Arm: Computex Update, CPU Core Demand Hinted at Being Higher   

Posted on by io-fund

We believe it is worthwhile to revisit Arm for a couple of reasons: the first being the fact that shares have meaningfully broken out post-earnings, at one point up more than 100% over the last month, and the second being to make sure the we have not overlooked any pieces of Arm’s story given the strengthening thematic tailwinds from agentic AI driving the CPU to GPU closer to parity due to higher orchestration needs.  

Computex Takeaways – AGI CPU in Production, New Customers 

There were a handful of notable updates from Arm regarding the AGI CPU at Computex, while discussion around the CPU industry provided further confirmation on the thesis that agentic AI is quickly driving the CPU-GPU ratio towards 1:1 (or better).  

At Computex, CEO Rene Haas confirmed that the AGI CPU is in production at TSMC, hinting that it could begin recognizing AGI CPU system revenue sooner and potentially accelerate its ramp (pending supply). Haas also revealed two other large-scale customers joining the fray for the AGI CPU – Oracle and ByteDance, complementing launch partners Meta, OpenAI, Cloudflare, Cerebras and others. Still, the challenge likely remains securing supply to push initial revenue forecasts higher, as Arm did not offer much on that front at Computex. Haas later stated on Bloomberg that he was “very confident” that Arm would reach its $15 billion target by FY31 with demand remaining strong, adding that he hopes Arm could reach that target sooner.   

Also at Computex, Nvidia revealed its RTX Spark, its Arm-based PC superchip, featuring a slimmed-down 20-core Grace CPU alongside a Blackwell RTX GPU offering  up to 1 petaFLOP of FP4 performance on Windows laptops. Microsoft says RTX Spark offers “industry-leading performance per watt for creative, AI and gaming workloads” on Windows, with the chip helping consumers build and run AI models, inference and agents locally on PCs with native CUDA support. First PCs built with Spark are expected this fall, which could alleviate concerns to Arm’s growth related to PC softness stemming from elevated memory costs cutting into demand. 

Key rival Intel added more color to the CPU-GPU ratio thesis, explaining that due to immense orchestration needs, agentic AI is “leading to a situation where the 1 CPU to 8 GPU ratio in frontier model training has shifted CPU density to 1:1 or better.” This is along the lines of what AMD had implied as well, with the ratio potentially moving beyond 1:1 in favor of CPUs. This is up from 1:4 to 1:8 today, a substantial shift that has to happen quickly considering agentic applications are rapidly proliferating. 

Source: Arm Arm 

The main takeaway here is that CPU demand is expected to explode – Arm outlined at Computex that the agentic ecosystem (based on Github stars) has risen roughly 5X in the past three months, wildly outpacing the growth of both Linux and Kubernetes, with CPU demand following shifting from following Linux to following closely behind this agentic curve. While CEO Rene Haas had outlined a 4X growth in CPU cores per GW in March, from 30M to 120M, he stated at Computex that “4X, 8X, 10X, it’s a hard number to predict just based upon the growth rates of these agents,” which at its core implies that there could be certain agentic applications or deployments that require much greater CPU density and thus a much higher CPU:GPU ratio.  

To put this more in dollar terms, what this suggests is that TAM estimates for server CPUs, including Arm’s $100B forecast and AMD’s recently doubled $120B forecast, could still have significant room to the upside if the CPU:GPU ratios moves quickly towards 1:1 or better, or if CPU core growth per GW starts advancing towards that >8X number Arm laid out.  

Touching on v9, CSS and Hyperscaler Deployments 

Given that we are still awaiting the broader ramp and strongest contributions from Arm’s AGI CPU, growth in the meantime will remain tied to royalty and licensing revenue. For royalties, the question here is whether v9 and Arm’s compute subsystems (CSS) designs can help accelerate growth in the near term. 

For reference, as we had pointed out in our post-Q4 FY26 earnings analysis, Arm FQ4: AGI CPU Demand Hits $2B, Revenue Outlook Stays at $1B, Q1 guidance was relatively in line while 1H is expected to be softer with revenue growth dipping below 20% YoY. This dynamic has not changed much since, with FQ1 revenue projected to be ~20% before decelerating to 18% in FQ2 and rebounding in FQ3. 

There are a handful of factors that could support stronger royalty revenue growth through the rest of 2026 into 2027 as the AGI CPU prepares to ramp, stemming from hyperscaler deployments of Arm-based chips. 

To start, Arm’s latest v9 and CSS architectures carry much higher royalty rates per core, with the subsequent v9 generation carrying a 1.5X higher price versus the first v9 gen, with a similar dynamic occurring with CSS; to note, Arm sees its subsequent gen CSS carrying a 3X higher rate than its first gen v9, emphasizing why CSS wins are increasingly crucial for royalty growth (with two deals signed last quarter, one of which is for data center networking chips and the other for smartphones).  

