The Information Machine
The edition

Sunday October 18, 2026

In this edition

  1. New today
  2. NewThe Great American AI Act discussion draftGreat American AI Act criticized as step back from FRONTIER Act
  3. NewThe Genesis Mission ConsortiumEleven tech firms pledge $2.4B in AI and compute to Genesis Mission
  4. Updates
  5. Day 6Meta's Muse Wonder AI model launchMeta Releases Muse Wonder, Claims Frontier Parity With OpenAI and Anthropic
  6. Day 24Data center power demand versus grid supplyAI data center power demand to triple by 2030; supply gaps persist
  7. Day 12AI infrastructure capex and the revenue gapWall Street's 2027 Hyperscaler Capex Estimate Jumps to $1.13T
  8. Day 11RTX Spark and MXC on WindowsRTX Spark Laptops Now Shipping; NVIDIA Expands Nemotron Model Family
  9. Day 3Meta's Muse Spark Open modelMeta Releases Muse Spark Open; Non-Standard License Draws Scrutiny

New today

01
New

The Great American AI Act discussion draft

  • Representatives Jay Obernolte and Lori Trahan released a bipartisan discussion draft of the Great American AI Act of 2026, drawing immediate criticism from policy commentators including Miles Brundage, who called it 'a big step back from the FRONTIER Act'.
  • Brundage cited the lack of auditor independence, no strong corrective-action mechanisms when auditors flag major issues, and preemption of state safety laws without a federal regime able to evolve alongside AI capabilities.
  • The bill arrived during what the Future of Privacy Forum described as 'an unusually active few weeks for AI safety policy,' alongside an Illinois frontier model safety bill and a White House June 2026 executive order on voluntary pre-release model sharing.
The gist

A federal AI bill that preempts state safety laws while setting a weaker regulatory floor could reduce the overall stringency of AI oversight across the country. Whether the bill's audit and corrective-action mechanisms are sufficient is a concrete dispute shaping how frontier AI will be governed.

Continue this story →2 sources
02
New

The Genesis Mission Consortium

  • Full contribution amounts from a White House science summit are on record for eleven technology companies' $2.4 billion pledge to the Genesis Mission Consortium, a Department of Energy AI platform: Nvidia at $1 billion over five years, AMD at $500 million, OpenAI at $200 million, and Anthropic and Google each at $150 million.
  • Wiley joined as the only scientific publisher among named members, and the DOE issued a March 2026 funding round of $293.76 million across 21 technical areas.
The gist

The consortium pools substantial private AI compute and tooling across more than 15 federal agencies for scientific research in energy, health, space, and other areas. The individual company commitment amounts are now on record, showing a wide spread from $50 million to $1 billion.

Continue this story →3 sources

Updates

03
Day 6

Meta's Muse Wonder AI model launch

  • A Meta paper published October 18 studied how frontier agents learn in games including chess and NetHack, finding GPT-5.6 Sol reached 298 points of gain per $1,000 versus 57 for Claude Fable 5.
  • Meta also reported a Muse-powered health scaffold over 5 million de-identified records produced 10 manuscripts across cardiology, oncology, infectious disease, and emergency medicine.
  • Interconnects.ai placed the models, launched October 12, just behind Anthropic and comparable to OpenAI and Google, the only American company with frontier open and closed models.
The gist

Meta entering the closed frontier model tier changes the competitive landscape for both developers and enterprises choosing between providers. The gap between the stronger Wonder Full model and what developers can access via API limits immediate practical impact.

Continue this story →9 sources
04
Day 24

Data center power demand versus grid supply

  • Eric Schmidt, Broadcom CEO Hock Tan, and Intel CEO Lip-Bu Tan each argued that capital, not power or chips, is AI infrastructure's binding constraint, with Schmidt estimating roughly $50 billion per gigawatt; Morgan Stanley estimated a 76.5 GW shortfall against roughly 160 GW of projected US demand by 2029, and Amazon CEO Andy Jassy reportedly told an investor compute capacity is sold out beyond 2028.
  • H100 rental prices held at roughly $2.70 per hour since July despite falling token costs, suggesting cheaper inference expands rather than reduces total compute demand.
The gist

The gap between projected demand and realistically deliverable supply means AI compute capacity constraints are likely to persist through the end of the decade. Physical infrastructure lead times, not capital or chips alone, set the binding timeline.

Continue this story →8 sources
05
Day 12

AI infrastructure capex and the revenue gap

  • Wall Street's combined 2027 capex estimate for Amazon, Google, Meta, Microsoft, and Oracle has jumped from $495 billion to $1.13 trillion over roughly 12 months, a $636 billion increase for the same forward year.
  • Google's 2027 estimate alone rose from $103 billion to $335 billion.
  • The 2026 estimate nearly doubled over the same period, from $465 billion to $892 billion, bringing the combined 2026-2027 projection to nearly $1.9 trillion.
The gist

The gap between infrastructure spending commitments and identified revenue streams has grown to a scale that requires entirely new markets to emerge. Total debt financing for AI hardware has crossed $90 billion as companies can no longer absorb chip costs from their own balance sheets.

Continue this story →4 sources
06
Day 11

RTX Spark and MXC on Windows

  • NVIDIA's RTX Spark laptops went on sale October 16 starting at $2,399 from ASUS, Dell, HP, Lenovo, Microsoft, and MSI, with Jensen Huang making a surprise appearance at the Berlin launch event and giving away five personally signed laptops.
  • NVIDIA also released Nemotron 3.5 Super VL, a 120B-parameter open model NVIDIA claims is five times faster than the next-best in its class, with Scale AI, CrowdStrike, and Perplexity among early integrators, and Bolt E4B 4GB for on-device agents launching October 23.
The gist

RTX Spark laptops are now shipping with a full software ecosystem, making high-memory local AI inference on consumer Windows hardware a commercial reality rather than a preview. The simultaneous release of open Nemotron models targeting the same hardware expands the range of workloads available to run locally.

Continue this story →4 sources
07
Day 3

Meta's Muse Spark Open model

  • Interconnects.ai placed Muse Spark Open, the 816-billion-parameter open-weight model Meta released October 15, between GLM 5.2 and GLM 5.3/Kimi K3 among global open models.
  • The analysis also raised concerns about the model's non-standard license, describing revenue gating for companies earning over $20 million per month, branding requirements, and legal ambiguity around synthetic data use as "an unnecessary strategic risk for labs trying to compete and regain adoption".
The gist

Meta's release of an 816B open-weight model competitive with frontier closed models expands what developers can run locally or self-host. The model's non-standard license, including branding requirements and revenue gating, may limit adoption compared to more permissive alternatives.

Continue this story →7 sources
← October 17, 2026

What is moving now · Every edition · Every story

The daily email

Want this in your inbox?

I send one email each morning with the stories that moved. If you would rather just read here, that works too.

Subscribe free