AI Infrastructure & Data Centers

Global – October 16, 2026 -- The Gazelle optical computing evaluation board performs 8×2 fundamental optical matrix computations and is described by its maker as the highest-precision optoelectronic computing device known to date.

Gazelle pairs a programmable optical matrix with an FPGA and a ready-made software stack

The research platform is built on optical principles. An onboard FPGA and an out-of-the-box software stack are meant to let developers start work without building their own tooling. The optical matrix is programmable, and the board supports a range of commonly used debugging interfaces.

Three target workloads span AI inference, scientific computing and combinatorial optimization

For neural networks, Gazelle targets inference of models used in image recognition and object detection. For scientific research, it is positioned to solve scientific computing problems more efficiently than classical paradigms.

Global – October 16, 2026 -- Lightelligence's PACE2 accelerator card combines a 128×128 optical matrix multiplier with more than 40,000 photonic devices, and the company rates both its optical and electronic cores at 32 TOPs peak performance.

Hybrid architecture pairs a photonic core with an electronic core

The card, released in 2025, uses a heterogeneous design. An optical tensor processing unit (PIC core) works alongside an electronic tensor acceleration unit and an optical computing control unit (EIC core). The chip runs at a 1GHz clock speed, and PIC latency is 200ps.

Lightelligence describes the 128×128 matrix as the largest optical computing matrix it has built. Matrix coefficients can be configured dynamically through APIs, so the card is programmable.

Global – October 16, 2026 -- Lightelligence, founded in 2017, reports 446 cumulative patents, with invention patents accounting for 80% of the portfolio, as it markets three optoelectronic computing products to AI and data center customers.

Lightelligence staffs 250+ employees, 170+ of them in R&D

The company counts more than 250 employees worldwide. R&D staff make up over 170 of them, roughly two-thirds of the workforce, which fits a patent-heavy, research-led business model.

PACE 2 accelerator card launched in 2025

PACE 2 is an optoelectronic accelerated computing card released by Lightelligence in 2025. The company lists high performance, programmability, versatility and advanced packaging technology as its key features.

San Antonio – October 16, 2026 -- Singtel Digital InfraCo has received Frost & Sullivan's 2026 Southeast Asia Transformational Innovation Leadership Recognition in AI infrastructure, backed by a portfolio of 26 internationally granted patents and early deployments of four successive NVIDIA GPU platforms.

Frost & Sullivan benchmarks strategy effectiveness and execution

The analyst firm assessed the company on strategy effectiveness and strategy execution. It credited Singtel Digital InfraCo with a unified platform combining connectivity, Nxera data centre resources, high-performance GPU capacity and RE:AI's AI services.

"Rather than approaching AI infrastructure as a collection of individual technology layers," the company built an integrated operating model, said Gautham Gnanajothi, Senior Vice President at Frost & Sullivan.

San Jose – October 16, 2026 -- LITEON Technology (2301.tw) demonstrated 800 VDC rack power architectures scaling from 330kW to 1.2MW at the 2026 OCP Global Summit, targeting the megawatt-scale AI data centers its partner NVIDIA is building around.

LITEON exhibits 800 VDC Power Rack and 800V-to-50V Power Brick in NVIDIA MGX area

The company was invited to the NVIDIA MGX Ecosystem Ceremony at the summit. It was represented by John Chang, General Manager of its Cloud Infrastructure Platform and Solution SBG; Bob Weng, General Manager of the AIDC Platform SBU; and Simon Ong, President of LITEON USA.

In the MGX exhibit area, LITEON showed its NVIDIA 800 VDC Power Rack and 800V-to-50V Power Brick. The products span rack-level power architecture to power conversion, and are designed for the efficiency, power density and scalability requirements of AI data centers.

San Jose – October 16, 2026 -- Midea's Clivet brand has unveiled a magnetic-bearing centrifugal compressor that cuts compressor power by up to 40% at the same cooling capacity, targeting GW-scale AI data centers, at the 2026 OCP Global Summit.

Clivet presented the compressor and a full-chain liquid cooling solution at the "GW-Scale Open AIDC" forum. Dr. Yunxiao Ding, Director of the Research Institute at Midea Building Technologies, delivered the presentation, and Clivet received the "2026 GW-Scale Open AIDC Excellent Technical Presentation Award."

Rising cold-plate temperatures of 45°C create three engineering problems

As cold-plate supply temperatures rise toward 45°C, free cooling opportunities expand, while water constraints favor dry heat rejection. Prioritizing free cooling with mechanical-cooling backup raises three challenges:

London – October 16, 2026 -- The global data center precision air conditioning market will grow from USD 4.88 billion in 2025 to USD 10.00 billion by 2035, a 7.43% CAGR, according to a new DC Market Insights report. Rising AI rack density is shifting spending toward fan walls, in-row coolers, rear-door heat exchangers and retrofit services.

New-build capacity of 12.0 GW a year supports USD 2.75 billion in equipment spending

New data center capacity of about 12.0 GW a year supports roughly USD 2.75 billion of new-build precision air-conditioning spending, the report states. The analysis covers four product types: perimeter CRAC and CRAH units, in-row and rack-based coolers, fan walls and thermal walls, and rear-door heat exchangers.

London – October 16, 2026 -- The global DDR5 AI server memory market is forecast to grow from USD 2.14 billion in 2025 to USD 6.80 billion by 2035, a 12.25% CAGR, according to a new DC Market Insights report. Servers configured with more than 2 TB of DDR5 host memory are projected to account for USD 4.42 billion of that market by 2035.

AI capacity additions are modeled to rise more than fourfold

The report models annual AI capacity additions rising from 4.2 GW to 17.8 GW, expanding the installed base of memory-intensive accelerator servers. AI servers are adding more host memory per node. Accelerator-rich systems raise the value of capacity, bandwidth and memory-channel count.

New York – October 15, 2026 -- Mindbeam raised the Level 2 KernelBench score of NVIDIA's Nemotron 3.5 Super VL from 22.8% to 82.1% through post-training, ahead of the 57% accuracy the company reports for GPT 5.6 Sol.

The New York-based AI infrastructure company announced a collaboration with NVIDIA covering two layers of the open source AI stack: a post-trained model that writes CUDA kernels, and an inference engine that serves NVIDIA's latest open models.

Litespark Nemotron 3.5 Super VL gains 11.2 points on Level 1 tasks

Mindbeam post-trained NVIDIA's Nemotron 3.5 Super VL (120B-A12B) to produce Litespark Nemotron 3.5 Super VL, an open-weights model for CUDA code generation. The task is to take a PyTorch module and generate a raw CUDA kernel that reproduces the forward pass within tight numerical tolerances while staying architecture-agnostic.

Portland – October 15, 2026 -- Mueon Corporation has raised $65 million in a Series A round led by Maverick Silicon, lifting its total capital raised to more than $80 million as it moves its CubeletsTM modular compute architecture toward production readiness.

Intel Capital, Geodesic Alliances Fund, Oregon Venture Fund, M-Ventures, Wellington Capital, IQT and Defined also participated. The round follows a $15.5 million seed financing in 2025, when the Portland-based semiconductor systems company emerged from stealth.

Mueon targets up to 10X power and cost efficiency in the same rack footprint

The company says its architecture is designed to scale toward megawatt-class density as demand grows. CubeletsTM are modular, three-dimensional compute-and-memory building blocks. They shorten the physical distance data travels, which cuts the power cost of moving it, and avoid the sprawl of a traditional server rack.