Altera's $2 Billion IPO Puts a Premium on the FPGA Comeback Story

The story around Altera this week is that the ailing FPGA maker is back. Reuters reported Thursday that Silver Lake and Intel-backed Altera is preparing to file confidentially for an initial public offering that could raise more than $2 billion, with a listing possible as early as this year, in what would rank among the largest semiconductor IPOs since Arm Holdings' roughly $5 billion debut in 2023. Barclays, Citi, JPMorgan and Morgan Stanley are the banks tapped as underwriters. The framing attached to it is that programmable chips are returning to the computing architecture "alongside GPUs" — the "nervous system" to the GPU's "brain," in CEO Raghib Hussain's phrasing, as AI moves to robotics and the edge.

That is a recovery narrative, and it deserves to be treated as one. The disclosed financial history that sits behind Altera's claim to growth is a business that just contracted by roughly half, in a market that barely has more than two real players. An IPO is a convenient place for a highly leveraged owner to mark the asset up; the question is what the underlying economics actually support.

The baseline the recovery is measured against is a trough

Altera's most recently disclosed numbers come from Intel, which still owns 49% after selling 51% to Silver Lake. Revenue fell from about $2.9 billion in 2023 to about $1.5 billion in 2024, a decline of roughly 48%, as customers diverted spending to AI GPUs, the company lost share to AMD's Xilinx, and the industry worked through inventory. On that $1.5 billion of 2024 revenue, Intel's disclosure showed a non-GAAP operating income of about $35 million and a GAAP operating loss of roughly $615 million.

This is the trap in the growth story. Hussain told Reuters in July that the company grew more than 20% in the prior year and expects growth of roughly 25% this year, and that operating income has more than doubled.But Altera is private and does not publish detailed financials, so those figures are self-reported and unaudited. They describe a recovery climbing off a floor whose last externally known level was $1.5 billion of revenue and roughly break-even operating margin on a non-GAAP basis. The improvement is real or not; as it stands, the reader has only the CEO's word for it, and that same disclosure window also reported the 48% decline. The S-1, when it lands, will be the first number that can actually be checked.

The mark being put on the table

The valuation history matters here, because it shows how much of the ask is narrative. Intel bought Altera in 2015 for roughly $16.7 billion. Ten years later, in the deal that closed last September, Silver Lake paid $4.46 billion for a 51% stake, valuing the whole company at $8.75 billion, about half of what Intel paid. Now an IPO is being prepared that could raise more than $2 billion. If that raise represents something on the order of a fifth of the company, it implies a valuation in the low double-digit billions — comfortably above the private mark Silver Lake set barely a year ago.

That is not automatically wrong. A duopoly leader with a plausible path back to growth can re-rate. But the re-rating is being engineered by the same owners who bought the asset at a 50% discount, at a moment when the U.S. IPO market sits at record fundraising levels. The structural claim concealed inside it is the one worth examining: whether the FPGA squeeze that produced the 2023–24 collapse was cyclical or permanent.

FPGAs are not one market — and that is Altera's real problem

The industry has bifurcated in a way that determines who captures value. On one side sits the data-center and communications-heavy core, where the AI capex boom reallocated spending decisively toward GPUs and custom accelerators — the flows that drained Altera's revenue. On the other sits the embedded, industrial and defense base — deterministic, often radiation-hardened programmable logic where Xilinx built a durable position and where Altera lost share. Altera's collapse happened precisely because its center of gravity sat in the segment that hyperscaler GPU spending crowded out, while its competitor held the segment that did not.

The robotics and edge-AI pitch is an attempt to plant a flag in a third sub-market that, so far, is mostly a roadmap: sensor fusion, protocol conversion, low-latency deterministic I/O loaded next to an AI accelerator. The company has produced working prototypes of six new chips, is developing on TSMC's 2-nanometer and 3-nanometer nodes alongside Intel Foundry, and has an expanded Arm partnership. It is also fabless, which cuts the other way from the usual supply-cycle reading: this is a market-position and share story, not a constrained-capacity story. Nothing about the capex-allocation mechanism guarantees the programmable layer reclaims the incremental spend that bypassed it.

There is some independent temperature check, and it is modest. AMD's embedded segment, which includes the former Xilinx, grew about 6% year over year in Q1 2026 after a prolonged downturn — stabilizing, but hardly the AI-adjacent boom the Altera IPO needs to justify a double-digit-billion mark.

The forward condition

The key issue is not whether Altera can sell a block of shares in a hot market; it almost certainly can. The more important question is whether the audited growth in the S-1 confirms a recovery off a $1.5 billion trough, and whether the edge-and-robotics spend that "nervous system" claim depends on actually shows up in the revenue of a two-player market. If the disclosed recovery is real, a re-rating above the $8.75 billion Silver Lake mark is supportable. If the growth turns out to be measured against a steeply depressed base and the FPGA share of AI investment stays where it was during the GPU upcycle, then the IPO is pricing a cyclical comeback that the same disclosures that marked the company down by half would not support. The S-1 is the first chance to tell them apart, and it is worth waiting to see its numbers before accepting the CEO's.