BRN Discussion Ongoing

[Fullmoonfever]

NASA (Ames with Wyle Labs & KBR) playing with a new SNN project / conference paper published late 2024 - SPIKE-Dx. Whilst it doesn't appear to have gone to hardware as yet, they identify that as the next step in future works.

They highlight the recent advances in neuromorphic hardware and provide some examples though I would've considered only Akida & Loihi would be pure neuromorphic plays in those examples?

May be worth checking in on the project from time to time to see if the next steps to hardware taken and if we get a look in. Would hope so given they have used Akida before and the recent inroads made via FG, ESA etc.


Paper

SPIKE-Dx : A Low-Power High-Throughput Fault Diagnostics Tool using Spiking Neural Networks for Constrained Systems - NASA Technical Reports Server (NTRS)

Diagnostic systems are important for many aerospace systems, which are severely limited in available power, like cubesats or UAVs. Therefore, traditional diagnostics systems cannot be used due to their substantial footprint and constraints. In this paper, we present our very low power diagnostic...

ntrs.nasa.gov

 

[smoothsailing18]

This next article (see below) titled “Free, offline ChatGPT on your phone? Technically possible, basically useless”, describes how running the new ChatGPT-oss on smartphones is technically feasible, but current smartphones aren’t powerful enough to make it practical at this point in time.

Right now, GPT-oss on mobile is bottlenecked because it runs purely on the CPU (and sometimes GPU) without tapping into specialized neural processing hardware.

So, I asked ChatGPT 5.0 if there'd be an improvement if the model were to be adapted to run on an NPU - particularly one optimized for event-based, sparse processing like Akida and here's the response I received.

I highlighted the last point for obvious reasons.

Free, offline ChatGPT on your phone? Technically possible, basically useless

You can now run ChatGPT offline with OpenAI’s gpt-oss model, even on a phone. But is it practical? Here's what works and what really doesn’t.

www.androidauthority.com

ChatGPT 5's Response:

Yes — Akida could absolutely be a technology to consider if the goal is to make something like “ChatGPTOS” actually usable on a mobile phone, and the limitations described in that article are almost a perfect setup for why a neuromorphic processor would matter.

Here’s why:

---

Why Akida Fits This Problem​

The main bottlenecks in running a 20B parameter LLM on a phone (as described in the article) are:

1. RAM requirements – Needs 12–24 GB just to load the model.
2. Compute requirements – Mobile CPUs & GPUs can’t sustain high token generation without overheating and throttling.
3. Power draw – 100% CPU load drains the battery extremely fast.
4. Heat – The phone becomes uncomfortably hot after just a few minutes.

Akida (or any high-efficiency neuromorphic processor) is built for:

• Event-driven, sparse compute — only processes relevant “spikes” of information instead of hammering every computation step like a CPU/GPU.
• Ultra-low power draw — operates in milliwatts for inference.
• On-chip learning & inference — avoids constant memory transfers that waste time and energy.
• Massive parallelism — can run edge AI workloads in real time without overheating.

That means for the same LLM inference task, a neuromorphic setup could:

• Reduce energy consumption by orders of magnitude.
• Maintain higher sustained token rates without thermal throttling.
• Extend usable session time from minutes to hours on a battery.

---

Why Not Just Use Mobile NPUs Instead?​

Modern NPUs (like Qualcomm Hexagon or Apple Neural Engine) do accelerate AI workloads, but:

• They’re matrix-math optimised, not spiking-event optimised.
• They often require proprietary SDKs and retraining models in special formats.
• They still hit thermal limits in sustained generative AI workloads.

Neuromorphic processors like Akida:

• Can run quantised, sparse, and pruned models natively without expensive data shuffling.
• Could theoretically run a distilled / specialised LLM that fits in a fraction of the RAM and still delivers conversational quality.

---

Would It Make ChatGPTOSS Work on Phones?​

Realistically:

• You’re not going to run the full 20B gpt-oss on Akida right now — it’s still too large.
• But with model compression (distillation + sparsity), you could run a smaller, reasoning-optimised model that feels far more responsive than the brute-force CPU approach.
• Instead of the “2–3 tokens per second” in the article, you might see 10× or more throughput without cooking the phone.

