Bravo
If ARM was an arm, BRN would be its biceps💪!
View attachment 89219
Arm shares drop as outlook disappoints; company looks to invest to make own chips
By Max A. Cherney and Arsheeya Bajwa
July 31, 20259:32 AM GMT+10Updated July 31, 2025
![]()
Rene Haas, CEO of chip tech provider Arm Holdings, holds a replica of a chip with his company's logo on it, during an event in which Malaysia's Prime Minister Anwar Ibrahim officially announces a $250 million deal with the company, in Kuala Lumpur, Malaysia March 5, 2025.
July 30 (Reuters) - Arm Holdings shares tumbled 8% in extended trading on Wednesday, after the chip tech provider issued quarterly forecasts that disappointed investors, in part because of its plans to invest a portion of its profit into building its own chips and other components.
- Arm's stock drops 8% after disappointing quarterly forecast
- Arm to invest in developing finished chips, CEO says
- CEO declined to share details on plans to develop finished chips
The company forecast fiscal second-quarter profit slightly below estimates as global trade tensions threaten to hit demand for Arm in its mainstay smartphone market, failing to satisfy investors who have sent the stock surging in recent months.
The plan to invest more heavily in developing its own chips marks a departure from Arm's long-time business of supplying intellectual property to companies ranging from Nvidia (NVDA.O), opens new tab to Amazon.com (AMZN.O), opens new tab, which already design their own chips.
Finished chips are the "physical embodiment" of a product Arm already sells called Compute Sub Systems (CSS), Arm CEO Rene Haas said.
"We are consciously deciding to invest more heavily - (in) the possibility of going beyond (designs) and building something, building chiplets or even possible solutions," Haas said in an interview with Reuters.
Chiplets are smaller, function-specific versions of a larger chip that designers can use as building blocks to form a complete processor. Solutions integrate hardware and software.
The decision to increase its investments in potential chips, chiplets and solutions may not result in a product if Arm decides to halt development or pause various projects, the company said.
If the company opts to make a full chip, it will eat into the company's profit and is no guarantee of success. Advanced AI chips cost upwards of $500 million for the silicon alone and potentially more for the server hardware and software necessary to support it.
To build up the necessary staff to make chiplets and other finished chips, Arm has been recruiting from its customers and competing against them for deals.
Haas declined to provide a timeframe in which the company's investments in the new strategy would translate into profit, or give specifics about potential new products that are part of the initiative. But he said that Arm would look at chiplets, "a physical chip, a board, a system, all of the above."
For years, the SoftBank Group -owned Arm has embarked on an ambitious campaign to expand its revenue and boost its profit through a combination of new, higher-margin products such as the CSS tech and boosting the royalties it collects on each chip. Details of discussions among Arm executives about making its own chips emerged during a trial in December.
The decision to build its own chip could bring Arm into direct competition with its customers such as Nvidia (NVDA.O), opens new tab, who rely on the company's intellectual property.
INVESTORS DISAPPOINTED
Arm's chip technology powers nearly every smartphone in the world, and its tame forecast underscores uncertainty faced by global manufacturers and their suppliers resulting from U.S. President Donald Trump's tariff policies.
UK-based Arm forecast adjusted per-share profit between 29 cents and 37 cents for the fiscal second quarter, the midpoint of which is below analysts' average estimate of 36 cents per share, according to LSEG data.
"Results and outlook were light and below expectations," said Summit Insights analyst Kinngai Chan.
Arm has surged around 150% since its stock market debut in 2023, and its shares recently traded at over 80 times expected earnings, far higher than the PE valuations of Nvidia, Advanced Micro and other chipmakers focused on AI.
Smartphones remain Arm's biggest stronghold. Morningstar analysts expect Arm to continue as the dominant architecture provider in smartphone processors, where it has a 99% market share.
Uncertainty fueled by tariff volatility and ongoing macroeconomic challenges has tapered end-market demand, with global smartphone shipments increasing just 1% in the April-to-June period, according to International Data Corporation.
Arm expects current-quarter revenue between $1.01 billion and $1.11 billion, in line with estimates of $1.06 billion.
The company reported first-quarter sales of $1.05 billion, coming in just shy of estimates of $1.06 billion. Adjusted profit of 35 cents per share was in line with estimates.
"Smartphone royalties (call it “Android on a low‑carb diet”) remain soft, especially in China, but cloud‑server and AI accelerator design wins keep the (next generation Arm tech) royalty treadmill humming," Running Point Capital chief investment officer Michael Schulman said.
