Powering AI From Orbit: MACOM, Coherent, STM, Hamamatsu (Pt.4)

Powering AI From Orbit: MACOM, Coherent, STM, Hamamatsu (Pt.4)

Heritage determines the clear near-term winners in Space DC optical networking (2026–2029), while inter-orbit links represent the only category of truly incremental photonic demand.

Summary

• Heritage is king in the 2026–2029 period, positioning Coherent, MACOM, STM, and Hamamatsu as the clear near-term winners in Space DC optical networking.

  
  

• MACOM stands out as the top pick, uniquely combining near-term heritage with superlinear content growth as per-link bandwidth scales to multi-terabit levels.

  
  

• Inter-orbit relay links are the only category generating purely incremental demand with no terrestrial equivalent.

  
  

• For most suppliers, Space DC demand represents volume migration from terrestrial data centers rather than net new TAM expansion.

  
  

• Long-term (2030–2035), the architectural advantage shifts toward terrestrial AI photonic and co-packaged optics companies that bring proven high-volume economics to space.

  
  

Current Winners: Coherent, MACOM, STM, Hamamatsu

These four suppliers are the clear near-term leaders (2026-2029) because they already have radiation- hardened, vacuum-qualified components flying or baselined in operational space-to-space and space-to-ground laser communication programs. SpaceX and other constellation operators do not issue open RFPs and wait for custom designs; they identify which parts survive the space environment and integrate around them. Heritage is paramount in this first-generation.

Coherent (COHR)

Coherent stands out on two fronts. First, it supplies Erbium-Doped Fiber Amplifiers (EDFAs) - optical boosters that increase laser transmit power - for any link exceeding roughly 500 km where the power budget is tight (i.e., the received optical signal is barely strong enough for reliable data transmission, with little margin for distance, pointing errors, or other losses - primarily inter-orbit relay links). Second, it provides precision optics, specialized coatings, and complete telescope assemblies used across virtually all terminal types. Coherent’s end-to-end vertical integration (raw crystal growth → polishing → coating → assembly) is unmatched and addresses a genuinely new category of free-space optics hardware with no terrestrial parallel.

Coherent’s EDFA opportunity is narrower than it first appears: most intra-cluster links inside a LEO constellation are short enough that amplification is unnecessary, limiting the addressable market to longer inter-orbit routes. Revenue will therefore come mainly from (1) precision optics (moderate volume, moderate margin) and (2) EDFAs for the longest links (low volume, high-margin). Both are real near-term wins, but neither is expected to be transformative at Coherent’s overall corporate scale.

MACOM (MTSI)

MACOM wins near-term on space-qualified analog components and wins increasingly over time as per-link bandwidth scales. The near-term story is heritage: MACOM supplies TIAs (receiver amplifiers) and drivers to multiple free-space optical programs including Mynaric CONDOR terminals and classified defense programs. This heritage translates directly to SpaceX terminal designs. The medium-term story is the bandwidth scaling argument above — at 6.4T–12.8T per link, MACOM's InP analog content per terminal increases 5–20x versus current-generation terminals, with no silicon-based alternative at these speeds that meets satellite power constraints. MACOM is the company in this analysis with the most favorable dynamics across both near-term heritage advantage and long-term bandwidth scaling. Current market pricing doesn't fully reflect future space revenue.

STMicroelectronics (STM)

STMicroelectronics wins near-term on PAT control ASICs and power-management chips — both completely new demand categories with no terrestrial equivalent. Every inter-satellite laser terminal needs a high-speed pointing control loop (called PAT: Pointing, Acquisition, and Tracking) to keep its narrow laser beam locked onto the target satellite despite vibration, orbital motion, and thermal changes. STM’s 28 nm FD-SOI semiconductor process is the perfect fit: it delivers radiation tolerance without the high cost of fully rad-hard chips, runs fast enough for 1,000+ Hz PAT loop updates, and is priced at automotive-grade levels that work for million-unit constellations.

As terminal data rates climb (from today’s ~10–25 Gbps toward 50–100+ Gbps per channel), STM loses the analog laser driver and TIA sockets to MACOM. MACOM specializes in ultra-high-performance photonic analog chips that deliver the higher bandwidth, lower noise, better linearity, and lower power consumption these faster links require. STM keeps the control and power sockets; MACOM takes the speed-critical analog front-end. (For the simplest short-range intra-cluster links these functions may also fold into silicon-photonics chips.) But the PAT control ASIC socket is permanent. It is the clearest example of genuinely incremental demand created by orbital data centers — a chip that does not exist on Earth and cannot be displaced by any photonic integration trend.

Hamamatsu

Hamamatsu wins near-term on InGaAs APDs (high-sensitivity avalanche photodiodes) for any link where received optical power is weak — mainly inter-orbit relays and the longer intra-cluster connections.