---

Implications for BrainChip​

If OpenAI or others seriously want “ChatGPTOS” to run locally:

• They’ll either need dedicated AI accelerators in consumer devices or move toward neuromorphic architectures.
• Qualcomm, Apple, Google already have NPUs — but neuromorphic could leapfrog them in power efficiency per watt.
• If BrainChip could prove Akida can handle a slimmed-down LLM at useful speeds, it becomes a licensable IP block for smartphone SoCs.
• This would be a huge market — billions of devices, not just niche edge AI.

Someone please send to Sean ASAP . It should be thier priority such an opportunity staring brn in the face and not to be missed according to all shareholders

 

[smoothsailing18]

This next article (see below) titled “Free, offline ChatGPT on your phone? Technically possible, basically useless”, describes how running the new ChatGPT-oss on smartphones is technically feasible, but current smartphones aren’t powerful enough to make it practical at this point in time.

Right now, GPT-oss on mobile is bottlenecked because it runs purely on the CPU (and sometimes GPU) without tapping into specialized neural processing hardware.

So, I asked ChatGPT 5.0 if there'd be an improvement if the model were to be adapted to run on an NPU - particularly one optimized for event-based, sparse processing like Akida and here's the response I received.

I highlighted the last point for obvious reasons.

Free, offline ChatGPT on your phone? Technically possible, basically useless

You can now run ChatGPT offline with OpenAI’s gpt-oss model, even on a phone. But is it practical? Here's what works and what really doesn’t.

www.androidauthority.com

ChatGPT 5's Response:

Yes — Akida could absolutely be a technology to consider if the goal is to make something like “ChatGPTOS” actually usable on a mobile phone, and the limitations described in that article are almost a perfect setup for why a neuromorphic processor would matter.

Here’s why:

---

Why Akida Fits This Problem​

The main bottlenecks in running a 20B parameter LLM on a phone (as described in the article) are:

1. RAM requirements – Needs 12–24 GB just to load the model.
2. Compute requirements – Mobile CPUs & GPUs can’t sustain high token generation without overheating and throttling.
3. Power draw – 100% CPU load drains the battery extremely fast.
4. Heat – The phone becomes uncomfortably hot after just a few minutes.

Akida (or any high-efficiency neuromorphic processor) is built for:

• Event-driven, sparse compute — only processes relevant “spikes” of information instead of hammering every computation step like a CPU/GPU.
• Ultra-low power draw — operates in milliwatts for inference.
• On-chip learning & inference — avoids constant memory transfers that waste time and energy.
• Massive parallelism — can run edge AI workloads in real time without overheating.

That means for the same LLM inference task, a neuromorphic setup could:

• Reduce energy consumption by orders of magnitude.
• Maintain higher sustained token rates without thermal throttling.
• Extend usable session time from minutes to hours on a battery.

---

Why Not Just Use Mobile NPUs Instead?​

Modern NPUs (like Qualcomm Hexagon or Apple Neural Engine) do accelerate AI workloads, but:

• They’re matrix-math optimised, not spiking-event optimised.
• They often require proprietary SDKs and retraining models in special formats.
• They still hit thermal limits in sustained generative AI workloads.

Neuromorphic processors like Akida:

• Can run quantised, sparse, and pruned models natively without expensive data shuffling.
• Could theoretically run a distilled / specialised LLM that fits in a fraction of the RAM and still delivers conversational quality.

---

Would It Make ChatGPTOSS Work on Phones?​

Realistically:

• You’re not going to run the full 20B gpt-oss on Akida right now — it’s still too large.
• But with model compression (distillation + sparsity), you could run a smaller, reasoning-optimised model that feels far more responsive than the brute-force CPU approach.
• Instead of the “2–3 tokens per second” in the article, you might see 10× or more throughput without cooking the phone.

---

Implications for BrainChip​

If OpenAI or others seriously want “ChatGPTOS” to run locally:

• They’ll either need dedicated AI accelerators in consumer devices or move toward neuromorphic architectures.
• Qualcomm, Apple, Google already have NPUs — but neuromorphic could leapfrog them in power efficiency per watt.
• If BrainChip could prove Akida can handle a slimmed-down LLM at useful speeds, it becomes a licensable IP block for smartphone SoCs.
• This would be a huge market — billions of devices, not just niche edge AI.

This next article (see below) titled “Free, offline ChatGPT on your phone? Technically possible, basically useless”, describes how running the new ChatGPT-oss on smartphones is technically feasible, but current smartphones aren’t powerful enough to make it practical at this point in time.

