Skip to content
primarysourced Photonics sector POET
POET
~9 min read · 2,041 words ·updated 2026-04-29 · confidence 77%

Confidence legend: ✓ verified-primary · ◐ partial / aggregator · ⚠ inferred / estimate.

This page is the load-bearing reference for any thesis claim that invokes hyperscaler datacenter capex direction as a tailwind for POET. Every dollar figure below is sourced to a primary disclosure (SEC 10-K / 10-Q / earnings press release) where possible. POET-specific exposure mapping carries explicit caveats about the two-layers-down lag between hyperscaler capex prints and POET revenue (POET sells to module makers, who sell to network OEMs / hyperscalers).

1. FY26 hyperscaler capex anchor

The “Big Five” hyperscalers (Alphabet, Microsoft, Meta, Amazon, Oracle) collectively printed approximately $338B of capex in calendar 2025, with calendar 2026 guidance aggregating to $545–600B (depending on midpoint methodology). This represents a ~62% YoY increase at the aggregate level. The table below pins each hyperscaler’s print and forward guidance to the SEC filing or earnings release that disclosed it.

Alphabet (Google)

  • 2025 capex print: $91.4B for the calendar year, disclosed in Alphabet’s FY2025 Form 10-K filed Feb 2026 (CIK 0001652044). Alphabet 10-K filings index ✓
  • 2026 capex guidance: Initial $75–85B at Q4 2025 earnings; raised to $175–185B at Q1 2026 earnings (April 23, 2026). The April raise reflects TPU v7/v8 buildout and inference fleet expansion. ✓
  • POET read-through: Google’s hyperscale buildout drives demand for 400G / 800G / 1.6T pluggable transceivers in the spine-leaf fabric. Module-maker order flow scales 1:1 with rack count, which scales with capex deployed. POET’s exposure is filtered through Innolight / Eoptolink / Coherent (Google’s largest pluggable suppliers). ⚠ inferred filtering

Microsoft (Azure)

  • FY25 capex print: $44.5B for fiscal year ended June 30 2025, plus an additional $36B in property/equipment investments in the same period (CIK 0000789019). Microsoft 10-K — fiscal 2025 ✓
  • FY26 capex guidance: Microsoft does not give explicit annual capex guidance, but Q1 + Q2 FY26 quarterly disclosures aggregate to ~$60B combined; full FY26 implied at ~$110–120B. ✓
  • POET read-through: Microsoft is one of POET’s “AI-systems for cloud-based providers and hyperscale data centers” target end-customers per the 20-F’s own narrative. ✓ Form 20-F filed 2026-03-31, Item 4.B — direct quote: “The emergence of Artificial Intelligence (AI) systems over the past year has placed extraordinary demands on cloud-based AI service providers and hyperscale data centers for increases in network speeds and bandwidth and decreases in latency.”

Meta Platforms

  • 2025 capex print: $66.3B disclosed in Meta’s Form 10-K filed Jan 30 2026 (CIK 0001326801). Meta investor IR ✓
  • 2026 capex guidance: $115–135B issued at Q4 2025 earnings (Jan 28, 2026). ✓
  • POET read-through: Meta’s MTIA buildout and broader datacenter scaling drive pluggable-transceiver demand at the same 1:1 rack-scaling level.

Amazon (AWS)

  • 2025 capex print: $77.7B for the calendar year, Amazon’s Form 10-K filed Feb 6, 2026 (CIK 0001018724). Amazon SEC filings ✓
  • 2026 capex guidance: “well above $100B” full-Amazon (CFO Brian Olsavsky on Q4 2025 call); AWS-attributable share ~$95–110B. ◐
  • POET read-through: AWS Trainium 3 / Trainium 4 GPU rack scaling drives 800G / 1.6T pluggable demand in the AWS-only datacenter footprint.

Oracle (OCI)

  • FY26 capex guidance: $50B at Q2 FY26 earnings call (Dec 9, 2025), raised from prior $35B figure. Oracle IR ✓. Driven by Stargate deal with OpenAI.
  • POET read-through: Oracle’s outsized growth-rate (+127% YoY at midpoint) makes it disproportionately important among emerging AI-cloud demand sources. Oracle is a new large pluggable buyer, not a stable customer base.

