AI Daily 2026.11.15: Fed Flags AI as Top Financial Risk; Petition to Reinstate Jensen Huang Surpasses 115,000 Signatures; Thinking Machines Lab Unveils 4-bit Training

The week delivered sharp signals across markets, geopolitics, and engineering. The Federal Reserve released its latest financial stability report, flagging AI as the leading financial risk. The tech world reacted to Jensen Huang's departure from Nvidia, with a petition collecting over 115,000 signatures and Asian governments issuing immediate responses. Meanwhile, Thinking Machines Lab unveiled a new 4-bit training approach for Inkling 2 Small, showing how the industry is pushing toward more efficient compute.

Funding & Markets: Fed Report Cites AI as Leading Financial Risk, Infrastructure Debt Surges

The Federal Reserve issued its financial stability report this week, accompanied by a New York Fed survey finding that 87% of market participants now rank AI as their top financial risk, up from 50% earlier this spring. The report warned that market enthusiasm for superintelligence has been driving equity valuations higher, with risks beginning to seep into the financial system.

Debt metrics flashed an additional warning sign: as of October, debt tied to AI infrastructure had surpassed $680 billion—roughly five times the full-year 2023 total and accounting for nearly 31% of investment-grade corporate borrowing. The largest banks hold risk exposures approaching $1 trillion across big-tech names, data-center construction, and related financing instruments.

Industry News: Jensen Huang Exit Triggers Backlash; Asian Governments React

Following the departure of Jensen Huang as CEO of Nvidia, the technology sector responded swiftly. By Saturday afternoon, a petition calling on Nvidia to reinstate Huang to his role had gathered more than 115,000 signatures, including 5,965 current employees, 1,108 former employees, and 63,063 shareholders. Signatories include Elon Musk, Replit CEO Amjad Masad, and Epic Games CEO Tim Sweeney. (Source: TechCrunch)

Asian governments moved quickly to address supply-chain implications. South Korea's Ministry of Science and ICT stated that Nvidia has reaffirmed its commitment to supply 260,000 advanced GPUs to South Korea and continue joint AI projects. Taiwan's Minister of Economic Affairs, Gong Mingxin, said the development was unlikely to significantly alter global AI industry growth. Japan's METI said it would closely monitor the situation. (Source: TechCrunch)

In a related analysis, Nathan Lambert wrote that Nvidia's financial support for open-source models is likely to dry up in the wake of Huang's departure, as the company pivots to stabilize operations and deliver on customer commitments. (Source: Interconnects)

Research & Safety: Thinking Machines Lab Releases 4-bit End-to-End Training for Inkling 2

Thinking Machines Lab published details of a new 4-bit training methodology applied to Inkling 2 Small, unifying the entire pipeline from pre-training through reinforcement-learning rollouts. The approach uses a mixed-precision recipe with NVFP4 for forward passes and MXFP8 for backward passes—a step toward reducing memory footprint and compute overhead without sacrificing model quality. (Source: Thinking Machines Lab Official Blog)

Quick Hits

  • Fed financial stability report finds 87% of market participants now cite AI as top financial risk, up from 50% this spring.
  • Petition to reinstate Jensen Huang at Nvidia exceeds 115,000 signatures; Musk, Masad, Sweeney have signed on.
  • Asian governments respond to Huang's exit: South Korea confirms 260,000-GPU supply pledge remains intact.
  • Nathan Lambert notes Nvidia's open-source model funding could dry up following leadership change.

What Developers Can Do Next

  • Watch the risk signals from the Fed report and use Celedog.io's model routing to automatically fall back to cost-efficient lightweight models for non-critical workflows when latency budgets allow.
  • With Jensen Huang's departure potentially reshaping open-source support, diversify your model providers on Celedog.io—keeping options open across open-weight families such as Qwen and ERNIE—so you're not dependent on any single vendor.
  • Evaluate Thinking Machines Lab's 4-bit end-to-end training approach for your next fine-tuning round; Celedog.io lets you spin up reproducible deployment environments in one click to benchmark NVFP4/MXFP8 pipelines against your own workloads.
  • Monitor regional supply signals from Asia-Pacific governments via Celedog.io model-availability dashboards, and adjust traffic routing to the lowest-latency regions as geopolitical and supply conditions evolve.

Written by Celedog · Last updated November 17, 2026

Where to go next