Bernstein initiates coverage on SoftBank and Arm, arguing that Agentic AI will re-rate the value of the AI full stack and CPUs
AI summary card
Bernstein initiates coverage on SoftBank and Arm, arguing that Agentic AI will re-rate the value of the AI full stack and CPUs
The report rates SoftBank and Arm Outperform, with the core thesis that Agentic AI, by moving from suggestion to execution, will significantly increase demand for tokens, CPU cores, AI infrastructure, and physical AI.
- Gen AI 2.0, namely Agentic AI, can autonomously plan, call tools, execute tasks, and iterate workflows, and the report argues that its commercial value and enterprise adoption potential are far higher than the chatbot stage.
- SoftBank is positioned as a full-stack AI investor covering infrastructure, silicon, models/agents, and physical AI, with core assets including Arm, OpenAI, SB Energy, DigitalBridge, SoftBank Robotics, and ABB Robotics.
- Arm is viewed as a structural beneficiary of the CPU revival driven by Agentic AI; the report states that each GW of Agentic AI data center capacity requires about 120 million CPU cores, roughly 4x that of traditional AI data centers.
- The report forecasts Arm revenue to grow to $26bn by 2030, EPS to rise to about $9.83, and assigns an Outperform rating with a $300 target price.
- The report expects SoftBank's NAV to grow to $390bn in one year and derives a ¥8,200 target price using a 25% NAV discount.
Report interpretation
Overview
This is Bernstein's initial coverage report on SoftBank and Arm. The report's main thesis is that Gen AI is moving from the 1.0 chatbot stage into the 2.0 Agentic AI stage, where AI is no longer just generating suggestions but can plan, act, call tools, and complete end-to-end workflows. The authors believe this shift will significantly expand monetizable AI use cases and drive token consumption, AI infrastructure buildout, CPU demand, and the development of physical AI. Based on this framework, SoftBank is seen as an investment vehicle covering the full AI stack, while Arm is viewed as the core beneficiary of Agentic AI-driven CPU re-rating.
Core views
The report's key views are: first, Agentic AI resolves the commercial bottleneck in execution that Gen AI 1.0 faced, shifting AI value from text output to measurable business outcomes. Second, the more complex orchestration of Agentic AI tasks will materially increase token consumption and CPU orchestration demand. Third, SoftBank covers the AI value chain through assets such as Arm, OpenAI, SB Energy, DigitalBridge, SoftBank Robotics, and ABB Robotics. Fourth, Arm's energy-efficient architecture and strategy to extend from IP licensing into higher-value portions of the chain give it structural opportunities in AI data centers, edge AI, and physical AI.
Analysis framework
The report combines a thematic framework, value-chain mapping, and valuation models: it first uses the evolution from Gen AI 1.0 to 4.0 to explain demand changes, then maps SoftBank's portfolio to four layers—AI silicon, infrastructure, models/agents, and physical AI—and finally values SoftBank using a NAV discount approach and Arm using high-growth earnings forecasts and a P/E multiple.
Methodology notes
The technology evolution framework from chatbot to Agentic AI, Physical AI, and AGI
The report defines Gen AI 1.0 as chatbot, Gen AI 2.0 as Agentic AI, Gen AI 3.0 as Physical AI, and Gen AI 4.0 as AGI, and uses this framework to explain the long-term benefit path for SoftBank and Arm.
SoftBank NAV discount valuation
The report expects SoftBank's NAV to reach $390bn in one year and applies a 25% NAV discount in the base case to derive a ¥8,200 target price; the bull case uses a higher Arm valuation and a 15% discount, while the bear case uses a 45% discount.
Arm earnings multiple valuation
The report forecasts Arm revenue to grow more than fivefold to $26bn by 2030, EPS to rise to $9.83, and derives a $300 target price using a 90x P/E on Q5-8 EPS of $3.33.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- SoftBank Group CorpAI full-stack investment vehicle
- Strengths
- Through assets such as Arm, OpenAI, SB Energy, DigitalBridge, SoftBank Robotics, and ABB Robotics, it covers silicon, models/agents, infrastructure, and physical AI; NAV has already been supported by gains from OpenAI and Arm investments.
- Weaknesses
- Valuation depends on the NAV discount, the valuation of core unlisted assets, and Arm's share price performance; the portfolio is complex and some projects still require substantial capital investment.
- Comparison
- Compared with a single AI hardware or software company, SoftBank looks more like a holding platform spanning multiple layers of the AI value chain.
- Risks
- A decline in OpenAI valuation, lower-than-expected Arm valuation, capital expenditure pressure from infrastructure projects such as Stargate, a wider NAV discount, and weaker-than-expected execution on ABB Robotics acquisition and physical AI deployment.
- ArmCore beneficiary of the CPU revival driven by Agentic AI
- Strengths
- The Neoverse and Arm architectures offer energy-efficiency advantages, making them suitable for power-constrained data centers; the report says Agentic AI will significantly increase CPU orchestration demand and drive Arm to extend from IP licensing into higher-value parts of the chain.
- Weaknesses
- The high valuation requires rapid revenue and EPS delivery; gains in server CPU share still face competition from x86, custom chips, and customers' in-house chip development paths.
- Comparison
- Compared with GPU-dominated Gen AI 1.0, Agentic AI relies more heavily on CPUs for planning, tool calling, file reading/writing, retries, and multi-agent coordination, thereby increasing Arm's relative value.
