Topic
Topic covering AI inference hardware beyond standard GPUs, such as wafer-scale chips, decode accelerators and inference cards, plus compute supply deals.
Intel named Crescent Island the GPU D-Series at OCP 2026, a 350W air-cooled PCIe card with up to 480GB of LPDDR5X. Tom's Hardware reported a 1.3 PFLOPS FP8 peak target and 1.5 TB/s of bandwidth; customer sampling is in progress.
Perspective Coverage
5 publishersReality
IBM's torch-spyre team made the Spyre inference accelerator a native PyTorch device on PrivateUse1, the allocator, streams and Inductor. Teams on IBM Z, LinuxONE and Power can reach the card from ordinary eager and compiled PyTorch code through one launch path.
Reality
The accelerator is in full production and the headline number is a single-request generation rate at 100,000 tokens of context. That is a different purchase order than throughput.
Perspective Coverage
3 publishersReality
Groq 3 LPX is in full production with Nebius as the named first customer. The benchmark is one model at one context length, and the 4x claim does not quite get from hours to minutes.
Reality
The 1.7x to 3.6x latency range is set by the baseline systems, not the chip, and the report's own publication date is unsettled. Read it as direction, not evidence.
Perspective Coverage
9 publishersReality
The $875 million round prices an inference chip whose performance figures come from cycle-accurate simulations and whose production date has moved from early 2027 into the second half. The system Positron ships today runs on HBM.
Reality
The first third-party benchmark of the LP30 rack came in at roughly four times the next-fastest public endpoint, measured one request at a time on a model small enough to fit.
Reality
d-Matrix says stacking compute on a co-designed DRAM die moves bits at roughly a sixth of HBM4's energy cost. It still has not said who fabricates the die.
Reality
Intel's Hot Chips disclosure details 32 Xe Cores, 32MB of L2 and 16-deep matrix engines. The figure it does not include is the one that decides whether the capacity is servable.
Reality
Talks over a partnership, investment or acquisition would give Nvidia an energy-efficient NPU line and the sovereign-AI customers already buying it, rather than leaving both to regional rivals.
Reality
The first multi-wafer Cerebras system pairs a doubled clock with rebuilt power delivery and interconnect. The compute claims rest on WSE-3 Turbo dies that are otherwise unchanged.
Reality
Etched has first-pass silicon, 400 staff and more than $1bn in booked orders. What it does not have in public is a peak FLOPS figure, a power draw, or a third-party benchmark.
Reality
OpenAI's invite-only Ultrafast tier runs the same GPT-5.6 Sol up to 14 times quicker, while Google halves Gemini Flash pricing until December 31. Latency is now its own budget line.
Perspective Coverage
3 publishersReality