AI Infrastructure

Mindbeam Lifts Nemotron CUDA Pass Rate to 82.1% on KernelBench L2

Image
Mindbeam Lifts Nemotron CUDA Pass Rate to 82.1% on KernelBench L2

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.

Mueon Raises $65M Series A to Push 3D Modular AI Compute

Image
Mueon Raises $65M Series A to Push 3D Modular AI Compute

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.

Colt DCS, RMZ Start Building 64MW Mumbai 3 AI Data Centre

Image
Colt DCS, RMZ Start Building 64MW Mumbai 3 AI Data Centre

Mumbai – October 15, 2026 -- Colt Data Centre Services (Colt DCS) and RMZ have begun construction of Mumbai 3, a 64MW AI-ready data centre at their Navi Mumbai Campus, after signing a first hyperscale customer and fully pre-letting the facility.

The project is part of the companies' joint venture (JV), announced in 2024, to invest $1.7 billion in the Indian data centre market. Ground preparation began in December 2025. The companies said the JV has reached this milestone within 18 months of signing.

Mumbai 3 lifts Navi Mumbai Campus capacity to 86MW

Mumbai 3 will join Mumbai 1 and Mumbai 2, which currently support large enterprise and cloud workloads. Upon completion, it will raise the campus's operational capacity to 86MW. The campus has the potential to reach up to 148MW at full development.

Both partners have committed to develop 250MW of data centre capacity across Mumbai and Chennai under the JV.

Local LLM Cost per Right Answer Jumps Up to 64x on Hard Questions

Image
Local LLM Cost per Right Answer Jumps Up to 64x on Hard Questions

Southern California – October 14, 2026 -- The cheapest local language model for easy questions costs 37 times more per correct answer on hard questions, according to power-metered tests of 27 models on a Strix Halo system. The model, qwen3-30b-a3b-instruct, delivers right answers at 3.6 cents per thousand on easy work. On hard questions the price rises to $1.35 per thousand. Two other models in the 11-model comparison only doubled in cost over the same jump.

Difficulty tax ranges from 2x to 64x across 11 models

The tester ran 11 of the 27 models on both an easy suite and hard questions, and calls the cost ratio between the two the "difficulty tax." It runs from 2x to 64x. The easy suite has 190 checkable tasks across seven categories, including a 20-step stack-and-total exercise. The hard set is the first 100 of 198 GPQA Diamond questions, which are expert-written physics, chemistry and biology problems.

LITEON Unveils 800 VDC Power and 2.6MW CDU for NVIDIA AI Racks

Image
LITEON Unveils 800 VDC Power and 2.6MW CDU for NVIDIA AI Racks

San Jose – October 14, 2026 -- LITEON Technology (2301) is debuting a 2.6MW Enterprise Cooling Distribution Unit (CDU) and a full 800 VDC power stack for AI factories at the OCP Global Summit 2026, running October 12-15 in San Jose, California.

LITEON debuts 2.6MW CDU from investment partner DCX Liquid Cooling Systems

The Enterprise CDU comes from DCX Liquid Cooling Systems, a strategic investment partner of LITEON. It delivers up to 2.6MW of heat transfer capacity and an approach temperature of 3°C. An N+1 redundant pump design is built in for high-density AI and HPC deployments. The unit also carries a next-generation HMI and an integrated software layer for at-a-glance visibility into key system parameters.

LITEON is also showing an 800 VDC Power Shelf integrated with a Battery Backup Unit (BBU).

xFusion Pitches 800V HVDC and 155 kW Liquid Racks at OCP 2026

Image
xFusion Pitches 800V HVDC and 155 kW Liquid Racks at OCP 2026

San Jose – October 14, 2026 -- xFusion is showing a 155 kW liquid-cooled rack, cold plates rated above 4,000 watts per chip and a prototype 800V high-voltage DC power supply for up to 1.2 MW at the 2026 Open Compute Project (OCP) Global Summit, which runs October 12–15 at the San Jose McEnery Convention Center (Booth A33).

The company is targeting power density, which it describes as the main constraint on AI data center scaling. Single-chip power consumption is trending toward 3,000 watts and beyond, while rack-level density is moving from hundreds of kilowatts toward the megawatt class.

Vertiv, Deloitte Launch Workload-Based Guide for Enterprise AI Sites

Image
Vertiv, Deloitte Launch Workload-Based Guide for Enterprise AI Sites

COLUMBUS, Ohio – October 14, 2026 -- Vertiv Holdings Co. (NYSE: VRT) and the Deloitte Center for Integrated Research have introduced a collaboration built around a guide that sorts enterprise AI workloads into six archetypes and maps each to one of four physical infrastructure models.

The Vertiv Designing Infrastructure for Enterprise AI Guide was developed by Vertiv and STL Partners, a digital infrastructure research firm. It is aimed at business, IT and facilities leaders who must decide where AI workloads should run and whether available infrastructure can support them before they commit to an infrastructure plan.

Permitting Becomes Top Risk as DCPI Revenue Hits $14B, Up 27%

Image
Permitting Becomes Top Risk as DCPI Revenue Hits $14B, Up 27%

Redwood City – October 14, 2026 -- The worldwide Data Center Physical Infrastructure (DCPI) market grew 27% year-over-year to $14 billion in manufacturer revenue in the second quarter of 2026, according to Dell'Oro Group. Growth has now run above 20% for consecutive quarters as demand for AI data center capacity outpaces supply. Dell'Oro attributes the modest slowdown from 1Q 2026 to a tougher comparison base, not weaker demand.

The research firm now ranks siting and permitting as the most serious near-term brake on the market. "The quieter and more consequential one is whether it will be permitted," said Alex Cordovil, Research Director at Dell'Oro, contrasting the issue with the louder debate over project cancellations. He said power availability, long-lead equipment and community opposition are stretching project timelines. He added that the timing of permits and grid interconnections is harder to forecast than the demand behind them.

DayOne Tops 1GW Billings as Malaysia Loan Rises to $7.2B

Image
DayOne Tops 1GW Billings as Malaysia Loan Rises to $7.2B

Singapore – October 14, 2026 -- DayOne Data Centers has upsized its senior secured multi-currency syndicated term financing in Malaysia to approximately US$7.2 billion and surpassed 1GW of Billings, the Singapore-headquartered data center platform said.

Seven-lender consortium doubles Johor debt package from May 2025 level

A seven-lender consortium arranged the upsized financing, which supports continued development of DayOne's data center platform in Johor. It builds on the approximately US$3.5 billion syndicated financing the company secured in May 2025.

Since 2024, DayOne has raised over US$6 billion in equity across several financing rounds. It has also drawn project financing from more than 30 local and global banking partners.

Frost & Sullivan Names PDG APAC Customer Value Leader for 2026

San Antonio – October 14, 2026 -- Princeton Digital Group (PDG) has won Frost & Sullivan's 2026 Asia-Pacific Customer Value Leadership Recognition in data centre infrastructure and operations, backed by a portfolio approaching two gigawatts of capacity at the time of assessment.

Frost & Sullivan scored PDG on strategy effectiveness and execution across varied APAC markets

The analyst firm benchmarks candidates on two dimensions: strategy effectiveness and strategy execution. It said PDG performed strongly on both, aligning growth with hyperscale customer demand while executing consistently in markets that differ in infrastructure maturity, regulation, power availability and construction ecosystems.