Could be a great market opportunity for AEVA!
Marvell acquired Celestial AI, whose Photonic Fabric previously relied on external laser sources from POET Technologies. After the acquisition, Marvell cancelled the remaining POET orders.
So what replaced POET?
Just a few months later, Aeva announced a development agreement with an unnamed “leading optical engine provider” to develop high-power laser sources for AI data centers.
No proof, but the pieces seem to fit surprisingly well.
720k shares, about 17M $ today.
What Exactly Happened Today — The Verified Facts
Aeva launched an Optical Connectivity business, expanding its proprietary high-power optical source and silicon photonics technology to next-generation AI data centers. A leading provider of high-speed optical engines for AI data centers signed a joint development agreement to integrate Aeva’s high-power optical sources into a new module for deployment by a major hyperscaler, with initial deployment targeted for H2 2027 and production ramp targeted for 2028.
Aeva’s SOA demonstrates optical output power exceeding 28 dBm while achieving wall-plug efficiencies greater than 20% at temperatures up to 50°C — a combination of high power, efficiency, and thermal robustness that drives lower system power dissipation, reduced costs, and significantly improves reliability in large-scale AI data center deployments.
Q2 2026 financial results: Revenue $6.1M (slightly below Q1’s $6.3M but within guidance trajectory). Full year 2026 guidance of $30-36M maintained. CFO transition announced — Saurabh Sinha transitioning, new CFO joining with 20+ years finance experience.
Why This Is Architecturally Different From Everything Else Aeva Has Announced
This is not an incremental win. Let me explain precisely why, because most retail investors will miss the magnitude of what happened today.
The core insight: Aeva spent a decade solving one of the hardest problems in physics — making coherent light sources that are simultaneously high-power, low-noise, thermally stable, and manufacturable at automotive-grade reliability. They did this because FMCW lidar requires it. No one else was willing to invest this deeply in photonics infrastructure because the automotive lidar market was too small and too uncertain to justify it.
The result: Aeva accidentally built what may be the world’s most capable high-power silicon photonics platform — and they built it to a reliability standard (automotive grade) that far exceeds what data center applications require.
Now the data center industry has a desperate, urgent need for exactly this technology. The analogy: Aeva is like a company that spent a decade building the world’s most reliable industrial pump for deep-sea oil drilling — and then discovered that the same pump, with minimal modification, is exactly what every hospital in the world needs for a new critical medical procedure. The R&D is already paid for. The manufacturing is already industrialized. The marginal cost of entering the new market is a fraction of what it would cost anyone else to start from scratch.
The NPO/CPO Market — Why It’s Enormous
The near-packaged optics market was valued at $3.8 billion in 2025 and is projected to reach $18.6 billion by 2034, growing at a CAGR of 19.3%.
The global co-packaged optics market is projected to grow from $328.6 million in 2026 to $3.374 billion by 2034, exhibiting a CAGR of 33.8%.
Hyperscaler firms aim to deploy clusters with over 1 million GPUs by 2027. These clusters demand extremely fast and low-latency interconnections, creating an absolute necessity for co-packaged optics technology.
The combined NPO + CPO market reaches approximately $22 billion by 2034. The laser source component — Aeva’s specific entry point — represents roughly 20–30% of total module value. That’s a $4–6 billion laser source TAM within the broader NPO/CPO market.
The Competitive Landscape — Where Does Aeva Actually Fit?
NVIDIA made strategic investments in Coherent and Lumentum as laser supply chain partners. Ayar Labs joined NVIDIA’s NVLink Fusion ecosystem. Marvell completed its Celestial AI acquisition. The key competitors: Broadcom, Intel, NVIDIA, Cisco, Lumentum, Coherent, Ayar Labs, Marvell.
These are $10B–$500B companies. Why would a hyperscaler work with $1B market cap Aeva instead?
The answer is specific and technical: Aeva’s SOA achieves optical output power exceeding 28 dBm at wall-plug efficiencies greater than 20% at temperatures up to 50°C. This performance profile — particularly the thermal efficiency at high temperatures — is precisely what data centers need as they struggle with power density and cooling constraints. Large incumbents (Lumentum, Coherent) make excellent lasers but haven’t been forced to optimize for automotive-grade thermal robustness. Aeva has.
Large-scale NVIDIA CPO production could slip to 2028–2029 on systems-engineering grounds — serviceability, reliability, and manufacturing-test yield — elevating near-package optics (NPO) as a pragmatic intermediate. This timing actually helps Aeva — NPO is the near-term transition architecture, and that’s exactly where Aeva signed its first customer agreement.
Optical provides greatly increases connectivity + processing speed but equally important optical deceases the three community hot spot issues around present copper AI data centers being increased noise, increased electricity and increase water needs. Very exciting!
