Key Takeaways
- Demand for AI chips is set to shape the future of the global semiconductor industry.
- Yet despite the opportunity, getting M&A deals over the line has proven challenging.
- In 2023, “hyperscaler” data center operators scaled back acquisitions in the space, but manufacturers emerged as major buyers.
Last year, there were fewer mergers and acquisitions in the AI hardware sector than in 2021/2022. Having dominated dealmaking in previous years, “hyperscaler” data center operators pulled back significantly, with neither Amazon, Google nor Microsoft making a single acquisition in the space.
In their place, manufacturers stepped up as the preeminent buyers of AI hardware startups, but they have also struggled to navigate the tumultuous geopolitical landscape of the global semiconductor market.
Governments Step in to Prevent Takeovers
Even before the AI chip business became the hottest opportunity in town, pulling off semiconductor M&A was notoriously difficult.
Consider the case of Intel’s plan to purchase the Israeli chip foundry Tower Semiconductor for around $5.4 billion. The deal appeared to be progressing along nicely until it was called off last August “due to the inability to obtain in a timely manner the regulatory approvals required under the merger agreement,” the American tech giant stated.
According to the Wall Street Journal, the deal was scuppered by the Chinese government, which has a vested interest in protecting its national semiconductor industry from disruptions to the status quo.
Running in the opposite direction, governments in the UK and Germany have blocked Chinese takeovers of European chip makers. In each instance, European leaders argued that it was in the interest of national security to ensure domestic manufacturers don’t fall under improper foreign control.
Of course, it’s not that there aren’t British companies making silicon chips (although there are fewer German ones). But more often than not, firms that design, assemble and distribute AI hardware outsource the manufacturing of silicon components to dedicated “foundries.”
Limitations of the Foundry Model
To understand how the foundry model affects AI hardware manufacturing, consider the world’s preeminent GPU maker – Nvidia.
The US firm is what is known as a “fabless” manufacturer. In other words, while it designs its integrated circuits in California, fabrication is outsourced to the Taiwan Semiconductor Manufacturing Company (TSMC).
Alongside Intel and Samsung, TSMC is one of just 3 foundries in the world capable of making the 5nm nodes needed for contemporary AI chips. Considering Intel doesn’t offer its manufacturing capabilities as a service to other companies, Nvidia’s options are extremely limited.
This bottleneck in the global silicon market is reflected in recent M&A activity, which has seen fabless manufacturers emerge as major buyers of AI startups.
AI Manufacturers Ramp up M&A Activity
Last year, Nvidia acquired OmniML, a San Jose-based firm focused on miniaturizing large AI models so that they can run on home computers or smartphones. The deal won’t reduce Nvidia’s reliance on Taiwanese factories, but it will help the company secure a foothold in the emerging market for consumer AI hardware.
As noted by Omdia’s Principal Analyst for Advanced Computing, Alexander Harrowell, “on a company basis, IBM is the biggest buyer and is maintaining a steady cadence of around one [AI] acquisition every quarter.”
Like Nvidia, IBM’s acquisitions haven’t targeted foundries, but have rather focused on software startups that could strengthen the firm’s position in the AI hardware sector.
Recent additions to IBM’s AI portfolio include Manta Software, StreamSets and webMethods Platforms. In each case, the firm said it would integrate technologies acquired through the deals into its flagship cloud AI platform WatsonX.
Reflecting the complex interrelations that define the contemporary AI business, WatsonX runs on Nvidia GPUs housed in data centers owned by Amazon Web Services (AWS).
Ultimately, however, Nvidia and IBM’s moves in the AI sector do little to disrupt the established manufacturing hierarchy. But others have ambitious plans to shake up the status quo.
Masayoshi Son and Sam Altman Propose Rival AI Chip Ventures
In recent months, the billionaire tech investors Masayoshi Son and Sam Altman have both pitched AI chip manufacturing ventures that could take on Nvidia and maybe even TSMC.
During SoftBank’s annual shareholder meeting last week, Masayoshi revealed that the Japanese bank was in discussions with Middle Eastern investors to build a $100 billion semiconductor business.
Meanwhile, Altman has also been busy raising funds for a secretive TPU project dubbed Tigris.
Ultimately, while dealmakers in the AI hardware sector face many obstacles, there are plenty of startups and potential buyers ready to snap them up. Alongside institutional investors like SoftBank, Big Tech players like Microsoft that have already spent billions are unlikely to remain inactive for long.
The Next Nvidia? Why Supermicro Could Explode in a Chip-Hungry World