Right now, GPT-oss on mobile is bottlenecked because it runs purely on the CPU (and sometimes GPU) without tapping into specialized neural processing hardware.

So, I asked ChatGPT 5.0 if there'd be an improvement if the model were to be adapted to run on an NPU - particularly one optimized for event-based, sparse processing like Akida and here's the response I received.

I highlighted the last point for obvious reasons.

Free, offline ChatGPT on your phone? Technically possible, basically useless

You can now run ChatGPT offline with OpenAI’s gpt-oss model, even on a phone. But is it practical? Here's what works and what really doesn’t.

www.androidauthority.com

ChatGPT 5's Response:

Yes — Akida could absolutely be a technology to consider if the goal is to make something like “ChatGPTOS” actually usable on a mobile phone, and the limitations described in that article are almost a perfect setup for why a neuromorphic processor would matter.

Here’s why:

---

Why Akida Fits This Problem​

The main bottlenecks in running a 20B parameter LLM on a phone (as described in the article) are:

1. RAM requirements – Needs 12–24 GB just to load the model.
2. Compute requirements – Mobile CPUs & GPUs can’t sustain high token generation without overheating and throttling.
3. Power draw – 100% CPU load drains the battery extremely fast.
4. Heat – The phone becomes uncomfortably hot after just a few minutes.

Akida (or any high-efficiency neuromorphic processor) is built for:

• Event-driven, sparse compute — only processes relevant “spikes” of information instead of hammering every computation step like a CPU/GPU.
• Ultra-low power draw — operates in milliwatts for inference.
• On-chip learning & inference — avoids constant memory transfers that waste time and energy.
• Massive parallelism — can run edge AI workloads in real time without overheating.

That means for the same LLM inference task, a neuromorphic setup could:

• Reduce energy consumption by orders of magnitude.
• Maintain higher sustained token rates without thermal throttling.
• Extend usable session time from minutes to hours on a battery.

---

Why Not Just Use Mobile NPUs Instead?​

Modern NPUs (like Qualcomm Hexagon or Apple Neural Engine) do accelerate AI workloads, but:

• They’re matrix-math optimised, not spiking-event optimised.
• They often require proprietary SDKs and retraining models in special formats.
• They still hit thermal limits in sustained generative AI workloads.

Neuromorphic processors like Akida:

• Can run quantised, sparse, and pruned models natively without expensive data shuffling.
• Could theoretically run a distilled / specialised LLM that fits in a fraction of the RAM and still delivers conversational quality.

---

Would It Make ChatGPTOSS Work on Phones?​

Realistically:

• You’re not going to run the full 20B gpt-oss on Akida right now — it’s still too large.
• But with model compression (distillation + sparsity), you could run a smaller, reasoning-optimised model that feels far more responsive than the brute-force CPU approach.
• Instead of the “2–3 tokens per second” in the article, you might see 10× or more throughput without cooking the phone.

---

Implications for BrainChip​

If OpenAI or others seriously want “ChatGPTOS” to run locally:

• They’ll either need dedicated AI accelerators in consumer devices or move toward neuromorphic architectures.
• Qualcomm, Apple, Google already have NPUs — but neuromorphic could leapfrog them in power efficiency per watt.
• If BrainChip could prove Akida can handle a slimmed-down LLM at useful speeds, it becomes a licensable IP block for smartphone SoCs.
• This would be a huge market — billions of devices, not just niche edge AI.

I took the opportunity to send to IR at Brainchip.
Why not I say

 

[TECH]

While Larry agrees with all the frustration about the lack of ASX announcements and the effect on the SP as a result and also the shorters doing their faecal trades...Larry has never sold and keeps on buying as he did again today....so thank you to everyone who is keeping the SP at this level.....Larry loves you

View attachment 89417

Really Happy as Larry

Hey Larry........you are one of the smarter shareholders, I can't remember a post of yours that has ever been very negative (maybe I'm wrong)
but having our share price held in such a tight band, yes, it's very frustrating, but the opportunity is sitting right in everyone's face, for the
ones who currently hold cash, what are you waiting for.

The support from 0.18 to 0.20 is the strongest I have witnessed for a long time, to me it's like we have finally found a true bottom, yet look
how much development has been achieved to date with the Akida models, any sign of negativity? absolutely NOT.