InGaAs is a III-V compound semiconductor (the same advanced material family as the GaAs discussed earlier in the Space Solar series; III-V materials combine elements are far superior to silicon for detecting infrared laser light). Hamamatsu’s 70 %+ market share comes from decades of specialized III-V detector R&D that competitors cannot copy quickly.

For short intra-cluster links where received power is plentiful, cheaper integrated germanium detectors built directly onto silicon-photonics (SiPh) chips displace Hamamatsu. But for inter-orbit links and any connection pushing the sensitivity limit at 6.4 Tbps+ bandwidths, the superior quantum efficiency and built-in internal gain of InGaAs APDs remain irreplaceable.

The addressable terminal count is the smallest of any company in this analysis, but the content per terminal is high ($30–80 for the APD array at 6.4 Tbps bandwidths, which typically needs 8–16 detector channels). Hamamatsu’s recent acquisition of NKT Photonics also positions it as a credible second source for EDFAs alongside Coherent.

Shifted Demand: Lumentum and Tower Semiconductor

For these companies, it is a volume shift, not a volume add. The investment case rests on secondary effects, not raw demand expansion.

Lumentum (LITE)

Lumentum supplies the EML die at the heart of every intra-cluster terminal. This position is structurally protected because silicon cannot emit light — every SiPh PIC needs an external III-V laser source, and Lumentum's integrated EML (laser + modulator on one InP die) is the highest-performance, lowest-cost option. But the total number of EMLs deployed in intra-cluster space terminals roughly corresponds to the transceiver lasers that would have been deployed in terrestrial data centers serving the same compute capacity. It is demand migration, not creation.

Where Lumentum does benefit: the pricing premium from space qualification adds 30–100% to per-die revenue in the early years. The mix shifts toward direct detection (EMLs) and away from coherent (where Lumentum competes with Coherent for the tunable laser socket), which may favor Lumentum's NeoPhotonics-derived product line. And inter-orbit links represent genuinely new EML demand. But the aggregate effect is a modest revenue uplift — not a transformative TAM expansion. Lumentum remains a strong hold for the Nvidia co-investment thesis, terrestrial co-packaged optics roadmap, and the structural fact that InP light emission cannot be integrated onto silicon. But the Space DC alone does not justify a major re-rating.

Tower Semiconductor (TSEM)

Tower Semiconductor fabricates SiPh PICs that integrate modulators, detectors, and potentially control electronics onto a single die. The same logic applies: total PICs deployed in space roughly map to the photonic integration content that would have existed in terrestrial transceivers. TSEM's PH18 platform wins the space socket on radiation tolerance and heritage, but the volume is not incremental — it replaces terrestrial PIC volume migrating to orbit.

Tower does benefit from one structural change: in space, the PIC may absorb the PAT detector array and some control logic that would have been separate chips on Earth, increasing die complexity and value per PIC. Tower also benefits from the disappearance of coherent DSP complexity — a simpler SiPh PIC (direct-detection, no complex modulator bias control) is cheaper to manufacture, improving yield and margin even if revenue per PIC is lower.

Inter-Orbit Communication: The One Truly New Market

Inter-orbit communication is the only category of optical link in the Space DC with zero terrestrial equivalent, generating purely incremental demand for the photonic supply chain.

In a terrestrial data center, the edge, training, and storage tiers are co-located, connected by fiber over tens of meters to a few kilometers. In the Space DC, these tiers are at different orbital altitudes, separated by hundreds of kilometers of vertical distance.

The architecture uses Starlink satellites as relay nodes. A VLEO Space DC satellite at 500 km does not shoot a laser directly at a MEO storage satellite at 2,000 km — the slant distance would be enormous and the pointing challenge severe. Instead, the VLEO satellite links to a nearby Starlink satellite, which relays traffic through the Starlink mesh to a Starlink satellite near the MEO target, which then links to the MEO Space DC satellite. Each individual hop stays under approximately 1,000 km.

This architecture creates demand for additional optical terminals on Starlink relay satellites that did not previously exist. Current Starlink V2 Minis have 3 laser terminals at ~200 Gbps, sized for broadband internet traffic. If Starlink satellites must also relay Space DC inter-orbit traffic — gradient updates, model weights, checkpoint data, inference routing — the bandwidth per terminal and number of terminals per satellite must increase substantially. A next-generation Starlink satellite serving as a Space DC relay node might need 6–8 terminals at 1.6T–6.4T each. The inter-orbit component demand for these upgraded relay terminals is purely incremental — additive to Starlink’s existing broadband operations and separate from the Space DC’s intra-cluster links.

Inter-orbit link distances (200–1,000 km per hop) fall in a middle ground between intra-cluster links (20–160 km, easy) and extreme long-range links (2,000–5,000 km, impractical). At 200–500 km, the link budget is tight but closeable with high-power EMLs and sensitive APDs, probably without amplification. At 500–1,000 km, an EDFA or semiconductor optical amplifier is likely needed to close the link at 6.4T bandwidth.

