AI Infrastructure

Cambridge, Mass. – September 04, 2026 -- Physical Superintelligence (PSI) launched with $58 million in seed funding to industrialize physics discovery at machine scale, starting with a founding role in the first privately funded interstellar space mission and commercial deployment of a data center optimization platform.

Breakthrough Energy Ventures leads $58 million seed round for AI-native physics lab

The round was led by Breakthrough Energy Ventures and joined by Dragon Global, Robot Ventures, Solari, Susa, Ron Conway's SV Angel, Valkyrie, Balaji Srinivasan, and Anthony Scaramucci, alongside individual investors from OpenAI, NVIDIA, SoftBank Energy, Oracle, Hugging Face, JUMP Capital, and the a16z Scout Fund. PSI was founded by Matt Pines, Alex Klokus, and Dr. Alexander Wissner-Gross to build higher-fidelity world models and engineer physical systems beyond human design capacity.

Vienna, Austria – September 04, 2026 -- Rankscale GmbH has launched a new analytics module that tracks paid ad placements inside ChatGPT, as OpenAI rolls out ChatGPT Ads across 31 European markets including Germany, France, Spain, Italy, Poland and Austria.

ChatGPT Ads expansion creates a new competitive layer for European advertisers

The rollout gives advertisers a new channel within AI conversations, but also a challenge: understanding when, where and in what context their ads appear. Rankscale, an Austrian AI-visibility platform, extends its existing organic-visibility technology to systematically analyze how ChatGPT Ads are served to users.

SAN JOSE, Calif. – September 04, 2026 -- Mixx Technologies, Inc. has launched the SxCTM Connector, an optical connector capable of terminating up to 24,576 fibers in a single rack unit, a density the company says is four times greater than MMC-VSFF, the highest-density connector platform currently qualified for deployment.

Connector density becomes the bottleneck for co-packaged optics deployment

Mixx, a venture-backed deep-tech firm building co-packaged optics (CPO) for hyperscale AI infrastructure, said existing connector families cannot support the fiber density that next-generation inference workloads require. A 400T switch brings thousands of fibers out of a single box, exceeding what pluggable transceivers can support at the front panel. Mixture-of-experts and agentic AI workloads demand high radix — the number of endpoints a single node can reach — to minimize latency and reduce switching layers, transceivers, and cables.

Dallas – September 04, 2026 -- Cango Inc. (NYSE: CANG) reported a net loss of US$81.6 million for the second quarter of 2026, driven primarily by non-cash impairment and disposal charges on its Bitcoin mining fleet, as the company accelerates a pivot toward AI compute infrastructure.

Revenue halves as company right-sizes mining fleet

Total revenue fell approximately 50% sequentially to US$50.8 million, with US$47.4 million derived from Bitcoin mining and US$3.4 million from other sources. The decline stemmed from Cango's decision to phase out older S19-series mining machines and shift some capacity to a hosted leasing model, a move management said improved operating costs and cash flow despite the top-line hit.

Riyadh – September 03, 2026 -- AMD Instinct MI355X GPU-based AI infrastructure, networked on Cisco Silicon One technology, is now live in Saudi Arabia and serving HUMAIN customers in the Kingdom and beyond.

Production system pairs AMD Instinct GPUs with Cisco's AI-optimized network fabric

The deployment combines AMD Instinct MI355X GPUs and AMD EPYC CPUs with Cisco's N9000 Series platform, built on Cisco Silicon One and 800G optics. The fabric is designed for scale, low latency and operational resilience, allowing HUMAIN to offer GPU-as-a-service for both model training and inferencing.

Companies plan up to 250 MW of new capacity starting in 2027

AMD, Cisco and HUMAIN plan to add up to 250 MW of AI infrastructure powered by AMD Instinct MI400 Series GPUs, AMD EPYC CPUs and AMD ROCm open software, alongside Cisco networking. Deployment is scheduled to begin in 2027, with capacity coming online in the second half of that year.

ARLINGTON, VA. -- September 01, 2026 -- Corvex, Inc. (Nasdaq: MOVE) said it will lift critical IT power capacity fivefold, from roughly 1.5 megawatts to approximately 8 megawatts by year-end 2026, backed by a $33 million private placement led by Goldman Sachs & Co. LLC, Morgan Stanley and Oppenheimer & Co. The Arlington, Virginia-based AI infrastructure provider is doubling its existing Mid-Atlantic data center and standing up a second facility in the Midwest, with both sites expected ready for service in the fourth quarter of 2026. A right of first refusal on 12.5 additional megawatts at the Midwest site could push total capacity past 20 megawatts, a 13-fold jump from current levels.

REHOVOT, Israel -- September 01, 2026 -- CollPlant Biotechnologies Ltd. (Nasdaq: CLGN) has signed a definitive agreement to acquire LightSolver Ltd., an Israeli developer of laser-based computing hardware, in a deal expected to close this week. The transaction moves CollPlant, a regenerative-medicine company known for recombinant collagen products, into supercomputing and photonic computing. LightSolver's technology has already been tested by Boeing, a top-tier global financial institution and unnamed defense, energy and engineering-software companies.

RIYADH, Saudi Arabia -- August 31, 2026 -- Data Dynamics announced general availability of Enterprise 2.0.5 at LEAP 2026 in Riyadh. The company said 60% of AI projects stall due to data curation and AI-readiness gaps, prompting the release of a platform that enforces governance directly at the data layer. Enterprise 2.0.5 targets models pointed at unmapped data, agents with standing access to content, and compliance teams left proving control after incidents occur. The software is available now across cloud, on-premises, sovereign and air-gapped deployments.

NINGBO, China -- September 01, 2026 -- Ningbo Longwell Electric Technology Co., Ltd. (LONGWELL) said its LWBE3G plug-fan platform cut CRAH fan energy consumption by 38% in a data center cooling deployment completed with a top-3 global precision cooling OEM. The Chinese EC fan and motor manufacturer released its AI-Era Data Center Cooling Pack on Aug. 31, 2026. LONGWELL delivered 12 engineering samples in 35 days and moved from specification lock to mass production in 90 days, compared with a 14-month development period plus a six-month ramp quoted by an incumbent European supplier for the same program.

SHERIDAN, WYOMING -- August 28, 2026 -- SuperX AI Technology Limited (NASDAQ: SUPX) said its subsidiary SuperX Microinference Pte Ltd signed a long-term equipment and services supply agreement with Australian compute provider Ezisight Australia Pty Ltd., trading as Ultimate AI Datacentre, and received an initial purchase order for 128 units of B300 AI server clusters. The Singapore-based company will deliver the servers in the fourth quarter of 2026 for deployment in Ezisight's Australian data centers. The order marks SuperX's first commercial entry into the Australian market.

Ezisight Deploys Servers For Local AI Training And Inference

Ezisight will install the 128 B300 units to expand its domestic GPU compute pool, supporting local large-model training and inference workloads. SuperX Microinference will handle supply chain coordination and delivery under the agreement.

SuperX did not disclose the order's dollar value.