MaxLinear (Nasdaq: MXL) raised its full-year 2026 optical data center revenue forecast for the third consecutive time Thursday, lifting the target to $250–$270 million from an initial projection of roughly $150–$170 million set at the start of the year — a cumulative upward revision of more than $100 million. The Carlsbad, California chipmaker simultaneously reported third-quarter net revenue of $218.1 million, up 72% year-over-year and 29% sequentially, driven by what CEO Kishore Seendripu called a “significant growth inflection” powered by surging demand for its AI data center optical chip platform.

Behind the headline numbers sits a supply-chain signal that carries more weight than a single-quarter beat: MaxLinear’s working capital expanded sharply in Q3, with accounts receivable climbing 44% sequentially to $73.4 million, prepaid expenses and other current assets jumping 60% to $121.6 million, and inventory rising to $113.7 million from $105.5 million the prior quarter. The company also amended its revolving credit facility with Wells Fargo and other lenders to add $70 million of capacity, bringing the total to $200 million. Working capital builds of this magnitude — especially the prepaid expenses line, which often reflects advance payments to semiconductor foundries to secure wafer capacity — are a durable demand signal: hyperscale customers pulling chip orders forward to lock in supply for committed GPU cluster build-outs.

Infrastructure Segment Posts Triple-Digit Growth for Second Consecutive Quarter

The infrastructure segment, which houses MaxLinear’s AI data center optical products, grew 210% year-over-year in Q3 — an acceleration from the 145% year-over-year growth that segment posted in Q2 2026 earnings results and the 136% growth in Q1. Infrastructure has been the company’s largest revenue category since Q2 and continued to pull away from MaxLinear’s legacy broadband and connectivity segments during the quarter.

Seendripu framed the results as confirmation of a durable platform: “Strong momentum in our optical AI data center business reflects the volume ramp of our Keystone PAM4 DSP platform, as well as the strength of our expanding infrastructure portfolio and customer activity centered around our Rushmore and Annapurna product families of 1.6T-capable optical and electrical products.” The full Q3 2026 earnings press release is available on the SEC website.

How Keystone’s 5nm Architecture Wins on Power — and Why That Matters for AI Data Centers

The chip at the center of MaxLinear’s run is Keystone, a family of digital signal processors (DSPs) and serializer/deserializer (SerDes) circuits fabricated at TSMC on a 5-nanometer CMOS process node. The platform uses PAM4 (Pulse Amplitude Modulation 4-level) signaling — a modulation technique that encodes two bits per symbol using four distinct amplitude levels rather than the two levels used in older NRZ (Non-Return-to-Zero) signaling. That doubles the data rate per lane without doubling the electrical bandwidth, enabling 400G and 800G optical transceivers to move more data through the same fiber infrastructure at 100 gigabits per lane (106.25 Gbps) on both the host electrical interface and the optical line side, as detailed in MaxLinear’s Keystone 5nm PAM4 DSP specification.

The tradeoff of PAM4 is that tighter amplitude spacing reduces noise margins, demanding more sophisticated equalization on the transmit and receive sides — exactly what MaxLinear’s integrated DSP and SerDes SoC provides. By combining both functions in a single 5nm chip, MaxLinear achieves what the company states is nearly 40% lower power consumption than competing solutions, as documented in MaxLinear’s JPC Connectivity partnership. That power advantage is increasingly decisive: as hyperscale AI data centers push toward rack densities of thousands of GPU accelerators, the thermal and energy budgets required to cool and power the optical transceiver layer become a hard engineering constraint, not just a cost consideration. A 40% reduction in transceiver power directly translates to lower cooling requirements and the ability to pack more optical ports per rack unit.

Q2 2026 earnings call commentary from CEO Seendripu reinforced the technical edge: “Delivering almost 40% lower consumption in power than competition, Keystone success serves as the foundation for multi-generational customer engagements that extend to the adoption of next generation 1.6 Tb and 3.2 Tb optical scale-up and scale-out architectures.” The Q2 2026 earnings call transcript with those remarks is available through MarketBeat.

Keystone is currently ramping in volume production at major hyperscale customers across the United States and Asia for 400G and 800G optical modules, spanning multiple optical transceiver form factors including pluggable optics (QSFP-DD, OSFP), linear pluggable optics (LPO), linear receive optics (LRO), near-packaged optics (NPO), co-packaged optics (CPO), and active electrical cables (AECs).

