Rumor: Meta is reportedly considering purchasing Google’s TPUs.
Regardless of whether the deal materializes, Google (through Broadcom) has been raising its planned TSMC CoWoS wafer starts every month for next year.
Regardless of whether the deal materializes, Google (through Broadcom) has been raising its planned TSMC CoWoS wafer starts every month for next year.
Rumor: Snapdragon 8 Elite Gen 6 will skip TSMC’s N2 node and go straight to N2P.
Memory support will go up to LPDDR6 + UFS 5.0.
Memory support will go up to LPDDR6 + UFS 5.0.
❤2
Apple reportedly shows no interest in TSMC’s A16 process… NVIDIA likely to be the first to use the A16 node (Korean local media report)
Rumor: About Galaxy Z Fold 8
• Battery capacity in the 5000mAh range
• To improve display crease, it will adopt a laser-drilled metal plate technology, similar to the one used in the foldable iPhone
• S Pen support to be reinstated
• Battery capacity in the 5000mAh range
• To improve display crease, it will adopt a laser-drilled metal plate technology, similar to the one used in the foldable iPhone
• S Pen support to be reinstated
⚡ TSMC CoWoS, Nvidia & Broadcom
🔹 What is CoWoS?
- TSMC’s 2.5D packaging tech (Chip-on-Wafer-on-Substrate).
- Connects logic + HBM via silicon interposer → ultra-high bandwidth, low latency, power efficiency.
🔹 Nvidia
- Uses CoWoS for Hopper & Blackwell GPUs.
- Shifting from CoWoS-S → CoWoS-L for better yield & scalability.
- By 2026: ~595k wafers, ~60% of TSMC’s CoWoS capacity (510k CoWoS-L).
- Creates supply bottlenecks for others.
🔹 Broadcom
- Applies CoWoS to networking chips (Tomahawk, Jericho) & custom AI ASICs for hyperscalers.
- Faces allocation pressure due to Nvidia’s dominance.
🔹 Industry Impact
- Global demand → 1M wafers by 2026.
- TSMC expanding capacity; rivals explore Intel EMIB/Foveros & Samsung I-Cube.
✅ Bottom line:
CoWoS is the backbone of the AI boom—Nvidia leads, Broadcom competes, and TSMC holds the keys.
🔹 What is CoWoS?
- TSMC’s 2.5D packaging tech (Chip-on-Wafer-on-Substrate).
- Connects logic + HBM via silicon interposer → ultra-high bandwidth, low latency, power efficiency.
🔹 Nvidia
- Uses CoWoS for Hopper & Blackwell GPUs.
- Shifting from CoWoS-S → CoWoS-L for better yield & scalability.
- By 2026: ~595k wafers, ~60% of TSMC’s CoWoS capacity (510k CoWoS-L).
- Creates supply bottlenecks for others.
🔹 Broadcom
- Applies CoWoS to networking chips (Tomahawk, Jericho) & custom AI ASICs for hyperscalers.
- Faces allocation pressure due to Nvidia’s dominance.
🔹 Industry Impact
- Global demand → 1M wafers by 2026.
- TSMC expanding capacity; rivals explore Intel EMIB/Foveros & Samsung I-Cube.
✅ Bottom line:
CoWoS is the backbone of the AI boom—Nvidia leads, Broadcom competes, and TSMC holds the keys.
📱 Apple vs Qualcomm on TSMC’s Roadmap
Apple (iPhone 2026):
- Uses TSMC N2 (2nm) with GAAFET transistors.
- Prioritizes yield stability and maturity.
- Delays A16 (1.6nm) adoption until later (2027+).
Qualcomm (Snapdragon 8 Elite Gen 6, 2026):
- Skips N2 entirely.
- Debuts directly on N2P (enhanced 2nm).
- Plans to stay on N2P for multiple generations.
TSMC Roadmap:
- 2025–26: N2 mass production.
- 2H 2026: N2P ramp-up (better efficiency & performance).
- 2026–27: A16 (1.6nm, backside power delivery).
Strategy Snapshot:
- Apple = cautious, stepwise adoption.
