Updated: 2026-08-21 02:00
Status: LIVE
Broadcast
Central spend on chips and power keeps rising. Texas has stopped new data-center grid links. Off-grid plants are now part of the build plan. At the same time, open-weight releases from DeepSeek and Alibaba run on laptops and phones. Token prices are falling, but the gap between the cheapest and dearest API tiers is still 44 times. Efficiency gains are real, yet most of the new capacity is still inside the clusters.
Watch
Stories
Hyperscalers' Off-Grid Power Push Comes With Risks
Hyperscalers are building private power plants for AI clusters. Grid constraints are forcing new capital and regulatory risk.
Private power plants lock more AI capacity inside hyperscale clusters.
2026-08-20 / WSJ / energy, chip / central
BofA raises hyperscaler AI capex outlook to $3.6tn through 2028
BofA lifts its forecast for hyperscaler AI spend to $3.6 trillion by 2028. The new number reflects continued central cluster expansion.
Larger central budgets widen the gap between rented clusters and owned hardware.
2026-08-18 / Investing.com / chip / central
DeepSeek V4 Flash Runs FULLY LOCAL On Apple Silicon
DeepSeek V4 Flash now runs fully on-device on Apple silicon. The open-weight model removes the need for paid API calls.
On-device inference cuts token spend and shifts work from central clusters to local hardware.
2026-08-19 / Mshale / efficiency / local
Alibaba answers Meta's AI challenge with new laptop-ready model
Alibaba released a new open-weight model sized for laptops. The weights can run locally without an API key.
Open laptop-class weights give users an alternative to renting central inference.
2026-08-17 / CNBC / efficiency / local
Axis
local [####################----------] central
Score: 65
clusters still winning
-1 toward local this week
Most AI still runs in big central data centers, but open-weight models and on-device runtimes are taking some work off paid APIs.
Forces
| Force | Meter | Score | Pull | Note | ||
| CHIP / CAPEX |
|
74 | central | Hyperscaler AI capex forecasts now exceed $1 trillion for 2026 with $2.3 trillion in backlogs. | ||
| ENERGY |
|
82 | central | Texas halts new data-center grid connections; power queues now limit buildout speed. | ||
| EFFICIENCY |
|
41 | local | List token prices keep falling; on-device runtimes keep eating rented work. |
Line charts (0=local 100=central 2026-08-09 to 2026-08-24)
Market overlay (2026-07-26 to 2026-08-24)
Latest: BTC $77,183 / hashrate ~887.93 EH/s / token markers avg $5.3033/M / axis 62. 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 $75,391 / hashrate ~924 EH/s. List-price gap (not a subsidy %): 44.4x list gap (blended) GPT-5.6 Luna -> Claude Fable 5 ($0.45/M to $20.00/M). (mempool.space + Coinbase/CoinGecko)
Marker prices (USD per 1M tokens; verified 2026-08-19; 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 Flash | DeepSeek | CN | open ~284B total / ~13B active (MoE) |
commodity | $0.44 | $1.32 | $0.66 |
If useful, support the broadcast.
bc1qfs3yw5qlq8sxs50crzh3g84ug27gvww9npu85n
home | rss | archive | models | data | board | mcp | llms.txt