Marvell Technology will acquire Lumentum on November 2, 2026, joining electrical and optical AI connectivity and lifting the combined addressable market to $550 billion1, the companies say.
Koda editorial team · 3 November 2026 · 6 min read
Listen · 15 min
Today: Make your agent workflows survive long, messy runs. The steps
Marvell Technology agreed on November 2, 2026 to acquire Lumentum in an all-stock transaction that combines electrical and optical AI connectivity. The companies say the deal expands the combined total addressable market to $550 billion1, gives line of sight to more than $1 billion in annual run-rate cost synergies, and adds to Marvell non-GAAP EPS in year one post-close.
Why it matters. The $1 billion1 synergy claim carries no stated timeline; check it against the closing date and Marvell's year-one EPS accretion promise before pricing in any savings.
Go further: today's paper, How I Aligned Your Model: Scaling Up Multilingual Safety Training with Templates (arXiv, 29 Oct).
Google's Gemini 4 Carbon leads SEC-bench Pro at 88.2%2, in the same week Anthropic prepares Claude Mythos 5.5.
Why it matters. Security teams choosing a model for vulnerability hunting now have a clear leader in Gemini 4 Carbon, so Claude Mythos 5.5 must top 88.2%2 on SEC-bench Pro to change that choice.
Anthropic is extending invisible text watermarking to older Claude models starting November 3. Coverage includes Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6.
Why it matters. Audit your disclosure policy before shipping Claude text after November 3 if you pinned Opus 4.6 or Sonnet 4.6, because those outputs will now carry detectable marks.
SemiAnalysis launched GDPx, an agent benchmark spanning 11 high-value domains from chip design to legal work. The best model scored just 12%3, with Claude Opus 5.5 leading overall.
Why it matters. Companies hoping to hand chip design or legal work to autonomous agents should keep humans in the loop, since even the leader, Claude Opus 5.5, fails most GDPx tasks.
The Rundown argues Claude changed more this fall than in the previous two years. Claude merged Chat and Cowork into one app, put Docs, Slides and Design inside the conversation, and now runs long tasks in the cloud while the user's laptop is closed.
Meta built Max, a 5ft, roughly 106 lb humanoid prototype designed for household chores, in three months with Chinese robotics firm Fourier, according to The Information. Max has a face display, a built-in speaker and hands for picking up small objects.
Subvocal's glasses use radar signals sent through the face to capture micro-muscle movements as the wearer mouths words, converting them to text at 200 words per minute. The glasses support silent dictation, AI prompting and communication with other Subvocal wearers, and the demo drew more than 11M views.
Republicans have controlled both chambers since Trump took office nearly two years ago; polls now show Democrats poised to take the House and within reach of the Senate in Tuesday's vote.
S&P 500 -0.51%, the Nasdaq -1.07% and WTI crude -7.19% on the session. The crypto Fear & Greed index reads 26, Fear.
S&P 500
7,541.40
-0.51%
Nasdaq
25,164.36
-1.07%
Bitcoin
$76,971.58
-1.63%
Ethereum
$2,254.07
-0.84%
Crude Oil (WTI)
$96.63
-7.19%
Crypto Fear & Greed
26 / 100
Fear
Levels from Yahoo Finance5, 2 Nov and 3 Nov; alternative.me6, 3 Nov.
Herdr's server owns your agent processes, so quitting the app leaves Claude Code and Codex running until you return. Anyone juggling several CLI agents on one machine should install it today; adoption costs a download and one click on the + button. Treat it as a keeper of live sessions and keep reading the output yourself, because its reporting on unattended work is thin.
Faster links will not fix fragile agents .
Marvell and Lumentum are betting that AI clusters need optical and electrical connectivity under one roof, yet the 12%3 GDPx result shows the models on top of that hardware still stumble over long tasks. Better infrastructure gives you more compute, and reliability still has to come from how you build. Put checkpoints in your agent chains, keep processes persistent with a tool like Herdr, and measure on your own tasks rather than leaderboards.
Every story links to its outlet. Below, by source, are the 12 figures we checked: 1 verified, 9 reported and 2 estimated.
Verified confirmed by an independent searchReported as published by the named sourceEstimated approximate, in the source or in our check
Signal 221 · 3 November 2026 · 14 sources · · Corrections
220 editions so far, one a day. Unsubscribe anytime.
Forward this to one operator you work with. Your referral link is in every email; milestones at koda.community/refer.