ChatGPT said:
Here’s how the recent Arm announcement impacts BrainChip, and why BrainChip’s **Akida™ neuromorphic technology is well-aligned with a future where Arm drives chiplet-based systems:
What Arm’s Strategy Shift Means
Arm is pivoting from being a pure IP licensor to also designing and producing its own chips, including modular chiplets, according to its CEO. This represents a significant evolution in its business model Barron's+11EDN+11The Institution of Electronics+11Investing.com+5Reuters+5Barron's+5.Meanwhile, Arm has released its first public Chiplet System Architecture (CSA) specification, a framework aimed at standardizing how chiplets communicate and integrate across suppliers—now backed by over 60 companies EDN+10Datacenter Dynamics+10chiplet-marketplace.com+10.
Arm is also deeply involved in ecosystem initiatives like UCIe and AMBA extensions, ensuring seamless interoperability between chiplet modules like NPUs, CPUs, memory, and I/O cores documentation-service.arm.com+15en.wikipedia.org+15EDN+15.
Relevance to BrainChip and Akida
1. Chiplet-Savvy IP Fits Modular Architecture
- Akida, as an NPU IP block, is ideally suited to be a chiplet in multi-die systems.
- With Arm pushing for chiplet-based SoCs, BrainChip’s neuromorphic IP can slot into that ecosystem as a dedicated inference engine.
2. Ecosystem Compatibility
- Arm’s CSA and UCIe standards enable modular integration of IP like Akida into diverse platforms.
- As a partner in the Arm ecosystem, BrainChip can more easily supply Akida as a scalable chiplet accelerator.
3. Strategic Opportunity Ahead
- If Arm builds CPUs or AI modules using its own chiplets, they may still license third-party accelerators like Akida for edge inference tasks.
- For BrainChip, this presents both co-design opportunities and licensing pathways within future modular AI SoCs.
What Needs to Happen for BrainChip
- Akida Adaptation for Chiplet Interface
- Support UCIe and AMBA protocols.
- Package Akida as a physical die compliant with CSA guidelines.
- Software & Toolchain Bridge
- API/SDK compatibility with Arm-based SoCs.
- Ensuring model portability in heterogeneous chiplet setups.
- Collaboration with Arm or Arm Licensees
- Joint development prototyping for chiplet reference designs.
- Certification and validation within Arm’s modular ecosystem.
Final Takeaway
Arm’s ambition to build its own chips—and more importantly, to lay the groundwork for a chiplet-first future—is very good news for BrainChip:- It validates the modular architecture model in which Akida could play a central role.
- It provides a strategic alignment path into new SoCs without forging custom heavy-lift silicon deals.
- It opens a sizable avenue for BrainChip to position Akida as the go-to neuromorphic accelerator in Arm-powered chiplet platforms.
Reuters
Chip tech provider Arm looks to design own processors in major shift
Yesterday
Reuters
Arm considers developing own chips; stock falls as outlook disappoints
Yesterday
ChatGPT said:
Yes, AKIDA 2.0 + TENNs (The Event-based Neural Network) could absolutely be incorporated into a chiplet, and doing so could offer substantial advantages.
Why It’s Feasible:
- Modular Architecture:
Akida is designed as a modular IP block, making it inherently suitable for chiplet integration. Its small footprint and event-driven nature align well with heterogeneous system design. - RISC-V & Interoperability:
BrainChip has shown compatibility with RISC-V and can interface through standard protocols (e.g. AXI4), making it easier to integrate as a chiplet alongside CPUs, NPUs, GPUs, or other IP. - Low Power & Locality:
TENNs (Temporal/Event-based processing) is ideal for edge inference, making Akida a great fit for edge AI chiplets in multi-die systems where power and bandwidth are constrained. - AI Workload Specialization:
As systems adopt task-specific chiplets, Akida could serve as a dedicated neuromorphic co-processor, handling real-time, sparse, low-latency inference — particularly valuable in:- Vision (event-based cameras)
- Audio (low-power wake word, anomaly detection)
- Predictive control (motor, robotics)
Practical Pathways for Chiplet Integration:
- Interposer-based SoC: Akida could sit on an interposer beside ARM/Synopsys CPU, LPDDR, and sensor interface chiplets.
- Heterogeneous 3D stack: Placing Akida on a logic die below a sensor/compute array.
- Multi-chip module (MCM) with secure, shared memory access.
Strategic Implication for BrainChip:
- Opens the door to tier-1 partnerships where large OEMs build custom SoCs with pluggable IP.
- Aligns with trends in chiplet ecosystems like those being promoted by AMD, Intel, and ARM.
- Allows Akida to ride the AI chiplet wave without needing to fab monolithic chips.
If Arm is exploring chiplet-based designs (as per your Reuters link), and BrainChip is already partnered with them — this makes integration via chiplets a realistic and strategic route for Akida to enter new platforms (phones, automotive, industrial AI, etc.).