Pinned Tweet
Glad to be quoted in this excellent @Reuters article by @chernandburn & @StephenNellis on @Arm's plight to change its biz model...
reut.rs/4g4FCbe
"It was news to me that $Arm is even thinking about (making its own chip)," said @TantraAnalyst founder Prakash Sangam, who attended the trial. "It should send a chill down the spine of their customers."
As often happens, public trials expose lots of confidential information about companies. The details in the article came from its trial against @Qualcomm.
@ReutersBiz #qcomtrial
Prakash Sangam retweeted
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
Glasses are a perfect device for running AI agents for autonomous hands-free operations...
I ditched my phone at Meta Connect 2026 and spent the entire day with Meta's new AI glasses.
Also sat down with Alex Himel, Meta's VP of Wearables. The man runs his personal AI agent across 3 phones and 23 pairs of smart glasses.
Full video is live. Here's everything we covered:
00:00 Can AI glasses replace your phone? The 24-hour challenge
00:51 Morning at Meta Connect 2026
01:56 Unboxing Meta's new AI glasses
02:38 Alex Himel on why glasses are the future of AI
05:19 Exploring Meta's newest smart glasses
06:15 Trying Meta's display glasses
06:59 Live AI translation demo
08:52 The viral leaked glasses and Project Hatch
10:01 Rapid fire with Alex Himel
11:50 Meta glasses Gen 2 vs Gen 3
13:52 Secret Meta VR glasses demo
14:17 Always-on AI, privacy and the Limitless deal
17:16 Testing Meta Muse on smart glasses
19:02 Behind the scenes at Meta Connect
20:12 Zuckerberg's keynote and the Muse Charm reveal
21:23 Why computer use matters for AI agents
22:06 Muse voice mode and why Meta could win AI
23:31 Why Meta built camera-free AI glasses
24:54 Alex's AI agent setup: 23 pairs of glasses
25:47 What AGI means to Meta's VP of Wearables
26:15 The one thing he'd never let AI do
26:34 The future of AI glasses
The latest edition of my #TantrasMantra #podcast is out:
go.tantraanalyst.com/Tantras…
In this episode, I talk to Thomas Barber of @GlobalFoundries about the evolution of the #datacenter #interconnect landscape, how quickly the speed and latency requirements within racks are increasing, from 200 Gbps to 1.6 Tbps and beyond, how optical connectivity is fast becoming a basic need: the “Copper where you can and optical where you must” approach, evolution from today’s optical pluggable to Co-Packaged Optics (#CPO), and more.
We also explore the role of a foundry like $GFS in CPO, its SCALE OCI CPO Module, the proprietary vs. standard-based solutions dilemma in the optical scale-up space, and what's next.
Index:
00:00 - Intro
02:40 - Guest intro (Thomas Barber, VP Comms - Infra & Data Center)
03:00 - AI changed data center function from data storage and retrieval to high-intensity compute, and moving data within the data center
04:05 - Changed scale-up definitions - from within a server to within a rack, expanding to multiple racks
05:08 - How long will "use copper where you can and optical where you must"
hold? Will most connectivity move to optical?
06:44 - Are active electrical cables (#AEC) for short term or long term too?
07:50 - Considerations for moving to fully optical - reliability of Direct Attach Copper (#DAC) vs. link flap of optical, cost & power
10:46 - Current state of Co-Packaged Optics (CPO); Pluggable vs. CPO
13:00 - All of today's CPO products are proprietary
14:07 - GlobalFoundries' industry's first SCALE CPO OCI MSA-compliant light module
17:20 - OCI MSA only defines optical; electrical left to implementation (members: $AMD , @Broadcom, @Meta, @Microsoft, @OpenAI, $NVDA)
18:25 - Will proprietary solutions coexist or merge with standard-based solutions in the long run?
21:46 - Both @nvidia's #NVLink and open #UALink supported by @AMD & others, can use #OCIMSA optical layer for transport
22:17 - Role of Global Foundries in CPO - from @IBM Microelectronics legacy to building complex solutions at scale
26:50 - Market impetus toward standard CPO solutions
29:44 - What's next in the evolution of CPO
33:27 - AI infra is not just #GPUs, but interconnect is also critical
34:57 - Closing
@TantraAnalyst #OCPSummit26
My thoughts on the jury verdict in the @Qualcomm vs. @Arm case:
The Judgment Day: Qualcomm wins ALA and Interference/Unfair Competition Claims, Arm Prevails in TLA Claim
bit.ly/3VLECUR
tl;dr Summary:
Despite the case's complexity and the even more complex verdict form, the jury made a quick decision. Unlike the last case, they were also decisive. Qualcomm won on its two most important claims, which its lawyer said were crucial to the company's future: breach of the ALA and interference and unfair competition. As their lawyer Karen Dunn put it in the closing argument, this was a case of principles and, if nothing else, the jury should side with Qualcomm and award nominal monetary compensation. And that’s exactly what they did.
Arm prevailed on the TLA claim, which is extremely important to them, because the vast majority of their customers have TLAs. They also felt vindicated with the mere $2 of total damages. All along, their lawyers had argued that there was no significant material harm.
Not to gloat, but this is exactly what I had predicted in my last day’s blog (Check Day 5 blog bit.ly/4jaEDLL):
“…Here is my take on the jury trials, FWIW:
1. On the ALA breach case, Qualcomm has made a strong case, except for Arm’s last-minute “customization” argument…
2. On the TLA breach, …Arm has brought some interesting questions on motivation, etc., to sow doubt in the jury’s mind…
3. On the interference claim, Qualcomm has an airtight case, with solid arguments and proof…”
$QCOM $ARM #qcomarm @TantraAnalyst
Quick clarification:
Regarding the CA Unfair Competition Law (UCL) claim, jury only decided how the relationship between Qualcomm and Arm should be considered (Q6c & Q7c).
Judge will decide the UCL claim in her bench trial...
This is pretty important. The biggest challenge running multiple subagents is access.... all of them colliding and locking files, ports, and databases.
Providing a computer for each of them, with a sandbox that is a copy of the box...
Lots of subagents with such a setup can definitely beat large models...
I am on the way back home from the @Qualcomm vs. @arm trial. While the jury is cooling their heels and gets back to deliberating on Tuesday, check out what has happened so far on my #TantraNotes #blog:
bit.ly/TA-Notes
Day 1: Opening Statements, CEOs take the stand
bit.ly/4zk6nSY
Day 2: ALA Commitments and Bloomberg leak in focus
bit.ly/4ARZedV
Day 3: Focus on License Rates, Judge’s Questions on Remedy, and XXXXXX
bit.ly/3Uflcar
Day 4: More XXXXX, Companies Rest Their Case, and Other Considerations
bit.ly/4yeso4D
Day 5: Case in Jury’s Hands, Arguments Conclude in Bench Trial, My Final Take
bit.ly/4jaEDLL
$QCOM $ARM #licensing @TantraAnalyst













