Broadcom and Apple just locked in a custom chip partnership that runs through 2031, and the real story isn’t the radios and Wi-Fi silicon they’ve always built together — it’s the new AI server chips buried in the fine print. The two companies confirmed on July 6, 2026 that they’re extending and expanding their application-specific integrated circuit (ASIC) collaboration, with Broadcom now supplying silicon for Apple’s in-house AI server chip, internally codenamed Baltra, that’s targeted to start rolling out as early as 2027.
I’ve been tracking Apple’s chip strategy since the first Apple Silicon Macs shipped, and this deal is a bigger deal than the headline “Apple renews a supplier contract” makes it sound. Apple isn’t just buying components anymore — it’s co-designing the compute layer that Apple Intelligence will run on for the rest of the decade.
What Broadcom actually builds for Apple
Broadcom has quietly been one of Apple’s most important suppliers for years, and most of it never shows up in a keynote slide. The company supplies:
- RF (radio frequency) chips that handle cellular connectivity in every iPhone
- Wi-Fi and Bluetooth combo chips used across iPhone, iPad, Mac, and Apple Watch
- Networking and connectivity semiconductors for Apple’s broader device lineup
- Custom ASIC silicon, now expanding into AI server infrastructure
None of that is glamorous. It’s the plumbing. But plumbing is exactly why this extension matters — Apple doesn’t renew seven-figure, multi-year supply agreements with companies it’s planning to walk away from.
The AI server angle: meet “Baltra”
The part of this deal that actually got my attention is Baltra. Apple has reportedly been developing its own AI server chips for cloud-side inference — the kind of hardware that would run Apple Intelligence’s heavier workloads in Apple’s own data centers instead of leasing capacity from Nvidia, AWS, or Google Cloud. Broadcom’s ASIC design expertise is going into those chips, and the current timeline points to an early rollout in 2027.
This matters because Apple has taken a very different path than Microsoft, Google, and Amazon on AI infrastructure. Those three have been racing to build (or rent) the largest GPU clusters they can get their hands on. Apple, true to form, is trying to design its way around that arms race with custom silicon tuned specifically for its own workloads — the same playbook that gave us the M-series chips beating Intel on efficiency.
| Fact | Detail |
|---|---|
| Deal announced | July 6, 2026 |
| Deal runs through | 2031 |
| Chip type | Custom ASIC (application-specific integrated circuit) |
| New AI chip codename | Baltra |
| Target rollout for AI server chip | As early as 2027 |
| Existing Broadcom products for Apple | RF, Wi-Fi/Bluetooth, networking chips |
Why Apple builds its own chips instead of buying off the shelf
If you’ve followed Apple’s hardware for a while, this fits a pattern you’ve seen before. Apple moved the Mac off Intel and onto its own Apple Silicon because it wanted tighter control over performance-per-watt. It’s done the same with the A-series and M-series chips inside iPhones and iPads for over a decade. Server-side AI silicon is just the next domino.
There’s a business logic here too. Nvidia’s H100 and Blackwell-generation chips are in such high demand that even the biggest cloud providers are rationing allocation. If Apple can design an AI server chip tuned specifically for the kinds of workloads Apple Intelligence needs — on-device-style efficiency, privacy-preserving inference, tight integration with its own software stack — it sidesteps both the GPU shortage and Nvidia’s pricing power. Broadcom, which already helped Google design its in-house TPU chips, is exactly the partner with the ASIC chops to pull that off.
What this means if you’re not an Apple shareholder
You don’t need to own AAPL stock for this to matter. A few practical takeaways:
- Apple Intelligence should get faster and more capable over the next few years as Apple’s own server infrastructure comes online instead of leaning entirely on third-party cloud compute.
- Your iPhone’s connectivity chips aren’t going anywhere. Broadcom staying locked in through 2031 means continuity in the RF and Wi-Fi hardware that’s been reliable for years — no surprise supplier switch that tends to introduce early bugs.
- The AI chip shortage story just got another data point. When even Apple, sitting on one of the largest cash piles in corporate history, chooses to design custom silicon rather than simply outspend everyone for Nvidia GPUs, that tells you how tight the high-end AI chip market really is.
For readers who’ve been watching component prices climb — we covered how AI data centers are driving RAM and SSD prices up as much as 500% — this is the same underlying pressure showing up from a different angle. Every major tech company is now trying to secure its own compute pipeline, whether that’s memory, GPUs, or in Apple’s case, custom ASICs.
How this compares to building your own hardware
It’s a strange parallel, but the logic Apple is using at the data-center scale is the same logic I’d give anyone shopping for a graphics card for their own PC right now: buying exactly what you need for your actual workload beats overpaying for generic top-of-the-line hardware you don’t need. Apple doesn’t need a general-purpose Nvidia GPU cluster if a custom ASIC tuned to its own software does the job more efficiently. The same reasoning applies whether you’re speccing a gaming rig or a hyperscale data center — just with a few extra zeros on the budget.
FAQ
What is Broadcom’s new deal with Apple?
Announced July 6, 2026, it’s an extension and expansion of their existing custom chip partnership, running through 2031. It covers Apple’s established RF, Wi-Fi, and Bluetooth chips, plus new custom ASIC silicon for Apple’s in-development AI server chips.
What is “Baltra”?
Baltra is the internal codename reported for Apple’s own AI server chip, designed with Broadcom’s help, aimed at supporting Apple Intelligence’s cloud-side infrastructure. It’s targeted for an initial rollout as early as 2027.
Does this affect my iPhone or Mac directly?
Not immediately. The connectivity chips (Wi-Fi, Bluetooth, cellular) you already rely on are unaffected — this deal mostly locks in that supply. The bigger long-term effect will be on how fast and capable Apple Intelligence’s cloud features become once Apple’s own AI servers come online.
Why doesn’t Apple just buy Nvidia GPUs like everyone else?
It’s doing some of that too, but Apple has a long history of preferring custom silicon tuned to its own software over generic off-the-shelf chips — the M-series Macs are the clearest example. Custom ASICs can also help Apple avoid competing for scarce, expensive Nvidia GPU allocation.
Sources: MacRumors, Bloomberg.
Jared is a tech journalist covering product launches, industry news, and the culture around technology. He has been reporting on the consumer tech beat for more than eight years.