Dr. JT has already given Sean a solid commitment to have the team produce AKD 3.0 successfully up and running within the first 8 weeks approximately of 2026, and correct me if I'm slightly wrong here, but from memory, the good Dr. stated that he felt this achievement would
propel Brainchip for the next decade!! that in itself is a huge call, clearly showing the confidence, the entire team has in what we have, which
obviously includes our current portfolio of Patents with more to follow.

Very exciting times indeed.......I seem to recall myself saying that over the first decade, must get it right eventually.

Tech (home once again in Perth) AKD

 

[Diogenese]

Hey Larry........you are one of the smarter shareholders, I can't remember a post of yours that has ever been very negative (maybe I'm wrong)
but having our share price held in such a tight band, yes, it's very frustrating, but the opportunity is sitting right in everyone's face, for the
ones who currently hold cash, what are you waiting for.

The support from 0.18 to 0.20 is the strongest I have witnessed for a long time, to me it's like we have finally found a true bottom, yet look
how much development has been achieved to date with the Akida models, any sign of negativity? absolutely NOT.

Dr. JT has already given Sean a solid commitment to have the team produce AKD 3.0 successfully up and running within the first 8 weeks approximately of 2026, and correct me if I'm slightly wrong here, but from memory, the good Dr. stated that he felt this achievement would
propel Brainchip for the next decade!! that in itself is a huge call, clearly showing the confidence, the entire team has in what we have, which
obviously includes our current portfolio of Patents with more to follow.

Very exciting times indeed.......I seem to recall myself saying that over the first decade, must get it right eventually.

Tech (home once again in Perth) AKD

Hi Tech,

Per the roadmap:

BrainChip Technology Roadmap - BrainChip ,

the Akida 3 tapeout is scheduled for Q1 2026, and the FPGA is due in Q2 2026:

Akida 3 capabilities:

Neuromorphic architecture is programmable, making the architecture more flexible than Akida 2.

The optional quantization enables 3rd party models to be used as is, without the need to convert to Akida format. Also INT16 /FP32 also opens up use to a broader range of high precision 3rd party models. In particular, that will open up the possibility to run all legacy open source CNN models on Akida 3 - that's decades of development.*

Akida 3 will be a universal NN, with spik-like function as an option.

Contextual awareness enables interplay between voice and video:

exciting developments this quater:

1. Akida 2 - TENNs - computer vision; [We already have TENNs model for speech denoising & speech recognition, and LLM + RAG - https://brainchip.com/ready-to-use-akida-neural-network-models/ ]

2. Akida GenAI - natural language processing.

In Q4, Akida 2 - advanced computer vision; signal processing.

Jonathan Tapson also foreshadowed patents for improved data memory movement.


* This could be useful where a customer has a library of old CNN models which they only need to access occasionally. They can run them in native CNN instead of having to convert to Akida SNN. Hence this extends Akida's backward compatibility.

 

[7für7]

Tomorrow — and I’m telling you, folks — it’s going to be a tremendous day, maybe the best day ever for a price-sensitive announcement. People are saying — many smart people — that this one could send the share price higher than it’s ever gone before. Skyrocketing. Like a rocket. And believe me, I know rockets. I had one. The best. Maybe I still do. Probably. Who knows? But I’ll tell you something: Musk never called me about it. Never! Why? Because he was afraid. Afraid of CHINA! And everybody knows it.

 

[IloveLamp]

#space #ml #ai #benchmarking #highperformance | Leonidas Kosmidis

Great to see the international impact of OBMark-ML at the IEEE Space Computing + Space Mission Challenges for Information Technology (SMC-IT/SCC) last week! The open source machine learning benchmarks which we developed for the European Space Agency - ESA were used in at least 5 publication...

www.linkedin.com

 

[Pom down under]

https://www.news.com.au/technology/innovation/elon-musk-says-openai-will-eat-microsoft-alive-as-gpt5-launches-to-consumers/news-story/f6425a40065d073d0c6ee39b3d251485

 

[TheDon]

Dear manipulators and downrampers,

Please keep the share price low until i get my tax return. Thanks


TheDon

 

[Space Cadet]

Second neurotic hackathon.

https://www.neurotum.com/post/the-second-neurotum-hackathon

A team that moved far was the SPADES team, they were tasked with using Brainchip neuromorphic hardware to estimate the future precise position of spacecrafts based on event-based data from the past. After a selective voting process, where both the mentors as well as student representatives voiced their opinions, the team of Roua, Jakub, Manoj and Jost won the challenge! Their team was most successful at hardware implementation onto the AKIDA chip and did wonderfully at predicting spacecraft pose based on the SPADES Dataset using Event Sensing.

