POET / MXL / HIMX: three layers of AI interconnect optionality

Investment question: these are not the same AI semiconductor trade

POET, MXL and HIMX can all be mapped to AI interconnect, optics, CPO or edge AI, but the first-principles layers are different. POET is a manufacturing-platform option: can Optical Interposer / EOI move optical-engine production closer to wafer-scale integration? MXL is the signal-chain layer: optical modules and AI data-center links need PAM4 DSPs, TIAs and retimers. HIMX is a display and ultra-low-power sensing company with AR, WiseEye and CPO optionality through FOCI.

The high-odds question is therefore not simply which name is an AI stock. It is which layer has already entered revenue, which remains a purchase-order framework, and which is still mostly a 2027 option.

POET: highest optionality, highest need for skepticism

POET is the nonlinear name. Q1 2026 NRE and product revenue was only $503,389, yet the same day the company disclosed a Lumilens initial $50M purchase order and a five-year framework that could exceed $500M. Relative to current revenue, that is the largest asymmetry in the group.

But an order is not revenue. The POET/Lumilens release says fulfillment and revenue depend on successful development, qualification and manufacturing scale-up; engineering samples are expected in late 2026 and production ramp aligns with 2027 hyperscaler deployments.

The three checks are Lumilens quality and funding, warrant dilution / fully diluted share count, and whether 2H26 brings verifiable sample, qualification, shipment and revenue-recognition milestones.

MXL: best financial proof, highest valuation hurdle

MaxLinear has the cleanest financial proof. Q1 2026 revenue was $137.2M, up 43% Y/Y; Infrastructure grew 136% Y/Y and became the largest end market; Q2 guidance is $160M-$170M with non-GAAP GM of 58%-61%. Management explicitly tied the inflection to optical data-center products ramping at multiple hyperscale AI platforms.

That makes MXL the name where the AI interconnect story is already entering the income statement. Keystone 400G/800G PAM4 DSP is ramping, while Rushmore 200G/lane DSP and Washington 200G/lane TIA point toward 1.6T.

The risk is valuation catching up first. The next test is whether Q2 is the start of a durable slope, not a one-quarter step function: optical revenue, operating leverage and 2027 Rushmore/Washington production language must keep improving.

HIMX: less pure, more like a portfolio option

Himax is not a pure data-center optics name. Its base is display drivers, automotive Tcon, WiseEye AI, LCoS microdisplays and CPO exposure through FOCI. Q1 2026 revenue was $199.0M, Q2 guidance calls for 10%-13% sequential growth and gross margin around 32%, and the company had $287.6M of cash, equivalents and other financial assets.

The upside comes from two option layers. WiseEye and LCoS give Himax a smart-glasses route: a leading brand has adopted WiseEye with mass production expected later in 2026. CPO gives the second route: Himax/FOCI Gen 1 supports 1.6T/3.2T with small shipments in 2H26, while Gen 2 targets 6.4T and 2027 volume ramp.

The failure condition is timing. If display cycles drag the base business or WiseEye/AR/CPO do not enter revenue quickly enough, the market may refuse to assign a premium for the option stack.

Ranking the checks, not issuing a trade signal

On pure asymmetry POET ranks first; on revenue proof MXL ranks first; on risk-adjusted holdability HIMX may be easier to underwrite. My review order is: POET order-to-revenue and dilution, MXL post-Q2 revenue slope and operating leverage, HIMX 2H26 automotive/WiseEye mass production and 2027 CPO readiness.

The common thread is that AI data movement is spilling value from GPUs into optics, electronics, packaging and edge sensing. The most dangerous mistake is pricing a 2027 option as if it were already 2026 revenue.