Latest

6/recent/ticker-posts

Header Ads Widget

Jensen wants to return 🔙, NVIDIA board says no 🚫, Tesla's Powercore ⚡

In his first statement since resigning, Jensen Huang said he made serious mistakes, will cooperate with the board’s independent ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

TLDR

Together With SiMa.ai

TLDR Hardware 2026-11-18

Building Physical AI? Run vision models, VLMs and LLMs concurrently on one chip (Sponsor)

Palette 3.0 is SiMa.ai's new software release for running any Physical AI model on one Modalix MLSoC within strict power limits.
With it, developers build complex Physical AI applications in days/hours, not months.

Join the team behind it in our live webinar to see:

  • Agentic camera (ISP) tuning that onboards new sensors without recompiling code
  • LLMs, VLMs and vision models running side by side on one chip
  • Speculative decoding that raises tokens per second by up to 3.5x
  • GPT-OSS-20B on the 16GB Modalix SoM

See it live on December 2 or check out the launch hub

📱

Headlines and Launches

Jensen Huang says he would be “honored” to return as Nvidia CEO if the board decides it is in the company's best interest (2 minute read)

In his first statement since resigning, Jensen Huang said he made serious mistakes, will cooperate with the board's independent investigation, and would return if the board decides it serves shareholders and the company. Nvidia has formed an executive committee to oversee export control compliance, and Huang will testify before Congress on December 1. A source on Nvidia's board told the Washington Post that there are no plans to reinstate him, and one senator called the idea among the worst he'd heard.
Nvidia guides Q4 to $130B with no China data center revenue as Vera Rubin ramps, and the board rules out Huang's return (3 minute read)

Nvidia's Q3 revenue hit $115.8 billion, with data center up 110% to $107.6 billion and Vera Rubin NVL72 racks shipping to CoreWeave, Google Cloud, Azure, Oracle, and Nebius. Q4 guidance of $130 billion, plus or minus 3%, assumes no China data center compute revenue, and gross margin dips to 71% to 72%. On the call, board chair Harvey Jones said the board does not intend to reinstate Jensen Huang, and a search firm is leading the CEO hunt.
🔨

Engineering and Applications

Tesla's Powercore packs transformer, inverter, and breaker into one 7.4 MW solid-state unit, taking on Heron Link (3 minute read)

Powercore uses silicon carbide power electronics to replace the low-voltage inverters and oil-filled iron-core transformer in a solar or battery site. It's rated 7.4 MW (9 MVA) continuous at 34.5 kV with grid-forming support up to 10 MVA peak, made in the US, operating in 2027. That's about 50% more than the 5 MW Heron Link from ex-Tesla exec Drew Baglino's startup, which claims 98.6% efficiency, a Morgan Hill factory, and customers like Vantage and Intersect. Tesla hasn't shared efficiency or pricing.
Backplane is an alpha, open-source KiCad fork where AI agents design boards through KiCad's IPC and CLI (1 minute read)

Agents drive KiCad directly through kicad_ipc and kicad-cli instead of rewriting files, so every board stays human-editable. Persistent bots run in threads with memory, routines, and webhooks. The built-in KiStack adds 11 auto-updating skills for schematics, placement, layout, panelization, BOMs, and fab outputs, with the export skill running ERC and DRC and generating gerbers, drill, pick-and-place, STEP, and renders. The demo is an STM32N657 SoM, and it ships with a 199-footprint library.
🧠

Research and Prototypes

Plaid built $382K in pipeline from one TLDR placement (Sponsor)

Close Q4 with proof. TLDR reaches 8 million tech readers across 14 newsletters. Book a Q4 campaign.
A 125-qubit chip teleports an error-corrected qubit using weaker entanglement than expected (3 minute read)

Teleporting a protected qubit between quantum processors needs a good-enough entanglement link, and below a certain quality it fails. By balancing the two kinds of errors in the code, researchers made their 125-qubit chip tolerate a much weaker link, raising the threshold from 0.122π to 0.175π. Theory suggests even very weak entanglement could eventually work, which would make linking error-corrected quantum computers easier.
Isomorphic Labs raises $5B to scale its IsoDDE drug-design engine and foundational biology models toward clinical trials (2 minute read)

The Alphabet-backed company raised a Series C co-led by Greenoaks and Sequoia, six months after a $2.1 billion Series B, and Bloomberg earlier reported a roughly $50 billion valuation. The money funds IsoDDE, its proprietary design engine, plus new foundational biology models and its first therapeutic programs moving toward clinical trials. Isomorphic reports positive preclinical results in oncology and immunology, with partners including Eli Lilly, Novartis, and Johnson & Johnson.
🎁

Miscellaneous

Waymo halts freeway rides in all 15 cities after California DMV cites recurring construction-zone incursions (3 minute read)

California's DMV restricted Waymo's freeway operations on November 11 after three robotaxis entered an active construction closure on US-101 near Burlingame and a separate incident involved CHP vehicles on Highway 280. The DMV calls it an imminent hazard, since similar incidents on May 18 led to a prior restriction and fixes. Waymo has halted freeway service in every market while it evaluates performance, and surface-street rides, over half a million trips a week, continue. The restriction will stay until the DMV sees effective corrective action.
Marvell and Penguin's photonic memory appliance pools 16 TB of DDR5 and 1 TB of HBM across 16 servers over optical links (3 minute read)

Marvell's module, from its $3.5B Celestial AI acquisition, stacks 72 GB of HBM3e and eight DDR5 channels on a silicon photonic interposer with 900 GB/s bidirectional bandwidth. Penguin's box holds 16 modules behind a 57.6 Tbps optical shuffle, serving up to 16 servers over NICs or CPO and NPO optics, bypassing CXL for the optical modes. An HBM cache-through mode gives HBM speed with DDR5 capacity, aimed at KV cache offload, with systems due in the second half of next year.
⚡

Quick Links

Genesis AI raises $500M at $3.5B to build Eno, a headless wheeled robot with in-house actuators, gearboxes, and hands (3 minute read)

Eno skips the humanoid form for three card-like body segments on a rotating wheeled base, and Genesis designs its own actuators, motors, gearboxes, and hands while running on Nvidia chips.
ULA says Vulcan's first-stage booster lost performance, and Centaur couldn't recover, as investigators weigh the BE-4 (3 minute read)

ULA says the solid boosters were nominal and the problem began in the first stage, while Centaur's RL-10C engines did their job but couldn't make up the loss.
iPhone Duo's thin foldable body has no room for a TrueDepth array, so Apple adds front-camera face unlock in iOS 27.3 beta 2 (2 minute read)

The feature uses the outer display's ordinary camera instead of Face ID hardware, so it only unlocks the closed phone and still needs Touch ID after about 45 minutes or for Apple Pay.

Want to advertise in TLDR? 📰

If your company is interested in reaching an audience of hardware professionals and decision makers, you may want to advertise with us.

Want to work at TLDR? 💼

Apply here, create your own role or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! TLDR is one of Inc.'s Best Bootstrapped businesses of 2025.

If you have any comments or feedback, just respond to this email!

Thanks for reading,
Andrew Tan, Adi Nikumbh, Kshitij Sudan, & Kat Scott


Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR Hardware isn't for you, please unsubscribe.

Post a Comment

0 Comments