SC

 

[IloveLamp]

Training for the operator or model i wonder

Proud to announce another successful deployment of the #DronutDD1 with a U.S. | Cleo Robotics

Proud to announce another successful deployment of the #DronutDD1 with a U.S. federal agency. In the most challenging GPS-denied environments, the Dronut® is redefining ISR — delivering real-time intelligence in places where no other drone can operate. From tight, complex indoor spaces to deep...

www.linkedin.com

 

[smoothsailing18]

Asking chatgpt, when a new product is launched in early 2026 what is the general time frames for advertising.

Share
For a new technology product released in early 2026, companies generally start promoting the product between 4 to 12 months prior to the launch date, with significant pre-launch activities intensifying around 2 to 4 months before the release. This timeframe allows for building anticipation, conducting market research, and preparing all necessary marketing and sales materials to ensure a successful debut [1] [2] [3] [4] [5].
According to www.iAsk.Ai - Ask AI

 

[7für7]

Jesus Christ … even ChatGPT know about the issue “lack of announcements” lol… I asked her how she would see the situation if she would be a shareholder of brainchip

“My opinion as a shareholder:

Technologically, BrainChip is stronger than ever – Akida 2 + Cloud are exactly what we needed. Customers can test instantly without ordering hardware, which massively shortens the time to real contracts.

Problem: The market barely knows this because there was no ASX announcement. Without major, recurring revenue deals, the share price will stay down here.

Opportunities: Energy efficiency, edge AI, data security – Akida is right on trend and not dependent on automotive alone.

Risk: If no big deals land in 2025, it’s going to be a grind.

I remain positive, buy on weakness, and wait for the moment when a real name + long-term contract drops. That’s when things will take off here.

 

[7für7]

Newsletter

Unleashing Akida in the Cloud and Beyond from BrainChip

A milestone but no ASX announcement… WHY???? What are you afraid of!?!??

 

[Space Cadet]

1 week ago

https://www.linkedin.com/posts/gais...uting-riscv-activity-7356631775965511682-_PHG

 

[7für7]

What I honestly don’t understand is this: What are they afraid of?

They call the Cloud system a milestone, yet they can’t manage to announce it properly through the ASX.
Where’s the problem? Is there something behind this that we don’t know about?
Is there a reason why they don’t officially share this milestone?

Or are they deliberately trying to prevent another hype?

To me, it makes no sense – and if anyone from BrainChip is reading this: Please give us a plausible answer without the usual platitudes.
Thank you.

MILESTONE!!! HELLOOOO!?!?

 

[TheUnfairAdvantage]

@7für7 Is there a reason?

Tony Lewis likes this LinkedIn post. Coincidence the Akida cloud is released or is it just good timing?

Be nice to think Brainchip can join this national network of AI cloud labs. Nice earner if they can.


https://www.linkedin.com/posts/nati...urce=social_share_send&utm_campaign=copy_link

 

[7für7]

@7für7 Is there a reason?

Tony Lewis likes this LinkedIn post. Coincidence the Akida cloud is released or is it just good timing?

View attachment 89494

Be nice to think Brainchip can join this national network of AI cloud labs. Nice earner if they can.


https://www.linkedin.com/posts/nati...urce=social_share_send&utm_campaign=copy_link

Interesting… BUT…
If BrainChip is involved in any way and there’s already a deal in place that is expected to generate significant revenue, there must definitely be a price-sensitive announcement.

Since this announcement has already been made public by them, there is no reason to assume that an NDA exists.

Edit…. COULD IT BE??

 

[MegaportX]

 

[Rach2512]

Tesla Simplifies AI Chip Strategy, Scraps Dojo | ASSOCIATION OF ELECTRONIC INDUSTRIES IN SINGAPORE (AEIS) posted on the topic | LinkedIn

🚗 Tesla Refines Its AI Chip Strategy Elon Musk announced that Tesla will streamline #AI #chip #design, focusing on versatile inference chips for both real-time decision-making and limited training — ending the costly split between multiple chip architectures. The Dojo supercomputer project has...

www.linkedin.com