Tag: Inference
Together lets coding agents rewrite the forward pass all day
A persistent serving engine turns kernel tuning into a continuous loop, and Together says it beats the best…
A British chip startup reaches a $6.5B value before shipping a wafer
Fractile closed roughly $700M with the UK's sovereign AI fund on the cap table while its in-memory inference…
Baseten adds small open model for fast decisions
Joey-1 returns a calibrated probability instead of prose, costs a cent per million tokens and ships as open…
FuriosaAI banks $620M by selling power as the bottleneck
The South Korean inference chip startup reached a $2.9B valuation on the argument that data centers are short…
An inference cloud bets on rival silicon while the bears circle
General Compute raised $100M at an $800M valuation by renting out chips nobody else wanted to build around.
A neocloud bets on inference silicon and walks away with fresh capital
General Compute reached an $800M valuation by renting bitcoin sites and buying chips everyone else ignored.
Nvidia backs a Bengaluru startup selling less GPU burn
Simplismart raised $14M on software that tunes model serving, even though its pitch implies customers need fewer chips.
Cloudflare’s model router prices risk from the last few turns
Auto Router classifies each request across 14 task types and scores it for stakes, ambiguity and context dependence…
IBM makes its Spyre chip a native PyTorch target
A rewrite of IBM's compiler moved Spyre onto PyTorch's Inductor backend, cutting the codebase from about 273,000 lines…
Cloudflare’s new preview returns structured data, not prose
Clairic answers structured questions from documents in one pass, with no generated tokens and no reasoning step.
IBM’s llm-d reaches 1.0 with controls enterprises kept asking for
The IBM-led inference framework adds workload arbitration, upgrade survival and better hardware use for teams running their own…
Tenstorrent doubles its value to $6.2B on open-weight inference
Jim Keller's chip company raised $1.5B after showing its Galaxy servers could serve DeepSeek models.