Fractile’s $720M Round, CoreWeave’s Prepayments, and AlphaGenome 2

Artificial intelligence infrastructure and scientific modeling recorded three significant milestones this week as British chipmaker Fractile secured a massive late-stage venture capital round, neocloud provider CoreWeave restructured its capital strategy around upfront customer deposits, and Google DeepMind introduced its next-generation single-cell DNA sequence model.

Fractile Raises $720 Million for Stacked-DRAM Inference Chips

British artificial intelligence hardware startup Fractile has raised a $720 million Series C round backed by Britain’s Sovereign AI Fund, with the British government stating the investment aims at “ensuring the U.K. has a greater stake in the hardware that will underpin its future economy.” Alongside semiconductor startups Axelera, Etched, and OLIX, the financing positions Fractile among the top-funded private companies building purpose-built chips to reduce the cost of running demanding production inference workloads.

The Series C round was led by Thrive Capital, Founders Fund, Lightspeed Venture Partners, and Redpoint. According to the British government, the financing valued Fractile at “over £5 billion,” a figure that both the Financial Times and a Reuters source reported as potentially reaching $7 billion. Beyond custom silicon, Fractile is pursuing a full-stack architecture that pairs a machine learning compiler with its hardware to maximize execution speed for large language models.

Following an architectural pivot, Fractile’s hardware design centers on stacking DRAM directly on top of each chip to bypass the conventional bottleneck between memory bandwidth and capacity—a constraint that has grown more acute as agentic workflows execute longer chains of tasks across large models. In a LinkedIn post, Fractile chief executive Walter Goodwin noted that “there is still a lot of work ahead” and stated that the capital will enable the company to ship its first chips and systems next year.

CoreWeave Pivots Toward $9 Billion in Customer Prepayments

As demand for compute capacity continues to climb, neocloud operator CoreWeave reported $3.66 billion in third-quarter revenue on Tuesday, up 168% from the prior year, while posting a $709 million net loss and $846 million in quarterly interest expense. With bond yields on some of its debt climbing above 15% last month, chief executive Michael Intrator told analysts during Tuesday’s earnings call, “We will not use the unsecured debt market to fund our business at current yields, nor do we need to.”

Instead, CoreWeave is financing its $132.5 billion contract backlog through loans tied to specific customer contracts, upfront deposits, and shorter-term agreements that command higher rates. The company expects about $9 billion in customer prepayments and deposits over the next three months, including more than $3 billion linked to new customer contracts for capacity already built that requires no additional spending on hardware. Rivals Nebius and IREN have similarly utilized Microsoft prepayments to fund buildouts.

CoreWeave also raised prices across its offerings by about 25% in July and by another 10% later in the quarter. Intrator noted that CoreWeave is selling short-term capacity at more than $40 million per megawatt annually, while pricing for new infrastructure on four- and five-year contracts is moving to $20 million to $25 million per megawatt. In August, the company closed a $2.6 billion delayed-draw term loan known as DDTL 5.5 designed specifically for shorter-term customer contracts.

DeepMind Unveils AlphaGenome 2 for Single-Cell Genetics

In computational biology, Google DeepMind researchers Adam Gayoso, Natasha Latysheva, Pushmeet Kohli, and Žiga Avsec announced AlphaGenome 2, a conditional DNA sequence model that predicts how genetic variations affect specific human cell types and states rather than relying on bulk tissue averages. Building on last year’s original model and the AlphaGenome Atlas covering 9 billion possible single-letter changes, AlphaGenome 2 was trained on bulk and single-cell data from human and mouse samples encompassing more than 100 million—and specifically over 107 million—individual cell profiles across 68 tissues.

Alongside a 1-million-letter DNA sequence, AlphaGenome 2 takes a target biological condition at every stage of its architecture, predicting gene expression, histone modifications, chromatin accessibility, and DNA methylation. On the TenK10K dataset spanning 28 primary immune cell types, the model identified nearly twice as many causal variants (30% vs. 16%) and directional gene effects (43% vs. 23%) as fine-tuned predecessors, and computationally reproduced how a regulatory switch drives inflammatory bowel disease by overactivating the ETS2 gene in gut macrophages.

As reported by The Verge, the model can also compress a single-cell RNA-seq snapshot from a few dozen cells to analyze unseen states without retraining—distinguishing COVID-19 mRNA vaccine-activated immune T cells from resting cells—and extends enhancer-gene mapping to primary human cells such as brain astrocytes while collaborating with the Wellcome Sanger Institute. AlphaGenome 2 is available for non-commercial research via the AlphaGenome API and in private preview for commercial research, with AlphaGenome 2 Skills slated for release in Google Antigravity alongside Gemini 4.