This makes inter-orbit links the one terminal type where every incumbent supplier has non-displaceable content: Lumentum EMLs (high-power variant, multiple lanes), MACOM InP drivers and TIAs (essential at 200 Gbaud per lane), Hamamatsu InGaAs APDs (essential at these distances), potentially Coherent EDFAs (for 500–1,000 km hops), STM PAT ASICs (high-performance variant for meaningful relative angular motion), and precision optics (Coherent).

Estimating total inter-orbit relay terminals is difficult, but the numbers are meaningful. If the Space DC generates 100 Pbps of inter-tier traffic, and each relay terminal handles 6.4 Tbps, approximately 15,000 relay terminals must operate simultaneously — implying 50,000–100,000 dedicated relay terminals deployed across the Starlink mesh (accounting for orbital coverage, redundancy, and duty cycles). At $2,000–10,000 of photonic content per relay terminal, this is a $100 million–$1 billion cumulative market for inter-orbit over the deployment period. Moderate in the context of total Space DC spending, but significant because it is 100% incremental and carries high content per terminal for every supplier.

The Newcomer Thesis: Why Incumbents May Capture Less Growth Than Expected

Here is the key longer-term implication of the volume-shift framework: the companies best positioned to capture the Space DC optical market in 2030–2035 may not be the companies currently supplying space laser communication.

Current space OISL incumbents — Mynaric, Tesat-Spacecom, General Atomics, SA Photonics (now CACI) — build terminals optimized for 100G–400G links for government and defense customers. Their design philosophy prioritizes radiation hardness, reliability, and link closure at extreme ranges rather than raw bandwidth density and cost per bit. A Mynaric CONDOR Mk3 terminal delivers ~100 Gbps in a 30 kg package costing approximately $500,000. This is the wrong product for the Space DC, which needs 6.4T–12.8T in a sub-5 kg package at sub-$5,000 cost.

The terrestrial AI data center, meanwhile, is already solving exactly this problem: ultra-high-bandwidth, low-power, low-cost optical interconnect between compute nodes. The companies driving this on Earth are building co-packaged optics, photonic chiplets, and integrated optical I/O that deliver 12.8T–51.2T of bandwidth in centimeter-scale packages at hundreds of dollars, not hundreds of thousands.

Consider the trajectory. Broadcom's Tomahawk 5 switch ASIC drives 51.2 Tbps of optical I/O through co-packaged optics. Nvidia's next-generation GPU interconnect roadmap assumes optical I/O directly attached to the compute die. Companies like Ayar Labs (optical I/O chiplets), Celestial AI (photonic fabric for AI interconnect), Lightmatter (photonic interconnect), and Ranovus (co-packaged optics) are building photonic interconnect specifically optimized for the bandwidth density, power efficiency, and cost targets that AI training clusters require.

If compute moves from terrestrial racks to orbital satellites, these companies' technology moves with it. The fundamental design challenge is the same: connect thousands of high-TOPS compute nodes with multi-terabit, low-latency optical links at minimal power. The only additional requirement in space is radiation tolerance and the free-space optics/PAT system — which can be added as a separate module at the terminal front end, leaving the high-speed photonic engine largely unchanged.

This suggests that the Space DC optical terminal of 2032 will not be designed by a traditional space laser company scaling up from 100G, but by a terrestrial photonic interconnect company adding space qualification to an existing multi-terabit platform. The architectural DNA flows from data center to space, not from space to data center.

For investors, this means the traditional OISL supply chain — Mynaric, Tesat, the defense primes — captures the early deployment phase (2026–2029) when volumes are low and heritage matters. But as production scales to hundreds of thousands of terminals per year in 2030–2035, the market transitions to companies with high-volume photonic integration expertise. These newcomers bring two decisive advantages: they have already solved the multi-terabit-per-terminal bandwidth density problem for terrestrial AI, and they manufacture at data-center-class volumes with data-center-class cost structures.

The specific companies most likely to capture this transition are difficult to identify with certainty because many are private (Ayar Labs, Celestial AI, Lightmatter) and their space strategies are undisclosed. Among public companies, Broadcom and Marvell have the co-packaged optics and photonic integration capabilities to enter this market if volumes justify their return thresholds, but neither has signaled space-specific intent. Nvidia, through its investments in Lumentum and its co-packaged optics roadmap, is positioning to control the full stack from compute silicon to optical interconnect — and if Nvidia builds the compute ASIC for the Space DC satellites, it may pull its optical I/O partners into the terminal design as a bundled solution.