Source: Arm Arm 

Also layering in to growth next year is a broad line-up of new data center chips based on Arm’s architecture – the company sees at least 8 new chips coming online in 2027, more than double the three new Arm-based data center chips that came online in 2026. Four of the eight feature substantially higher cores than 2026’s launches, providing a direct outlet for royalty growth, with three of these being among the top five highest-core count chips launched since 2018.  

Source: Arm Arm 

For a rough, speculative estimate on what 2-4X growth in CPU core demand could suggest for server CPU royalty revenue growth through 2028:  

Assuming server CPUs account for roughly two-thirds of Cloud AI royalty revenue (as this also includes DPUs, networking, etc, but with significant concentration in hyperscalers’ custom CPUs), this would project server CPU royalty share of around 8-9% in FY26, or ~$220 million at midpoint, based on Cloud AI taking roughly 12-13% share.  

Estimating 2X growth in CPU core demand from here by 2028 combined with ~2X higher royalty rates from blending v9, CSS v3 and upcoming CSS v4 (slated for 2027 though exact release data), this would project server CPU royalties to $880 million. A 4X increase in core demand combined with the same ~2X increase in royalty rates would roughly estimate server CPU revenue of up to $1.76 billion. This would roughly estimate server CPU share of overall revenue for Arm to rise from the 4-5% range to the 18-19% range under the 4X core growth assumption by the end of 2028. 

Hyperscaler Chip-Based Demand Signals 

Notably, Google’s newest TPUs, 8t and 8i, will both see the Arm-based Axion CPU replace x86 chips at the head node, which Google says will “remove the host bottleneck caused by data preparation latency.” TPU shipments are expected to see a rapid ramp in 2027, with UBS modeling shipments rising nearly 139% YoY from 4.13 million in 2026 to 9.87 million in 2027. Reports have suggested that the new generation will adopt a 1:2 CPU-to-TPU ratio, which would imply demand of close to 4 million Axion CPUs in 2027 if the v8 TPU accounts for roughly 80% of UBS’ estimated shipments.  

Amazon highlighted in early April that its Arm-based Graviton CPU was seeing exceptionally strong demand, noting that “two large AWS customers have already asked if they could buy all of our Graviton instance capacity in 2026.” Amazon also signed a deal with Meta, letting Meta access tens of millions of Graviton cores for at least three years; in terms of chips (based on 192 cores for Graviton5 and assuming 30-50M cores), this would represent 156K-260K individual chips. Amazon also noted that Graviton accounted for more than half of the CPU capacity it added last year for the third year in a row. 

Nvidia’s Vera CPU cannot be forgotten either, as Nvidia recently outlined a $200 billion TAM for the new CPU with visibility into $20 billion in total CPU revenue this year, as it plans to utilize it in four different ways (Vera Rubin, standalone CPUs, Vera CPU plus CX-9 and storage, and Vera CPU plus CX-9 and security/confidential computing). The standalone Vera racks also offer 7X more CPUs in one rack, at 256 compared to the 36 CPUs in the Vera Rubin NVL72 (and thus 22,528 cores vs 3,168 for the NVL72), offering room for royalty growth for Arm if the rack does indeed scale towards $20 billion in revenue.  

Analysts Increasingly Bullish on Arm 

Analysts look to be getting increasingly bullish on Arm despite the rather lukewarm Q4 report a month ago.  

The most notable (and active) is Mizuho, which has increased its price target on Arm three times since May 28, taking it from $290 to $360 on growing agentic AI demand, then again to $425 on June 1 due to its view on supply constraints driving server CPU upside. Mizuho raised this to $500 on June 4 on increased confidence in Arm’s ability to hit its $15 billion AGI CPU goal.  

Wells Fargo recently hiked its price target on Arm from $255 to $410, with its main takeaway from its ‘Bus Tour’ being that the “proliferation of AI inferencing / Agentic AI [is] driving significant incremental server CPU demand.” Bernstein also believes Arm is the “structural beneficiary of the renaissance” of CPUs with agentic AI, stemming from its “unparalleled power efficiency.”  

Conclusion 

Despite a rather lackluster earnings report and soft guide, Arm’s shares meaningfully broke out with a nearly 100% rally in the back half of May on increasing momentum and optimism on agentic AI’s tailwinds for CPU growth. Arm hinted at Computex that CPU core demand per GW could move meaningfully higher than the 4X it described at the launch of its AGI CPU, depending on how agentic AI deployments unfold, while key rivals have also laid the foundation for CPU:GPU ratios to quickly move towards 1:1 or better.

Please note: The I/O Fund conducts research and draws conclusions for the company’s portfolio. We then share that information with our readers and offer real-time trade notifications. This is not a guarantee of a stock’s performance and it is not financial advice. Please consult your personal financial advisor before buying any stock in the companies mentioned in this analysis. Beth Kindig and the I/O Fund own shares in ARM at the time of writing and may own stocks pictured in the charts.

Damien Robbins, Equity Analyst at I/O Fund contributed to this analysis.

Recommended Reading:

Leave a Reply

Your email address will not be published. Required fields are marked *