Deepsemi

Semiconductor | Technology | Silicon Valley | Startup

[Deepsemi @Deep_semi]

Government of India (GoI) has decided to invest $10B for semiconductors. This scheme comes as USA is still waiting to pass the CHIPS act.

Below is what I think GoI should do with this scheme and further such schemes 🧵

https://t.co/Aiu7lbNqiP

[Deepsemi @Deep_semi]

$GFS gets $300M to CPO

[Deepsemi @Deep_semi]

https://t.co/WAfgTvU8v0

CHIPS act money put to good use

[Deepsemi @Deep_semi]

This is called domination!
No wonder all the top talents in the world are going here right now!

If in 3 months we don't see $GOOG / OpenAI or open source come up with something competitive, then Anthropic is going to runaway with this.

[Deepsemi @Deep_semi]

For those confused, Alphawave had a growing ASIC business before acquisition

[Bloomberg @business]

Qualcomm reached a deal with TikTok owner ByteDance to supply chips for artificial intelligence data centers, according to people familiar with the matter https://t.co/jimjquWnQJ

[Deepsemi @Deep_semi]

Some interesting roadmaps from TEL

I predict HBM will move to hybrid bonding on HBM5E generation potentially sooner if demand holds

[Deepsemi @Deep_semi]

@runaway_vol What about Samsung? If Intel can make memory, Samsung can design a GPU

[Deepsemi @Deep_semi]

@jukan05 Alphawave relationship is the gateway

[Deepsemi @Deep_semi]

It's Smart NIC , now known as IPU

[Aaron @Aaronwei3n]

Wow, $INTC to supply server CPU to $GOOGL!

[Deepsemi @Deep_semi]

Something we all expected. Terafab will also announce collaboration with DRAM manufacturer and maybe a packaging company as well

[Intel @intel]

Intel is proud to join the Terafab project with @SpaceX, @xAI, and @Tesla to help refactor silicon fab technology.

Our ability to design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab’s aim to produce 1 TW/year of compute to power future advances in AI and robotics.

It was fun hosting @elonmusk at Intel this past weekend!

[Deepsemi @Deep_semi]

@dylan522p @var_epsilon @dylan522p - By this flawless logic, you must be related to Kash Patel as well 😅

[Deepsemi @Deep_semi]

How government uses $PLTR, quite impressive

[Bilawal Sidhu @bilawalsidhu]

Probably the most current look at Palantir’s maven smart system software. Here’s the DoW’s Chief AI officer showing how it works:

[Deepsemi @Deep_semi]

While $AAPL launched MacBook pro today with M5 series, all my friends are just refreshing the Mac studio page to see when that will get an update. $AAPL has surprising product market fit for AI using Mac studio, by far the best option to run LLM locally
https://t.co/74RlXH0Bvo

[Deepsemi @Deep_semi]

@austinlyons You forgot TPUs , pretty sure they will use that as well. People are not talking yet about CoWos capacity forcing some of these decisions

[Deepsemi @Deep_semi]

$META has always been a big $AMD buyer. While announcing deals is in fashion these days, Meta has been buying lots of instinct over the years. Infact, $META is quite open to new hardware, I bet they would have tried out falcon shores as well if it existed $INTC

[AI at Meta @AIatMeta]

Meta 🤝 AMD

Today we’re announcing a multi-year agreement with @AMD to integrate their latest Instinct GPUs into our global infrastructure. With approximately 6GW of planned data center capacity dedicated to this deployment, we’re scaling our compute capacity to accelerate the development of cutting-edge AI models and deliver personal superintelligence to billions around the world.

Learn more: https://t.co/GFggxfNZ8w

[Deepsemi @Deep_semi]

The market keeps looking for the AGI/ASI while ignoring that the 'killer app' is currently just the build-out itself. We’ve moved from the era of 'software is eating the world' to 'hardware is constraining the world.' If you aren't watching the lead times on semi supply chain, you aren't watching the cycle.

[Deepsemi @Deep_semi]

@taalas_inc For those who are curious

[Deepsemi @Deep_semi]

@dylan522p Infer-no ?

[Deepsemi @Deep_semi]

SK , Samsung and Micron will sell all the HBMs they can make through 2027.

Demand is so high that even if one companies memory is less competitive, it will still sell out

$MU

[Ben Bajarin @BenBajarin]

$MU execs dispelling any myth's about their HBM business. Calls recent reports "innaccurate reporting" and states all HBM for 2026 is sold out.

Demand is more than all of memory capacity and sounds like HBM 4 is fine and on time.

[Deepsemi @Deep_semi]

Between this and $AMKR earnings, I am convinced that both CoWos and EMIB capacity is probably completely booked until 2027-2028 .

Packaging companies will become very critical piece in the supply chain

$INTC $TSM $ASX

[Jukan @jukan05]

UBS’s TPU report — what surprised me:

1. In the TPU v8 generation, MediaTek’s ASIC service fee per TPU chip is less than half of Broadcom’s.

2. Google appears satisfied with MediaTek’s 224G SerDes IP quality, even though it still doesn’t fully match Broadcom’s.

3. Not only MediaTek’s TPU v9X, but also Broadcom’s TPU v9AX could adopt Intel’s EMIB solution.

4. Ibiden is entering Google TPU substrate business in a meaningful way.

5. Google TPU is expected to skip HBM4 and move directly to HBM4E.

[Deepsemi @Deep_semi]

Here's what this means:
- Hybrid bonded multi layer stack
- WoW (Wafer on Wafer) stacking instead of singulated dies
- "Via in one" uses one shot etch that goes across the wafers, this is fundamentally new.
- Z memory probably refers to direction of stacking and sidewall angle of the etched via
- This stack reduces power compared to HBM due to absence of microbumps and reduced resistance due to via in one

Reminder- HBM is supposed to move to hybrid bonds in the same time frame, around 2029/2030. So the added benefit will only be from via in one

$INTC $SFTBY https://t.co/bgFbbXHxls