Volantis raised an $88M Series A co-led by Lachy Groom and Abstract Ventures, with John Doerr, VXI Capital, Triatomic and Susa Ventures participating, to build A-1, an inference system that uses custom integrated micro-VCSEL photonic links to connect compute directly to large memory pools. The company says A-1 is designed to run models above 20T parameters at up to 10,000 tokens per second per user, and it plans to deliver its first integrated inference engines to customers in 2027. Rather than extending chip-to-chip links, the company aims photonics at the compute-to-memory path inside a system, where bandwidth sets the ceiling on inference speed. (PR Newswire)
Private Companies
Inferizeother
Nebius acquired Inferize, a startup founded in January 2026 whose software cuts the idle GPU time that builds up while models load, scale or update weights, and folded its team and technology into Nebius Token Factory. Terms were not disclosed, and the announcement gives no measured utilization gain or named customer. The deal follows Nebius's $643M Eigen AI acquisition, completed in June, and extends its push to own more of the optimization stack behind Token Factory. (Nebius)
Emerging
Photonic prefill: A paper from Arulselvan Madhavan, Peter Carson, Taylor Groves and Thomas Graham, Scaling Inference Prefill with High-Radix Photonic Interconnects, models 3D-integrated photonic interconnects against copper scale-up for mixture-of-experts prefill at 1K to 8K, 128K and 1M-token contexts on a 1,152-GPU deployment. It reports 2.1–3.2x lower latency in high-batch cases and 2.2–4.5x speedups once electrical systems reach their scaling limits. The result gives system builders a way to date when copper scale-up runs out for long-context prefill. (arXiv)
Leased compute: Tencent signed a five-year agreement worth about $7B to lease roughly 100,000 advanced AI chips in Oracle data centers across Southeast Asia, with about 30% paid upfront, the Financial Times reported. Neither company has commented. U.S. rules restrict direct sales of advanced AI chips to China but do not currently bar Chinese firms from renting capacity abroad, so Tencent's largest overseas compute agreement shows Chinese demand moving into foreign cloud backlog instead of chip exports. (Financial Times)