Qualcomm Inside: SpaceX's 5G Bet Starts in 2027 With Custom Silicon

SpaceX plans to deploy Qualcomm-chip-based 5G basestations across tens of thousands of US mobile sites starting in 2027, a land-based move that Deutsche Telekom CEO Timotheus Höttges says would cost at least €200 billion (US$223 billion) to complete.

Starlink Mobile confirmed the plan with unusual brevity. "Starlink Mobile is building the first fully integrated US satellite and terrestrial network that will keep people connected on their phones no matter where they are. Our 5G basestations will leverage Qualcomm chips to accelerate deployment, starting in 2027," the company said on X. Elon Musk added: "At last there will be one network with good signal everywhere!"

Nearly every hard detail since has come from Qualcomm, not SpaceX. On last week's results call, Qualcomm CEO Cristiano Amon told analysts: "Starlink is building one of the most ambitious new cellular networks in the country, and our Dragonwing infrastructure platforms will bring together advanced massive MIMO, purpose-built RAN processing, and commercial-grade Layer 1 software."

What is radical and what is reactionary?

The plan hands the hardest engineering work to a 40-year-old American chipmaker that holds more of the most valuable 5G patents than any other player, by some measures, yet has never led the US network equipment market, where Ericsson, Nokia and Samsung dominate. It is also a tacit admission that satellites alone cannot disrupt mainstream 5G — and that the small cells SpaceX touted earlier this year appear to have been shelved.

SpaceX also shows no interest in the GPU-based AI-RAN concepts Nokia has promoted with Nvidia. Based on the planned investment, more money could be spent on Qualcomm's 5G network products in 2027 and 2028 than on GPU RAN.

Custom silicon has fallen out of favor at Nokia and Samsung, neither of which had market share to justify it — hence Nokia's Nvidia partnership and Samsung's move to Intel CPUs. Operators such as Verizon like the flexibility of general-purpose processors, whose software can shift between Intel, AMD and Arm platforms. But custom ASICs still outperform them in efficiency. Ericsson and Huawei keep investing in them; the investment case only works if volume follows. Qualcomm, which already sold a similar stack to Vietnam's Viettel on a much smaller footprint, is now betting SpaceX provides that volume.

What is missing from the picture?

Qualcomm supplies Layer 1 chips and software, used in both the distributed unit and the radio unit. It does not make RUs, nor does it provide Layers 2 and 3 software. Analyst Gabriel Brown of Omdia told Light Reading a contract manufacturer or white-label vendor could build the radios, and Layer 2/3 code would likely run on third-party CPUs from Intel, AMD or Qualcomm's own Arm-based range, with servers from Dell, HPE or Supermicro.

"Is Starlink going to build a system itself and be an integrator and write software?" Brown asked. "It can definitely do that because it's showing the abilities technically. But the big challenge is not just building it once and being happy with it. You have to be constantly churning it out, and that's where it doesn't scale that well unless you've got more than one customer or more than one market."

Rakuten's Japanese venture — whose Altiostar-developed Layer 1 is used outside Japan only by Germany's 1&1 — is the cautionary tale.

Why are incumbents nervous?

Qualcomm plans massive MIMO products with up to 64 transmitters and receivers per RU — double today's typical 32 — engineered for SpaceX's 50MHz of 2GHz spectrum rather than the usual 3.5GHz. That could partly offset SpaceX's spectrum shortage, Brown said. New spectrum arrives at government auction next year, and a fourth bidder would push prices up.

Höttges cited EchoStar's failed fourth-network attempt before its bankruptcy this summer: "We have just seen Charlie Ergen coming to the market and then disappearing unsuccessfully." Of Musk, he allowed: "He is in another category, maybe."

The economics matter beyond the US. T-Mobile earns about $50 per postpaid user monthly; Deutsche Telekom earns €19 ($21.20) in Germany. GSMA figures put capex per customer at €70 ($78.20) in leading markets versus €35 ($39.10) in Europe. Ericsson drew almost 40% of its first-nine-month revenues from US operators. If SpaceX compresses US ARPU toward European levels, that cushion disappears — a consequence more far-reaching than any rocket landing.