What the Guidance Raise Progression Reveals About AI Capex

MaxLinear’s three upward revisions to its 2026 optical data center revenue outlook tell a story about how hyperscaler demand visibility has improved across the year, as documented in the Q3 2026 full-year guidance table:

• Q1 2026 guidance: $150–$170 million (initial 2026 forecast)
• Post-Q2 revision: $210–$230 million, driven by stronger customer orders and program ramp visibility
• Post-Q3 revision: $250–$270 million (current), alongside Q4 revenue guidance of $240–$260 million

The pattern is consistent with how semiconductor supply chains respond when hyperscale customers commit to specific GPU cluster build-outs: as the customers formalize their infrastructure programs, they extend firm chip purchase orders further into the future, giving component suppliers like MaxLinear earlier and more confident revenue visibility. Visible Alpha consensus data, as of September 2026, had already forecast infrastructure revenue growing 169% year-over-year to $398 million for the full year — suggesting sell-side analysts had anticipated an acceleration beyond even MaxLinear’s raised guidance.

Q3 Profitability: Operating Leverage Flows Through

The revenue ramp is flowing to the bottom line with meaningful operating leverage. Non-GAAP operating margin expanded to 29.5% in Q3, up from 22.3% in Q2 and 12.1% in Q3 2025 — a 1,740-basis-point year-over-year improvement. On a GAAP basis, operating income reached $30.2 million, swinging from a GAAP operating loss of $41.3 million in the same quarter a year ago — a turnaround of more than 4,600 basis points.

Total GAAP operating expenses declined 4% sequentially to $97.8 million even as revenue scaled 29% sequentially — the textbook definition of operating leverage in a semiconductor product cycle. GAAP diluted earnings per share reached $0.30, compared to $0.02 in Q2 and a loss of $0.52 per share in Q3 2025. Non-GAAP diluted EPS came in at $0.62, up from $0.35 in Q2 and $0.14 in Q3 2025.

MaxLinear’s nine-month 2026 cumulative revenue of $524.1 million already substantially exceeds the company’s total 2025 revenue.

What Is MaxLinear’s Roadmap Beyond Keystone?

MaxLinear has two next-generation platforms already named and in development, with a third complementary product in market, as outlined in the Q2 2026 earnings call roadmap commentary:

Rushmore is a 1.6-terabit PAM4 DSP operating at 200 gigabits per lane — double Keystone’s 100G per lane. Rushmore targets next-generation 1.6T optical modules and is expected to begin contributing revenue in 2027. At OFC 2026, MaxLinear showcased Rushmore alongside Washington, its 200G-per-lane transimpedance amplifier, which can pair with Rushmore or deploy as a standalone analog front-end in LPO and NPO architectures that do not require a full DSP.

Annapurna is a 224G-per-lane scale-up copper retimer that delivers up to 1.6 Tbps across copper within AI data center racks — an electrical interconnect complement to the optical DSP business for shorter distances where copper remains practical. The three product families together address pluggable optics, LPO, LRO, NPO, CPO, and AEC form factors, a deliberately wide posture designed to capture demand regardless of which optical architecture hyperscalers ultimately standardize.

MaxLinear also secured its first XGS-PON hyperscaler design win for dedicated data center control-plane architectures, with qualification for a 2027 ramp already completed — evidence that the company is broadening its footprint inside hyperscale customers beyond pure optical DSPs.

How MaxLinear Competes Against Broadcom and Marvell

MaxLinear operates in a market dominated by significantly larger rivals. Marvell Technology (Nasdaq: MRVL) has reportedly held more than 80% of the 800G optical module DSP market by unit share, while Broadcom (Nasdaq: AVGO) in March 2026 announced its Taurus BCM83640 3nm optical DSP — a monolithic 3nm PAM4 optical DSP targeting 1.6T transceiver modules — to compete directly in the next optical generation. Marvell itself announced an expanded 1.6T DSP portfolio under the Ara family, currently shipping in volume to hyperscale customers.

MaxLinear’s differentiation strategy does not rely on matching the breadth of either competitor’s portfolio. Instead, the company has focused on winning specific high-volume platform engagements through energy efficiency, speed-to-market with 5nm, and deep technical integration with a focused set of hyperscale customers. The 210% year-over-year infrastructure growth in Q3 indicates that approach is yielding meaningful market share gains in the current 800G cycle. Whether the same strategy maintains traction in the 1.6T cycle — where Broadcom’s 3nm Taurus and Marvell’s Ara family are already in volume production — will be among the defining competitive questions of 2027.