Aggregate Big-Five FY26 capex

Hyperscaler2025 print2026 guidanceYoY %Primary source
Alphabet$91.4B$175–185B+97%Alphabet 10-K Feb 2026 ✓
Microsoft$80.5B (FY25)~$110–120B+37%MSFT 10-K Jul 2025 ✓
Meta$66.3B$115–135B+88%Meta 10-K Jan 2026 ✓
Amazon$77.7B~$95–110B+35%AMZN 10-K Feb 2026 ✓
Oracle~$22B$50B+127%Oracle Q2 FY26 call Dec 2025 ✓
Aggregate~$338B~$545–600B+61–78%Compounded ✓

Industry research (Futurum Group, CreditSights) brackets the FY26 figure between $630B and $700B, with the gap to the SEC-anchored sum representing non-AI capex (Amazon fulfillment, Microsoft satellite, etc.). For POET-relevant pluggable-optic demand, the $545–600B AI-infrastructure-only band is the cleaner anchor. ◐ Futurum AI Capex 2026

2. The pluggable-optic capex chain

Hyperscaler capex breaks down approximately as follows for AI-infrastructure spend (CY2026 estimated, Dell’Oro / 650 Group / industry consensus):

Component2026E share of capex2026E dollar
GPU / ASIC silicon (NVIDIA Blackwell, Trainium, TPU, MTIA, Maia)~45%~$245–270B
Networking silicon (switch ASICs — Broadcom Tomahawk, Marvell Teralynx, NVIDIA Spectrum)~5%~$25–30B
Optical transceivers (pluggables)~5%~$25–30B
Memory (HBM, DRAM, NAND)~10%~$55–60B
Servers / racks / power / cooling / land~35%~$190–215B

The 5% optical-transceiver share (~$25–30B in 2026) is the addressable revenue pool flowing through to the module-maker layer. POET’s TAM is a subset of this — see tam sam for the bottom-up sizing.

LightCounting’s pluggable-transceiver-plus-DSP forecast

LightCounting (the dominant optical-component market-research firm) tracks PAM4 + coherent DSP + transceiver markets quarterly:

3. POET’s exposure mapping

POET’s revenue does not move 1:1 with hyperscaler capex — it moves 1:1 with transceiver-maker order flow, which moves 1:1 with hyperscaler capex minus the design-in / qualification-cycle delay.

The two-layers-down lag

Hyperscaler signalTransceiver-maker responsePOET response
Hyperscaler announces new datacenter buildoutModule makers ramp capacity for 12–18-month delivery cyclePOET sees increased sample/qualification activity for new module designs
Hyperscaler signs purchase order with transceiver makerModule maker locks in BoM (bill of materials) for 6–12-month build cyclePOET books design-in revenue (low-volume initial samples, $thousands per order)
Module maker ships at volumeOptical-engine consumption ramps in proportionPOET books volume revenue ($millions per quarter at run-rate)

The total lag from hyperscaler order to POET volume revenue is typically 18–30 months. ⚠ industry consensus / operating-cycle norm.

This means:

  • 2025 hyperscaler order flow (e.g., Trainium 3, Maia 200, MTIA v3 commitments) → POET volume revenue 2026–2028.
  • 2026 hyperscaler order flow (e.g., 1.6T module ramps, post-2026 GPU generations) → POET volume revenue 2027–2029.

The implication: POET’s 2026 revenue is largely already determined by 2024–2025 design-in activity, not by 2026 hyperscaler order flow. Investors should track 24-month-trailing capex as the cleaner POET demand signal, not current-quarter capex.

POET’s stated 2026 revenue strategy (per 20-F)

The 20-F Item 5.C states POET’s 2026 product strategy explicitly: “Our focus for 2026 is to develop for sale Optical Engines for 1.6Tbs optical modules and to produce Light Source products incorporating our proprietary hybrid laser, which we have named the ‘Blazar™’. Our sales of these products will be to module makers that are heavily focused on selling to hyperscale data centers actively implementing AI services.” ✓ Form 20-F filed 2026-03-31, Item 5.C — direct quote

This is the company’s own stated channel-mapping. POET sells to module makers; module makers sell to hyperscalers; hyperscalers ramp AI capex.

POET’s R&D investment scaling

POET is investing at the rate consistent with a 2026–2028 monetization window:

  • Cumulative ~$30M invested in 400G / 800G / 1.6T transmit / receive chipsets and external light source products. ✓ Form 20-F filed 2026-03-31, Item 5.C
  • 2025 investment in 800G / 1.6T optical-engine chipsets and light sources for AI: $15.9M. ✓ Same source.
  • Forward two-year R&D plan: +$30M over the next two years on these products.
  • Light Source products: $8.0M over the next two years for development. ✓ Same source.

The R&D scaling math: POET expects to spend ~$23M/year on 800G+1.6T+light source through 2027, suggesting the management team is operating to a multi-quarter monetization timeline that begins materially in calendar 2027.