- Qualcomm = aggressive leap to N2P.
Apple (iPhone 2026):
- Uses TSMC N2 (2nm) with GAAFET transistors.
- Prioritizes yield stability and maturity.
- Delays A16 (1.6nm) adoption until later (2027+).
Qualcomm (Snapdragon 8 Elite Gen 6, 2026):
- Skips N2 entirely.
- Debuts directly on N2P (enhanced 2nm).
- Plans to stay on N2P for multiple generations.
TSMC Roadmap:
- 2025–26: N2 mass production.
- 2H 2026: N2P ramp-up (better efficiency & performance).
- 2026–27: A16 (1.6nm, backside power delivery).
Strategy Snapshot:
- Apple = cautious, stepwise adoption.
- Qualcomm = aggressive leap to N2P.
❤4
🚀 NVIDIA & TSMC A16 Breakthrough
NVIDIA is set to be the first customer for TSMC’s A16 process (1.6nm-class), entering production around 2026–2027. Unlike past launches led by Apple, NVIDIA will debut its “Feynman” GPUs on A16, aimed at AI and datacenter workloads.
🔑 Why A16 Matters
- Gate-All-Around (GAA) transistors → better current control & scaling
- Backside power delivery → reduced resistance, higher efficiency
- Enables GPUs with 100B+ transistors
📅 Roadmap
- Rubin GPUs (2026–27): refined N3P nodes
- Feynman GPUs (2028): first on A16
⚔️ Impact
- Performance-per-watt leap for AI chips
- Strategic edge over AMD & Apple
- TSMC cements lead over Intel & Samsung
In short: NVIDIA’s early A16 adoption signals a new era where AI drives bleeding-edge chipmaking, securing its dominance in the AI hardware race.
NVIDIA is set to be the first customer for TSMC’s A16 process (1.6nm-class), entering production around 2026–2027. Unlike past launches led by Apple, NVIDIA will debut its “Feynman” GPUs on A16, aimed at AI and datacenter workloads.
🔑 Why A16 Matters
- Gate-All-Around (GAA) transistors → better current control & scaling
- Backside power delivery → reduced resistance, higher efficiency
- Enables GPUs with 100B+ transistors
📅 Roadmap
- Rubin GPUs (2026–27): refined N3P nodes
- Feynman GPUs (2028): first on A16
⚔️ Impact
- Performance-per-watt leap for AI chips
- Strategic edge over AMD & Apple
- TSMC cements lead over Intel & Samsung
In short: NVIDIA’s early A16 adoption signals a new era where AI drives bleeding-edge chipmaking, securing its dominance in the AI hardware race.
❤1👍1
This media is not supported in your browser
VIEW IN TELEGRAM
Today we are launching 100 early pre-orders for Mr. Tensors @eyecandyrobots, an inexpensive opensource lifelike robot character that's cool enough toy for kids (and a human-robot-interaction + reinforcement learning research tool for academia and hobbyists).
We are looking for parents and researchers to adopt our handmade cute little dinos!
The early 100 people would more or less be our design partners and early supporters to help us manufacture, ship and iterate on the product experience.
We are looking for parents and researchers to adopt our handmade cute little dinos!
The early 100 people would more or less be our design partners and early supporters to help us manufacture, ship and iterate on the product experience.
This media is not supported in your browser
VIEW IN TELEGRAM
Axon Mini by Vicharak : First Modular Single Board Computer
Ladies and gentlemen, as we all know, @Qualcomm’s IoT SoC will empower the entire ecosystem out there. We’re introducing something exceptional in the single-board computer domain, with it also we’ve also explored modularity, much like @FrameworkPuter. It’s powered by the QCS6490, delivering 12 TOPS of performance.
Ladies and gentlemen, as we all know, @Qualcomm’s IoT SoC will empower the entire ecosystem out there. We’re introducing something exceptional in the single-board computer domain, with it also we’ve also explored modularity, much like @FrameworkPuter. It’s powered by the QCS6490, delivering 12 TOPS of performance.
❤2
20th Anniversary iPhone May Replace Mechanical Buttons With Haptics