Nvidia Wants a Helping ARM
With its $40 billion bid, the chipmaker looks to corner the AI market.
After reflecting on Nvidia’s (NVDA) announcement that it plans to acquire ARM from SoftBank in a transaction valued at $40 billion, we foresee nice potential revenue synergies in the data center and the leveraging of Nvidia’s graphics processing unit and artificial intelligence intellectual property into ARM’s treasure trove of IP.
We have raised our fair value estimate for narrow-moat Nvidia to $300 per share on a probability-weighted basis. Should the deal go through, our fair value estimate for the combined company would be $350 per share. However, based on the regulatory risk associated with this deal--the largest in chip history if it closes--we assign a 50% probability of the acquisition closing. If the deal does not close, our stand-alone fair value estimate for Nvidia would probably remain at $250, all else equal. We believe that from a valuation perspective, Nvidia is paying a high multiple for ARM’s earnings. But given that the GPU leader’s share price is trading at a significant premium to our stand-alone $250 fair value estimate, we like that Nvidia is using its rich shares to fund a large portion of the deal.
The company’s focus will be to offer a comprehensive data center portfolio that includes ARM-based central processing units while leveraging Nvidia’s AI expertise into ARM’s vast ecosystem spanning the data center to mobile and “Internet of Things” devices. Similar to SoftBank’s justification when it bought ARM, Nvidia expects to bolster ARM’s research and development budget to realize its data center vision. ARM was a wide-moat-rated firm when we covered the stand-alone entity, and we believe it is likely this cash/stock deal will enhance Nvidia’s narrow moat even further, should it close.
Nvidia is financing the deal via $21.5 billion of its common stock (44.3 million shares) and $12 billion in cash from its balance sheet, which includes $2 billion payable at signing. SoftBank may receive as much as $5 billion in cash or common stock subject to ARM meeting certain performance targets, while Nvidia will issue $1.5 billion in equity to ARM employees for retention purposes.
ARM’s pro forma revenue and adjusted EBITDA margin are $1.8 billion and 35%, respectively, which implies a lofty price/EBITDA ratio of 63 times. SoftBank purchased ARM for $32 billion in 2016, which implies a modest return (6% compound annual growth rate) for the technology conglomerate. For background, SoftBank paid a 43% premium to ARM’s share price at the time of announcement. The deal does not include ARM’s Internet of Things services group, which would have been margin-dilutive relative to the pro forma adjusted EBITDA margin of 35%.
While there is nothing stopping Nvidia from developing its own ARM-based server CPU without a merger (its Project Denver in 2014 sought to accomplish just that), we assume an outright purchase may help the company accelerate any server CPU ambitions it may have. There has been mixed success in the ARM-based server CPU camp, with Qualcomm and Broadcom retreating from the endeavor while Marvell/Cavium and Amazon have had some traction. Beyond Nvidia focusing ARM’s R&D efforts on server CPUs, we also imagine opportunities to incorporate Nvidia’s AI and GPU expertise into ARM’s chip designs, particularly in the smartphone realm. This could be an interesting dynamic as Apple, Qualcomm, and others have been investing in similar initiatives. There is a risk that ARM customers such as Apple could try to use an alternative such as open-source RISC-V, though this would be an arduous process that could take years.
Automotive (self-driving cars) is another area that Nvidia is keen on dominating, and there are a host of chip firms using ARM-based silicon for their automotive processors. ARM’s open-licensing model and customer neutrality are likely to be areas of concern in these key end markets if Nvidia can close the deal, in our view.
Currently, Nvidia is a licensee of ARM; its Tegra-based processors found in automotive infotainment systems and the Nintendo Switch are ARM-based. CEO Jen-Hsun Huang noted the ARM acquisition would double Nvidia’s target addressable market to $250 billion. He also highlighted the vastness of ARM’s ecosystem, with 22.8 billion ARM-based chips shipped in 2019 relative to about 100 million Nvidia chips sold in a year. Nvidia’s goal is to leverage ARM to reach customers that Nvidia’s core GPU business can’t reach, including offering Nvidia’s GPU and AI IP along with ARM IP. Many ARM licensees are also focused on GPU and AI applications, though, and we are unsure how Nvidia would be able to assuage any reservations those customers would have about how Nvidia and ARM could license said IP without unfairly stifling competition.
Regulatory risk (particularly from escalating U.S.-China tensions), potential pushback from major ARM licensees (including Apple, Qualcomm, and Huawei, among many others), and competitors such as Intel are formidable hurdles to the consummation of a deal, in our view. The deal isn’t expected to close for 18 months, and although Nvidia succeeded in attaining regulatory approval from China for its Mellanox acquisition earlier in 2020, we expect the magnitude and significance of this potential deal will attract heightened scrutiny.
We remind investors that multiple semiconductor-related deals we had viewed favorably have fallen through in recent years, including Applied Materials/Tokyo Electron, Lam Research/KLA-Tencor, and Qualcomm/NXP. Also, adding a licensing component to Nvidia’s dominant GPU and AI chip business could lead to some regulatory and industry concerns comparable with the relationship between Qualcomm’s chip and licensing segments, which has faced intense customer and regulatory backlash in recent years. ARM customers and competitors such as Apple, Qualcomm, and Intel are likely to be less enthusiastic about the Nvidia/ARM combo.
Abhinav Davuluri does not own (actual or beneficial) shares in any of the securities mentioned above. Find out about Morningstar’s editorial policies.