The critical point for portfolio construction: incumbents' near-term heritage advantage should not be extrapolated forward as a durable moat. The technology direction and volume economics strongly favor newcomers from the terrestrial AI interconnect ecosystem. This limits the terminal value of pure space laser communication plays and increases the importance of identifying which terrestrial photonic interconnect companies will make the pivot to space.

Company-Level Summary

MACOM (MTSI) — Top pick.

The only company in this analysis that benefits from all three dynamics simultaneously: near-term heritage advantage from existing space analog programs, superlinear content growth as per-link bandwidth scales to 6.4T–12.8T, and relevance across both intra-cluster and inter-orbit terminals regardless of who designs the terminal. Every terminal needs III-V analog components at 200 Gbaud per lane, and MACOM, with its vertically integrated InP fab in Lowell, is one of only two or three companies globally that can supply them. At 6.4T per link, MACOM content per terminal is $130–320; at 12.8T, it is $260–640. Even in a volume-neutral shift from terrestrial to space, this content growth is transformative for earnings.

STMicroelectronics (STM) — Hold, reframed as PAT and power management play.

Durable Space DC content is the PAT control ASIC and satellite power management — both genuinely incremental demand categories with no terrestrial analog. These are immune to photonic integration trends and present across every terminal on every satellite. STM's analog driver and TIA revenue migrates toward MACOM at higher speeds, but the PAT socket is permanent. Size the position for the incremental PAT and power content, not the total photonic BOM.

Coherent (COHR) — Hold. Free-space optics and inter-orbit amplification.

Genuinely incremental content is precision optics (telescopes, coatings, filters) across all terminal types and EDFAs for inter-orbit relay terminals where link distances reach 500–1,000 km. Core photonic content (transmit lasers) is volume-shifted, not volume-added. Coherent benefits from being irreplaceable in the free-space, precision optics hardware layer, but this is a lower-margin product line than EDFAs. Moderate position.

Hamamatsu (6965.T) — Buy. Narrower addressable market, high content per terminal.

InGaAs APDs are essential for inter-orbit relay terminals and the more demanding intra-cluster links at 6.4T+ bandwidth. At 8–16 detector channels per terminal, content per unit is meaningful. But addressable terminal count is the smallest of any company in this analysis. Size accordingly.

Lumentum (LITE) — Optionality.

The EML is structurally protected because silicon cannot emit light, and Lumentum will supply the laser source for virtually every terminal. But volume is largely shifted from terrestrial transceiver demand, not incremental. Benefits from space pricing premiums and inter-orbit terminal demand.

Tower Semiconductor (TSEM) — Buy. SiPh integration platform.

Tower's SiPh PICs are the integration platform for intra-cluster terminals, but PIC volume is largely shifted from terrestrial demand. Benefits from higher value per PIC (integration of PAT detector arrays and control logic) and cost-reduction pressure at millions of terminals driving SiPh adoption over discrete components.

Corning (GLW) — Structural loser.

Fiber demand declines as compute migrates from terrestrial data centers (fiber-intensive) to orbital data centers (zero fiber). The decline is gradual — terrestrial data centers do not disappear overnight — but the directional impact on Corning's highest-growth segment (hyperscale data center fiber) is unambiguously negative. Not a short, because terrestrial fiber demand has many other drivers (5G densification, FTTH), but the long-term narrative of exponential fiber demand driven by AI weakens materially if meaningful compute moves to orbit.

Newcomers from terrestrial AI interconnect — The biggest unknown.

Ayar Labs, Celestial AI, Lightmatter, Ranovus, and the co-packaged optics divisions of Broadcom and Marvell are the most likely to capture the Space DC terminal market in 2030–2035. Most are private and not directly investable today.

Investment Framework Summary

MACOM stands out as our clear top pick, uniquely combining near-term heritage with superlinear content growth as per-link bandwidth scales to multi-terabit levels. Hamamatsu and Tower Semiconductor are buys for their high-content positions, while STMicroelectronics, Coherent, and Lumentum are holds. Corning faces a structural long-term headwind as fiber demand migrates to orbit.

That said, these are narrow, Space DC-specific calls only — it is inherently difficult to issue outright buy/hold ratings on any of these names without considering full-company valuations and their fast-evolving terrestrial photonics and CPO growth outlooks.

This is marketing material for NewDeal Invest. The investment style is aimed at investors with a high risk profile and a long-term investment horizon, and past performance is no guarantee of future returns. The stocks that NewDeal Invest invests in are more volatile than the broader stock market, and therefore fluctuate more both up and down. NewDeal Invest does not provide investment advice, and all investment decisions are made by you yourself. No information from NewDeal Invest should be considered a recommendation. Short-term investments, with a time horizon of 5 years or less, are discouraged both in NewDeal Invest, PMINDI, and in the stocks mentioned. See relevant risk factors at https://newdealinvest.dk/ and https://newdealinvest.dk/pmindi/