- Believes every entry car will be L2+ over time - Auto cloud business is managed by the auto team internally but reported in DC - Auto OEMs will likely develop in-house capability in 20 years, not near-term

- Full stack, flexible and modular auto platform - from car design, SoC and software to the factory floor and retail design - $14b pipeline for the next 6 years; a mix of software and chip revenue - Works with Tesla on the cloud side - Focus on L2+ or higher

NVIDIA auto latest - MediaTek partnership about installed base growth/not exclusive to MediaTek or auto - data center roadmap features pulled into auto (multi-instance GPU) - Software compatible across generations (Thor, Orin, Xavier etc.); auto stack compatible with CUDA APIs

Synopsys on multi die silicon complexity. Says it increases IP complexity as vendors use various nodes from various foundries. Says it has 7,500 IPs across 340 nodes. (comments at the BofA conference)

Link:

Microchips claims to have $100k+ semiconductor content per commercial aircraft. Would have never guessed it. Aviation is 8-9% of revenue. Also, says it has 88 chips in Hyundai Ioniq 6 car.

TSMC's IP library portfolio grew 37.5% to 55,000 in 2022 (from the annual report). Samsung and GlobalFoundries each have 4000+ IP libraries, per their disclosures. IP libraries help chip designers integrate various functionalities into their designs.

Updates financial model. Eur 2.1 b revenue by FY26. Auto and industrial portion to grow from 13% in FY23 to 20% by FY26.

Wafer capacity will by 1.5 m to 4.5 m by FY26. Half of it driven by 300mm.

Soitec’s content opportunity: 200mm2 in smartphones and 2500mm2 in auto.

Says Emerald Rapids, Sierra Forest and Granite Rapids on track.

Intel on the importance of software investments and how it offers performance boost vs process and architecture improvements.

Qualcomm aims to get revenue from software upgrades on its auto chips. A new revenue avenue.

Qualcomm says it no longer uses smartphone chips for auto and has a dedicated roadmap now for auto chips.

Xiaomi on Oppo's chipset division shutdown ''And for our peers, based on their own strategies, they have decided to terminate this business. And we all feel that it's quite a pity, but it fully reflects the difficulty that the chips business would present.''

Intel will have a foundry webinar in later June. Likely to separate profit and loss model for foundry and design operations. Already communicated this earlier. Official segmenting next year, perhaps.

Broadcom's Hock tan on foundry relationships ''we don't move just because of prices. We stay put because of support, service and abiding sense of -- a very abiding sense of commitment mutually.'

- Broadcom says most of AI revenue comes from networking and will cross $1b+ in 3Q - Non-AI revenue growth implies low single-digit but still growing - No revision in ASIC revenue target ($3b) - Says standard lead time for its products is 50 weeks

Broadcom 2QFY23: - Semiconductor revenue down -4% q/q & 9% y/y to $6.8b -Semiconductor gross margin: 71%; OPM: 59% - Except for wireless, all other segments up y/y - 3Q semi revenue will grow 5% q/q and 8% y/y to $7.1b - AI revenue: 10% of FY22; 15% of FY23; 25%+ of FY24

Fwiw, Intel CFO comments at the Cowen conference that it was able to execute a private enterprise LLM model using Gaudi and Sapphire Rapids. Claims Gaudi pipeline up 2.5x.

Amon claims Qualcomm is the largest customer of Arm. Nuvia chip specs will be out at Tech Summit this year, per Amon.

Qualcomm’s Cristiano Amon talks about hybrid LLM models on the client side. 10 b/20b parameter LLMs on phones and PCs and 60b on auto chips. ( comments at the Bernstein conference)

Ambarella claims it is evaluating AI server market and claims its performance is as good as AI100.

Ambarella 1QFY24: revenue down -25% q/q and -31% y/y to $62m. Gross margin: 63.1%. 2Q revenue guidance: $60-$64m.

Intel says 18A getting to structural cost parity with the best (comments from 1Q23 earnings call).