- Risks
- Adoption of Agentic AI is slower than expected, the CPU:GPU ratio does not change as expected, data center customers migrate slowly, and there are execution risks if Arm self-develops or participates more deeply in the chip value chain.
- OpenAISoftBank's core exposure to the AI model and agent layer
- Strengths
- The report says OpenAI's ARR will reach $25bn by March 2026, Codex showed an agentic breakthrough similar to a ChatGPT moment in April 2026, and a potential IPO could provide additional valuation upside.
- Weaknesses
- There may still be high cash burn, competition, and valuation volatility.
- Comparison
- Within SoftBank's portfolio, OpenAI is the key asset at the model and Agentic AI platform layer.
- Risks
- IPO valuation falls short of expectations, commercialization slows, regulatory or competitive pressure increases, and cash burn exceeds market tolerance.
- SB Energy, Stargate and DigitalBridgeExposure to the AI infrastructure and energy layer
- Strengths
- As Agentic AI drives token and compute demand higher, the report believes Stargate and other data center and power-related assets will benefit from AI infrastructure expansion.
- Weaknesses
- Infrastructure projects are capital intensive and long cycle, with returns dependent on utilization and financing conditions.
- Comparison
- Compared with pure software AI assets, this layer benefits more directly from demand for compute, power, and data center construction.
- Risks
- Capital expenditure exceeds budget, financing costs rise, construction progress is delayed, and AI compute demand is below expectations.
- SoftBank Robotics and ABB RoboticsExposure to physical AI and embodied intelligence
- Strengths
- The report argues that Physical AI is Gen AI 3.0 and serves as an important bridge for Agentic AI to enter the physical world; if ABB Robotics acquisition is completed, it will enhance SoftBank's capabilities in industrial robotics.
- Weaknesses
- Physical AI commercialization is still at an early stage, and there are uncertainties in technology, supply chain, and real-world use cases.
- Comparison
- This asset layer extends AI from digital workflows to real-world operations, forming a combination of models, chips, and execution endpoints with OpenAI and Arm.
- Risks
- Acquisition integration risk, volatile robotics demand, technology progress slower than expected, and uncertainty around the AGI/Physical AI path.
Key data
- Agentic AI economic impactat least $456bnCapgemini estimates that the economic impact of Agentic AI during 2025-2028 will be at least $456bn.
- Agentic AI token consumptionabout 1,000x AI 1.0The report believes more complex autonomous workflows will drive token demand and AI monetization.
- OpenAI ARR$25bnThe report says OpenAI's ARR as of March 2026 increased more than 10x from two years earlier.
- SoftBank OpenAI stake11%The report says SoftBank has invested about $45bn and plans to invest another $20bn by October 2026, lifting its stake to 13%.
- SoftBank NAV forecast$390bnThe report expects next year's NAV to be mainly driven by Arm and OpenAI.
- SoftBank target price¥8,200.00Based on a 25% NAV discount, implying about 47% upside.
- Arm target price$300.00Based on Arm's revenue and EPS growth, plus a high multiple valuation.
- Agentic AI data center CPU core demandabout 120 million CPU cores/GWThe report says traditional AI data centers require about 30 million CPU cores/GW, while Agentic AI requires about 4x as much.
- Server CPU TAM$137bn by 2030The report argues that a shift in the GPU:CPU ratio from 8:1 to 2:1 or 1:1 will expand the server CPU TAM.
- Arm 2030 revenue forecast$26bnThe report forecasts Arm revenue to grow more than fivefold by 2030.
- Arm 2030 EPS forecast$9.83The report forecasts EPS to expand by about 5.5x as well.
Impact & implications
If the report's view is correct, AI investment opportunities will no longer be concentrated only in GPUs or the model layer, but will spread to CPUs, data center power and real estate, AI agent software, robotics, and automation in the physical world. SoftBank's portfolio could therefore be re-rated by the market as a full-stack AI platform, and Arm's valuation logic could shift from a mobile IP licensing company to a high-efficiency CPU and higher-value-chain participant in the AI era.
Risks
- Enterprise adoption of Agentic AI and ROI realization are below the report's expectations.
- OpenAI valuation, revenue growth, or IPO timing falls short of expectations.
- Arm's revenue realization in data center CPUs, edge AI, and higher-value chain segments is slower than expected.
- The GPU:CPU ratio changes less than expected, and server CPU TAM expansion is smaller than assumed in the report.
- SoftBank's NAV discount widens or core portfolio valuations are revised down.
- Capital expenditure and financing risks rise for AI infrastructure projects such as Stargate, DigitalBridge, and SB Energy.
- Execution risks exist for the ABB Robotics acquisition, SoftBank Robotics, and physical AI commercialization.
- AI regulation, competition, model costs, and changes in technology paths could weaken the investment thesis.
What to watch
- The actual commercialization progress of OpenAI ARR, Codex, and other Agentic AI products.
- OpenAI's IPO timetable and whether its valuation exceeds the latest funding round's $852bn.
- Arm's data center CPU share, Neoverse adoption rate, and hyperscaler customer penetration.
- Whether Agentic AI data center CPU:GPU ratios move toward 1:1 or 1:2.
- Whether Arm can move beyond IP licensing to capture CPU or more complete chip value-chain revenue.
- Changes in SoftBank's NAV discount and Arm/OpenAI's contribution to NAV.
- Financing, construction progress, and order visibility for Stargate, SB Energy, and DigitalBridge projects.
- Completion of the ABB Robotics acquisition and SoftBank's integration capability in physical AI and robotics.