The call came in quietly. Over the past weeks, executives from Apple’s silicon group have reached out to a handful of semiconductor startups to sound out interest in an acquisition, according to people familiar with the discussions. The targets are small teams working on the kind of technology Apple wants for its own AI computing hardware. None of the conversations has produced a deal, the people said, but the outreach signals how far the company is willing to go to fix a problem inside its most ambitious hardware project.
The problem is called Baltra. That is the internal code name for the follow-on version of Apple’s AI server chip, which the company had planned to release this year. The project has been pushed back, according to people familiar with the matter, and Apple has concluded that its in-house AI server performance is falling short of what its engineers need. The acquisition talks, first reported earlier this week, are one response to that shortfall.
Apple’s interest in startups is a shift in method, not in strategy. The company has spent years moving its AI workloads off other vendors’ chips and onto silicon it designs itself. The effort, which grew out of work that began around the launch of its on-device AI features in 2024, is meant to give Apple control over cost, power and capability in the data centers that run features such as Siri and its cloud-based writing tools. Building that capability from scratch has proven slower than planned.
Buying it has a precedent. Apple has a long history of absorbing small chip companies rather than building every capability internally. In 2008 it bought PA Semi, a processor designer of roughly 150 people, for about $278 million, a deal that laid the groundwork for its homegrown A-series chips. In 2010 it acquired Intrinsity, which had worked with Samsung on fast mobile processors. In 2012 it purchased Anobit, an Israeli flash-memory controller specialist, for a reported $500 million. Each acquisition brought engineers whose work showed up in later products.
The current shopping list is different. The startups Apple has approached are working on technologies relevant to AI servers, including accelerators, memory interfaces and the high-speed links that move data between chips, according to people familiar with the matter. These are precisely the components where Apple’s internal effort has struggled to keep pace with the demands of training and running large language models.
The delay matters because Apple’s competitors have not stood still. Microsoft, Google, Amazon and Meta have all built or commissioned custom AI chips, and Nvidia has kept its stranglehold on the broader accelerator market. Analysts said Apple’s AI server program is running years behind the internal chip programs of some rivals, and that the gap shows up in the features Apple can ship to customers.
For Apple, the stakes extend beyond the data center. The company has framed its AI features as a reason to buy an iPhone, and its ability to run those features on its own hardware affects both cost and performance. A server chip that misses its schedule leaves Apple leaning on rented capacity from cloud providers, a position its top executives have said they want to exit.
Acquiring a startup would not solve everything. Small teams bring intellectual property and a handful of engineers, but integrating them into Apple’s process-heavy development culture takes time, and the technologies Apple is shopping for are exactly the areas where talent is scarce and valuations have run up. Some of the companies approached have already been bought or are tied to exclusive deals with other chipmakers, people familiar with the matter said.
There is also the question of regulators. Apple’s acquisition of smaller chip companies would face scrutiny in the U.S. and Europe, though deals of the size under discussion would be unlikely to trigger the same review as a large merger. The company has largely sat out the acquisition spree that reshaped the semiconductor industry over the past two years, a period when consolidation among chip designers accelerated sharply.
Apple is also keeping its options open. The company continues to hire for its server chip team and has explored partnerships with outside designers, according to people familiar with the matter. A purchase, a licensing deal and an extended internal push are all on the table, and executives have not settled on one path.
The broader lesson for the industry is that even Apple, with one of the largest cash piles in technology, cannot will an AI chip into existence on schedule. The company’s server silicon program shows how hard it is to build advanced chips from scratch, and how expensive the alternatives have become.
People familiar with the matter cautioned that the outreach is early and could come to nothing. Apple typically evaluates many companies before moving, and the talks described this week may simply reflect that process. The company declined to comment on the discussions.
What is clear is that Apple’s AI infrastructure is no longer a background project. The company has committed resources and public statements to its AI push, and its data center plans are now central to that effort. Whether the gap gets closed by a deal, by engineering or by patience will be visible in the products Apple ships over the next two years.
For now, the phone calls keep going out. In Cupertino, the message to the startup world is simple: Apple is in the market for chip talent, and it is willing to pay for it. The only question is whether any of the conversations turns into a signature.
