Hot Chips is an annual symposium where semiconductor and systems companies unveil new processor, accelerator, and memory chip designs.
Samsung told Hot Chips 2026 it intends to hollow out the accelerator by relocating its non-compute blocks into a 4nm HBM base die. The stated endpoint of that plan is a memory stack with no interposer under it.
Reality
More memory on a Cerebras wafer arrives with CS-6, two generations out. Everything shipping before then, Nexus included, is rack engineering around the SRAM budget the wafer already has, and long contexts are where that budget hurts.
Reality
The zettaFLOPS figure is an estimate. The measured one, sitting in a GB300 example, is a 25 percent density gain, and it is paid for out of the safety margin.
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
Fujitsu halved its vector width, moved the whole last-level cache onto a 5nm die, and dropped HBM. The published estimates scale exactly with base clock, which tells you what kind of number they are.
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
OXMIQ's Hot Chips model gives a flash-backed rack 14 times the capacity and 58 percent of the bandwidth. Two of the three HBF grades cannot keep pace with today's HBM.
Reality
Two 70-core N3P chiplets, 12 DDR5 channels and a 2 TB/s UCIe link give capacity planners something to price. The 844.8 GB/s peak assumes DDR5-8800 modules Tom's Hardware says have not reached market.
Reality
At Hot Chips 2026 Samsung detailed a validated LPDDR5X-PIM part claiming 614 GB/s of internal bandwidth and 2.28x to 3.01x inference gains over plain LPDDR5X.
Reality
At Hot Chips 2026 Intel said an open die-to-die link, not advanced packaging, is what made its entry-level 18A chip possible. The bill came to 70% more interconnect area.
Reality
Micron's HBM design fellow says the memory's roughly threefold wafer-area penalty against DDR5 grows with every generation. The pricing that follows does not mean-revert on its own.
Reality
One SKU, one compute die, 1.2 TB/s of LPDDR5X, and benchmark wins over a 96-core EPYC. The comparison with AMD's Venice is being set on Nvidia's terms rather than on core counts.
Reality
Intel's Hot Chips disclosure gives capacity and licensing planners real numbers for 2027, but leaves out the one figure that would let anyone model per-core performance.
Reality
Copper-to-copper slips to HBM5 at the earliest, so HBM4E capacity has to come out of die thickness inside a 775-micron package. iHBM heat blocks cannot be retrofitted to parts already in design.
Reality
The dual-ISA core runs AArch64 and z/Architecture on the same core with nanosecond switching. x86 is named in the pitch and absent from the silicon, and the uptime arithmetic does not close.
Reality