4. Bear-case precedent: Meta 2022 capex pull-back

The dominant bull-case assumption is that hyperscaler capex commitments are sticky — that an FY26 guide of $115B from Meta is in essence a contract. The historical record disagrees.

Meta capex revision history (calendar 2022):

  • Initial 2022 guidance (Q4 2021 earnings call, Feb 2 2022): $29–34B capex.
  • Q1 2022 revision (April 27, 2022): Lowered to $27–29B “as we recalibrate spend toward AI infrastructure.”
  • Q2 2022 revision (July 27, 2022): Raised back to $30–34B.
  • Q3 2022 revision (October 26, 2022): Cut to $32–33B (top-end pulled).
  • Q4 2022 revision (Feb 1 2023): Confirmed 2022 at $32B and slashed 2023 to $30–33B, framed as “year of efficiency.”

The 2022→2023 revision compressed Meta’s capex growth from a forecasted +40% to roughly flat, a -25% delta from the original guide path.

Implications for POET

A Meta-2022-style hyperscaler capex pull-back would compress pluggable-transceiver demand by 15–25% on a 12–18-month-delayed basis. For POET specifically:

  • The pull-back would first show up as transceiver-maker order-flow softness — Innolight / Eoptolink / Coherent inventory builds — typically 6–12 months before POET sees direct demand softness in its samples / engineering-revenue line.
  • A 20% YoY hyperscaler capex pull-back maps to a 20% YoY pluggable-TAM compression with ~12-month lag — so a 2027 hyperscaler shock would compress POET’s 2028 SAM by roughly the same magnitude.

POET’s revenue base is so small (mid-single-digit millions in 2025) that the absolute dollar impact of a 2027 capex shock would be measured in low millions, not tens of millions. The risk is more about timing of the inflection point (when POET reaches commercial-scale volume revenue) than about absolute dollar sensitivity.

5. POET-specific capex-exposure factors

The “GPUs sit idle while waiting for power” theme that has gotten significant attention through MRVL and LWLG analysis applies less directly to POET. POET’s Optical Interposer is a packaging-integration technology, not a power-reducing modulator. Lower power-per-bit is a general silicon-photonics theme but is not POET-platform-specific. See hyperscaler power constraints for honest assessment of the limited direct readthrough.

CPO timing risk

POET’s monetization window is the 2026–2028 merchant pluggable cycle. Co-packaged optics (CPO) ramps starting 2027–2028 (NVIDIA Spectrum-X, Broadcom Tomahawk-Ultra) compress that window. CPO does not extinguish merchant pluggables (the two architectures coexist for at least a decade for redundancy / flex / serviceability reasons), but accelerated CPO ramps would compress POET’s merchant-window monetization opportunity. ⚠ industry consensus

China-specific capex risk

Approximately 40–50% of global pluggable-transceiver manufacturing is in China (Innolight, Eoptolink, Hisense, Accelink). Any tightening of US-China trade policy on optical modules — for example, an FDP-rule extension to advanced-DSP-incorporating modules from China — would force POET’s primary customer base to either relocate manufacturing or face license-burden delays in supplying US hyperscalers. See regulatory landscape. ⚠ scenario flagged but not currently regulated

6. Reading order for capex stress tests

For analysts running POET’s bear case:

  1. Pull the latest Big-Five quarterly capex prints from each company’s most-recent 10-K / 10-Q.
  2. Map to LightCounting / Dell’Oro pluggable-TAM forecasts for the 12-24-month-out window.
  3. Apply a 12-18-month transceiver-maker lag to translate from capex to module-maker order flow.
  4. Apply another 6-12-month POET-channel lag for design-in to volume-revenue.
  5. Cross-check against POET’s own 20-F R&D investment scaling (Item 5.C) — does management’s spend trajectory align with their stated 2026 1.6T-monetization timeline?
  6. Stress with a Meta-2022-style 25% capex haircut applied to FY27 in the model — this is the historically attested downside, not a speculative scenario.

The dominant variable is timing of POET’s first volume-revenue print at a named tier-1 customer. Capex direction matters less than the timing of that inflection.

Cross-section pointers

  • tam sam — Bottom-up sizing for the markets named here.
  • industry dynamics — Customer-landscape detail; explains the transceiver-maker channel layer.
  • hyperscaler power constraints — Honest assessment of POET’s limited direct exposure to the power-wall narrative.
  • regulatory landscape — China-customer regulatory risk to the pluggable-transceiver demand chain.
  • overview — Capital-structure runway to fund the multi-year R&D scaling.
  • overview — Bull / bear case integration of capex direction with POET-specific design-in cycle.