Chip stocks surge as AMD tops $1 trillion, but US project resistance grows
Bernstein says AI demand has not slowed, with semiconductor earnings revised up 113%, while Jamie Dimon warns hyperscaler spending could hit $1 trillion next year. A record $68 billion in US data-center projects now sits blocked or delayed, and New Jersey fined its largest AI site $1 million over pollution.
JPMorgan's Jamie Dimon expects hyperscaler AI spending to reach $1 trillion next year. The forecast, echoed across several outlets, frames the scale of capital now committed to AI infrastructure.
Bernstein says AI demand has not slowed. Its research report analysis found semiconductor earnings revised up 113%, and shares including ARM (+17.16%) and Intel (+12.14%) rose sharply.
Resistance to the buildout is mounting. A reported $68 billion in US AI data-center projects are blocked or delayed as opposition grows, with cracks forming in the boom.
Nscale's IPO filing revealed $103.4 billion in active and contracted AI infrastructure agreements. It landed alongside Crusoe's $3.9 billion Series F at a $30.9 billion valuation and Accelevation's US IPO targeting up to $5.4 billion.
New Jersey fined its largest AI data center $1 million over pollution from natural gas generators. Separately, a Malaysian investor committed $10.5 billion to gas power for the AI rush.
Data centers · 59 storiesSites, colocation, cooling, capacity leases.Funding and land deals flow as US permitting resistance hardensNew capacity was announced across Indonesia, Australia and the Nordics, while blocked projects and a pollution fine underscored the policy friction.Frontier AI sites: power against capital costEach point is a site in Epoch AI's frontier register with both a stated power capacity and a stated capital cost — the further above the trend, the more each megawatt cost to build.BDx broke ground on an AI data center in West Java, with its first 120MW building due by early 2027.W.MediaAustralian AI data centre funding surged to a record A$35 billion, and European neocloud Verda raised $189 million in a Series B.Yahoo Finance AustraliaSiliconANGLEDataSection signed an $876 million AI infrastructure contract as builders lined up for the sector's next IPO wave.Telecompaperobserver.comOpen 59 stories →
Hyperscalers · 9 storiesCapex, power procurement, model-capacity deals.Dimon's trillion-dollar forecast dominates as Alibaba weighs hyperscaler ambitionsSpending projections framed the day, while Alibaba and Huawei detailed expansion plans.Hyperscalers, indexed over the last 10 sessionsClosing prices on each company's home exchange, indexed to 100 at the first session shown, for the 5 hyperscalers that moved most over the window.Jamie Dimon said hyperscaler AI spending may rise to $1 trillion next year.PYMNTS.commarketscreener.comAlibaba rose after giving further details on its Qwen models and full-stack AI infrastructure, prompting questions on whether it can become a serious hyperscaler.Seeking AlphaThe Daily UpsideHuawei unveiled new AI infrastructure breakthroughs as the Middle East scaled its AI ambitions.TahawulTech.comOpen 9 stories →
Semiconductors · 34 storiesAccelerators, memory, foundry, packaging.AMD tops $1 trillion as Bernstein says AI demand has not slowedEarnings revisions and India investment pledges pointed up, while ASML warned of an eastward shift.Chipmakers this window, by move8 of 8 tracked chipmakers rose over the window; the change is close to close on each company's home exchange, in its own currency.AMD reached a $1 trillion market cap, while Bernstein reported semiconductor earnings revised up 113%.TradingViewtechflowpost.comApplied Materials plans a $5 billion, decade-long investment in India's semiconductor sector, and AMD said it has invested $245 million of a $400 million India commitment.LinkedInInspur MetaBrain SD200 Ultra Packs 128 Domestic AI Chips for 2.8T Kimi K3 Under 5.85ms/TokenPandailyOpen 34 stories →
Semi equipment · 7 storiesLithography, deposition, etch, test, export rules.Renishaw posts record on chip demand as Teradyne, Applied expandTool makers reported strong demand, with new test-center partnerships and an India commitment.Toolmakers from window open to close1 of 6 tracked toolmakers closed the window above where they opened it; each line is one company's close, indexed to 100 at the open.Renishaw slightly beat FY revenue estimates on strong demand for semiconductor tools.TradingViewTeradyne partnered with GS Microelectronics on a new semiconductor test center targeting AI, silicon photonics and auto chips.TradingViewInvesting.comApplied Materials committed $5 billion over ten years to India's semiconductor sector.Global SourcesOpen 7 stories →
New research
Inference-efficiency papers dominate, with memory tiering and expert pruning claiming large serving gains
Samsung and Seoul National reported a 24× increase in concurrent agentic sessions per GPU via hot-cold HBM and flash tiering, while a Tsinghua–Huawei study cut MoE inference cost by pruning a third of selected experts for up to 1.7× speedup. Flash-dLLM claimed 5.1× on diffusion LLMs, and UC Berkeley released SWE-Serve, a 53-task benchmark for agentic production inference engineering, as code only.
Which fields movedpapers today ● vs 30-edition average ○
01234 papers
How open was it
Weights released0 / 8
Code released1 / 8
Who published0 lab · 1 mixed · 7 academic
Reported gains today, as multiples
3 of 8 kept papers report their headline result as a speed-up multiple; papers reporting in other units are not put on this axis.
The papers · headline result, scale, what was released
AI + Energy & Sustainability
Frontier AI power on the record, by country8 rows · 1 source
Stated power capacity summed across the frontier sites Epoch AI records in each country — where the grid load of the build-out actually sits.Epoch AI · Frontier Data Centers Hub
2 charts built from 5 stated figures across 20 stories, the first planned by the composer from the record. Also had the data for what today covered · topic by region, when the energy stories landed, pressure and relief by topic. Held back: power committed in deals that state a figure (no deal in this edition states the capacity it commits); power disclosed per edition · last 7 (only 0 of the last 7 editions put a number on it).