According to LightCounting market data, 800G PAM4 chipset shipments nearly tripled in 2025 and are expected to more than double again in 2026, while 1.6T shipments are growing from a small 2025 base to tens of millions of ports. MaxLinear’s third guidance raise is therefore not an outlier — it reflects a broad market expansion in which multiple suppliers are reporting strong growth.

Other participants in the AI optical connectivity supply chain include Credo Semiconductor (Nasdaq: CRDO), Astera Labs, Coherent Corp., and a range of optical module makers, all reporting elevated demand as hyperscalers invest in the networking fabric interconnecting their GPU clusters.

Balance Sheet: Building Inventory, Securing Capacity

MaxLinear ended Q3 with cash and equivalents of $81.1 million, up from $64.8 million at the end of Q2, despite funding a working capital expansion required by the rapid revenue ramp. The $70 million revolving credit facility expansion (bringing total capacity to $200 million) provides additional liquidity headroom as the company pre-pays foundries and builds inventory ahead of continued production ramps.

The $45.7 million increase in prepaid expenses and other assets on the cash flow statement reflects the advance payments — a normal feature of a semiconductor ramp cycle in which wafer starts must be placed weeks or months ahead of when finished chips ship to customers. Long-term debt stood at $164.1 million as of quarter-end.

Q4 2026 guidance projects revenue of $240–$260 million, with non-GAAP gross margin of 59–62% and GAAP gross margin of 58–61%. The company guided non-GAAP operating expenses of $65–$70 million and GAAP operating expenses of $96–$101 million for Q4.

A webcast replay of the October 22 earnings call remains available at investors.maxlinear.com through November 5, 2026.

Frequently Asked Questions

What is MaxLinear’s Keystone chip and why does it matter for AI data centers?

Keystone is MaxLinear’s fifth-nanometer CMOS PAM4 digital signal processor (DSP) and serializer/deserializer chip family, designed for 400G and 800G optical transceivers deployed in hyperscale AI data centers. PAM4 signaling doubles data throughput per optical lane compared to older modulation schemes, while the 5nm manufacturing process enables what MaxLinear states is nearly 40% lower power consumption versus comparable competing solutions. In AI data centers deploying tens of thousands of GPU accelerators, the optical transceiver layer connecting those GPUs becomes a significant energy budget item; a 40% power reduction per transceiver directly improves the density and cost economics of large-scale GPU clusters. Full technical specifications are detailed in the Keystone 5nm PAM4 DSP announcement.

Why has MaxLinear raised its full-year optical revenue guidance three times in 2026?

Each successive upward revision reflects improved visibility into hyperscaler purchase commitments. As large cloud and AI operators finalize their GPU cluster programs, they extend firm chip orders further into the future, giving suppliers like MaxLinear progressively clearer revenue foresight. The progression — from an initial $150–$170 million annual target to $250–$270 million over three quarters — also reflects the working capital signals in MaxLinear’s balance sheet: prepaid expenses to foundries, inventory build, and an expanded credit facility all indicate the company is scaling production ahead of customer demand that hyperscale customers have already committed to. The complete Q3 2026 guidance progression is available in the SEC filing.

What comes after Keystone in MaxLinear’s roadmap?

The next generation is Rushmore, a 200G-per-lane, 1.6-terabit PAM4 DSP expected to begin contributing revenue in 2027. It will serve customers deploying next-generation 1.6T optical modules requiring double the per-lane data rate of Keystone’s 100G lanes. Alongside Rushmore, MaxLinear’s Annapurna copper retimer handles the electrical interconnect layer inside AI racks, and the Washington transimpedance amplifier provides the analog receiver front-end for LPO and NPO modules. Together, these three upcoming platforms are designed to keep MaxLinear positioned across the full spectrum of AI data center optical architectures through the next switching generation. Details on the platforms are available in the Q2 2026 earnings call roadmap discussion.

How does MaxLinear compare to Broadcom and Marvell in the AI optical chip market?

Marvell has held a dominant share of the 800G optical module DSP market, and Broadcom entered the 1.6T generation in March 2026 with its Taurus BCM83640 3nm optical DSP targeting next-generation 1.6T transceivers. MaxLinear’s strategy is narrower: it competes on energy efficiency (the Keystone 5nm power advantage) and deep engagement with a focused set of hyperscale customers rather than matching Broadcom’s or Marvell’s full AI portfolio breadth. The 210% year-over-year infrastructure growth in Q3 2026 suggests this focused approach is gaining share in the current 800G cycle, but the 1.6T transition starting in 2027 will be the true test of whether Rushmore can compete against Broadcom and Marvell platforms already in production at that generation. An overview of Marvell’s Ara family 1.6T expansion provides further competitive context.