ComputeLabs Research
Edition of · 19 stories
Liquid Compute emerged from stealth on September 15 with an announced $15 million seed financing. FirstMark and Chemistry co-led the round, with participation from K8 Capital, Night Capital, TrueBridge, Brainchild Holdings, UFO Holdings, and angel investor Dmitry Balyasny.
Financial Times reporting relayed through Telegram describes early discussions with investors about a new private financing, not a completed transaction. The target valuation is approximately $1.2 trillion, while the final valuation and fundraising amount could change (“相关谈判仍处早期阶段，最终估值和融资规模可能发生变化”).
Altera Corporation announced that its parent company confidentially submitted a draft registration statement on Form S-1 to the U.S. Securities and Exchange Commission, or SEC. The proposed offering concerns the parent company’s common stock.
The San Diego Supercomputer Center, or SDSC, received a $10 million National Science Foundation award for deployment of Expanse2. This is an award supporting a named supercomputer project, not a report that the system has entered service.
Prime Minister Mark Carney announced expanded immediate expensing for most new capital investment under the Productivity Mega Deduction (“生产力超级抵扣”). The reported C$36 billion over five years is the tax measure’s budgetary cost, not a dedicated AI-infrastructure investment fund.
Micron announced a successful demonstration of a 512GB DDR5 registered dual in-line memory module, or RDIMM, with speeds up to 9,200 megatransfers per second (MT/s). The capacity and transfer-rate figures apply to the server-memory module, not to GPU memory capacity or complete-server performance.
Astera Labs announced three memory-connectivity solutions: Leo X-Series, Leo 2 E-Series, and Leo 2 P-Series. These extend its Leo Smart Memory Controller family across direct fabric-attached GPU memory, CPU-attached expansion, and pooled or shared memory architectures.
SiFive and AMD showcased AMD’s ROCm software running on SiFive’s recently announced BigSky Datacenter Development Platform at the AI Infra Summit. The announcement concerns software execution on a RISC-V datacenter development platform, rather than a newly disclosed accelerator product.
The supplied Telegram updates identify two distinct steps in Meta’s in-house chip roadmap. Arke is scheduled for deployment in the first half of 2027, while next-generation Astrid is scheduled for rollout at the end of 2027.
Axelera AI announced the Europa architecture, positioned as a purpose-built inference accelerator for enterprise and data-center applications. The supplied announcement identifies inference as its intended workload, rather than reporting a training-system launch.
The U.S. Environmental Protection Agency, or EPA, repealed power-plant greenhouse-gas (GHG) rules associated with the Biden administration. Utility Dive distinguishes that repeal from additional standards the agency is moving to eliminate.
The U.S. Energy Information Administration, or EIA, reports that Cheniere Energy completed the Corpus Christi Liquefaction Stage 3 Project in Texas on August 28, 2026. September 15 is the publication date of this update, rather than the project’s completion date.
Utility Dive reports a strategic battery-supply agreement between B2U and Waymo. The agreement concerns older batteries that remain suitable for stationary storage, rather than newly manufactured batteries.
Amazon Web Services, or AWS, reportedly said it remained unable to restore access to affected facilities more than six months after drone-related damage, in its first public update since April (“无人机袭击造成设施损坏六个多月后……这是自 4 月以来的首次公开更新”). The update concerns Bahrain and parts of the United Arab Emirates, or UAE, rather than a stated global AWS disruption.
Rebellions and ai&, described in the announcement as a vertically integrated AI technology company, announced a partnership to deploy RebelRack in Japan. RebelRack is the named infrastructure product, distinct from Rebellions, the company providing it.
Google announced the availability of Agent Substrate on Google Kubernetes Engine (GKE). The open-source agent-execution runtime can run on any Kubernetes infrastructure, while being optimized for GKE.
Google announced the availability of Filestore agent volumes, a fully managed file-storage capability for agent workspaces. It was co-designed with Agent Substrate and complements the compute environments supplied by Agent Substrate on GKE and GKE Agent Sandbox.
The supplied Telegram messages report the release of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking as conversational models. Both are reported available through the Live application programming interface (API) (“Gemini 3.8 Live 和 Gemini 3.8 Live Extended Thinking 现已通过 Live API 开放”).
Salesforce announced Koa, described as its first customer relationship management (CRM) reasoning model. NVIDIA’s account states that the model is built on NVIDIA Nemotron 3 Super.
