# NVIDIA earnings land as Anthropic IPO prices and AMD arms HPC

> NVIDIA reports Q3 after the bell, Anthropic prices its IPO Wednesday, AMD open-sources an HPC assistant, and Cerebras ships Tungsten for wafer-scale code.

Source: https://ai.dosa.dev/blog/today-in-ai-2026-11-17
Published: 2026-11-17
Author: Dosa AI Tools

---

This brief covers AI news from 2026-11-17 UTC.

Earnings, an IPO and a pair of open-source pushes for scientists: here is what landed for builders today.

## NVIDIA reports Q3 today with a $108B guide on the table

**What happened:** NVIDIA releases fiscal Q3 results after the close today, with a 5 p.m. ET call. The company guided $108.0B revenue plus or minus 2% at a 74.0% gross margin; consensus sits near $109B and $2.47 EPS. The prior quarter was a record $96.2B, up 18% quarter over quarter.

**Why it matters:** The Q4 guide sets expectations for GPU supply and pricing that teams building on rented compute plan their budgets around.

[Source: AI Daily Dev](<https://www.ai-daily.dev/digest/2026-11-17/>)

## AMD open-sources an HPC assistant for LUMI and Frontier

**What happened:** AMD released AMD HPC Assistant, an open-source AI assistant that brings system-aware HPC knowledge into the coding tools researchers already use. It debuts on the LUMI and Frontier supercomputers and understands system directories, storage quotas, purge policies and scheduler requirements.

**Why it matters:** If you run jobs on shared clusters, an assistant that knows the scheduler and quota rules cuts the trial and error before a job ever queues.

[Source: AMD](<https://www.amd.com/en/blogs/2026/amd-brings-agentic-ai-to-supercomputing-with-open-source-amd-hpc.html>)

## Cerebras ships Tungsten to program wafers at a high level

**What happened:** Cerebras released Tungsten, a high-level language and compiler for its Wafer-Scale Engine, alongside Aviary, a portfolio of scientific applications and tutorials. The company says a plasma simulation written entirely by a coding agent ran across 98% of a wafer.

**Why it matters:** Moving wafer resource allocation from manual intrinsics to compiler passes means scientific code can be ported and resized without rewriting kernels by hand.

[Source: Unite.AI](<https://www.unite.ai/cerebras-releases-tungsten-for-high-level-wafer-scale-programming/>)

## NVIDIA bundles Nemotron for Science with an open chemistry set

**What happened:** NVIDIA introduced Nemotron for Science, combining the Nemotron 3.5 Super VL open model, post-training recipes and LitmusBench v0.2, an open dataset of nearly 50, 000 chemistry reasoning questions. Argonne National Laboratory and life sciences companies Lila Sciences, Terray Therapeutics and Therna are using it.

**Why it matters:** An open chemistry benchmark plus recipes gives teams a starting point for domain agents instead of building evaluation sets from zero.

[Source: NVIDIA](<https://nvidianews.nvidia.com/news/nvidia-introduces-nemotron-for-science-to-accelerate-discoveries-with-agents>)

## TokenEconomizer cuts coding-agent costs up to 30.9%

**What happened:** Microsoft Research published TokenEconomizer, an open-source plugin from UC Berkeley, Microsoft Research and Microsoft Security that learns from an agent's past sessions in a repository. Across 277 real bug-fixing tasks it cut costs 18.4% to 30.9% across multiple agents, with success rates broadly stable.

**Why it matters:** It works without modifying the agent or its model, so teams can drop it into existing coding workflows to trim token spend per repository.

[Source: Microsoft Research](<https://www.microsoft.com/en-us/research/blog/tokeneconomizer-learning-from-experience-to-make-ai-coding-agents-more-efficient/>)

## Google posts 18.02-minute result on MLPerf coding-agent training

**What happened:** Google says 640 Ironwood chips completed MLPerf's first coding-agent training benchmark in 18.02 minutes, reaching a 0.765 software-repair score against a 0.69 target, with training chips idle under 3% of the time. The benchmark caps each attempt at 30 turns and a 65, 536-token context.

**Why it matters:** The benchmark scores time to a quality bar rather than raw throughput, giving infrastructure buyers a different number to compare systems on.

[Source: Superpower Daily](<https://superpowerdaily.com/posts/google-publishes-ai-training-results-for-mlperf-s-first-coding-agent-benchmark>)

## Digital Science acquires Jenni AI and its 8 million users

**What happened:** Digital Science acquired Jenni AI, an AI-native research and writing platform used by more than 8 million students and researchers in 189 countries. Jenni AI will keep operating as a standalone product and is expected to connect with Dimensions, Digital Science's database of 171 million publications.

**Why it matters:** Citation checking against the published record is the hard part of research tooling, and tying Jenni to Dimensions puts that data behind its writing workflows.

[Source: RSM Ebner Stolz](<https://www.ebnerstolz.de/en/financial-and-tax-due-diligence-for-digital-science-115822.html>)

## Sources

- [AI Daily Dev](<https://www.ai-daily.dev/digest/2026-11-17/>)
- [AMD](<https://www.amd.com/en/blogs/2026/amd-brings-agentic-ai-to-supercomputing-with-open-source-amd-hpc.html>)
- [Unite.AI](<https://www.unite.ai/cerebras-releases-tungsten-for-high-level-wafer-scale-programming/>)
- [NVIDIA](<https://nvidianews.nvidia.com/news/nvidia-introduces-nemotron-for-science-to-accelerate-discoveries-with-agents>)
- [Microsoft Research](<https://www.microsoft.com/en-us/research/blog/tokeneconomizer-learning-from-experience-to-make-ai-coding-agents-more-efficient/>)
- [Superpower Daily](<https://superpowerdaily.com/posts/google-publishes-ai-training-results-for-mlperf-s-first-coding-agent-benchmark>)
- [RSM Ebner Stolz](<https://www.ebnerstolz.de/en/financial-and-tax-due-diligence-for-digital-science-115822.html>)