The Race to Secure Data

Why It Matters

Ensuring data remains encrypted while being processed removes a critical vulnerability for AI‑driven workloads and helps enterprises meet looming compliance deadlines, opening new revenue streams for hardware vendors.

Key Takeaways

• Broadcom's SecureHBA adds PQC encryption to 64‑Gb/s Fibre Channel adapters
• 128‑Gb/s silicon taped out for 2027, targeting 1‑2 M ports by 2027
• Niobium's FPGA‑based FHE chip promises up to 50× speedup for encrypted AI
• Optalysys uses photonic chiplets to accelerate FHE, claiming 100× speed gains
• US/EU regulations mandate full data‑in‑flight protection by 2030

Pulse Analysis

The push for end‑to‑end data security is reshaping the datacentre hardware landscape. Traditional at‑rest encryption no longer suffices as generative AI models, like Anthropic’s Mythos, can rapidly uncover hidden vulnerabilities. Chip manufacturers are responding with integrated solutions that protect data in transit. Broadcom’s SecureHBA family embeds post‑quantum cryptographic (PQC) algorithms directly into 64‑Gb/s Fibre Channel adapters, and its upcoming 128‑Gb/s silicon promises line‑rate encryption for the next generation of servers. With an estimated 1‑2 million ports slated for deployment by 2027, the company is positioning itself as the de‑facto standard for secure storage networking, even sharing the technology with rivals to accelerate ecosystem adoption.

Beyond transport‑level protection, the industry is tackling the harder problem of securing data while it is being processed. Fully homomorphic encryption (FHE) allows computations on encrypted data without ever exposing the plaintext, a breakthrough for sensitive AI workloads in finance, healthcare, and government. Niobium Microsystems leverages an FPGA‑based prototype that can accelerate FHE operations up to 50× faster than software‑only approaches, while Optalysys takes a photonic route, using laser‑based chiplets to achieve claimed 100× speedups. Both firms aim to deliver drop‑in accelerators that sit alongside CPUs or AI ASICs, enabling enterprises to run encrypted inference and training without compromising performance.

Regulatory pressure is a key catalyst: the US and EU have announced mandatory in‑flight encryption requirements by 2030. This creates a clear market incentive for vendors that can deliver line‑rate, low‑latency encryption and FHE capabilities. As AI adoption accelerates, the ability to keep data encrypted from storage through network transport to compute will become a competitive differentiator, driving investment in next‑gen silicon, photonic accelerators, and standardized security stacks across the cloud and on‑premise environments.