other
High-Bandwidth Memory (HBM) is a stacked DRAM chip technology offering higher bandwidth than conventional memory, used with GPUs and AI accelerators.
No evidence-backed relationships are recorded.
DDR4 contract prices rose 4% in September to a record $26 after a 14% gain in July, while MLC NAND rose 0.44%. Analysts tie Micron's shrinking earnings beats to multi-year supply deals, so the peak may show up in results before it shows up in contract prices.
Reality
Volantis, a San Francisco chip startup, closed an $88 million Series A to replace the electrical links between AI chips and memory with optical ones. Its A-1 system, promised to customers in 2027, puts a date on the bet that memory, more than compute, now limits AI inference.
Perspective Coverage
6 publishersReality
A Nanya-backed designer says DRAM hybrid-bonded to the processor belongs between on-die SRAM and HBM for inference. Design fees are almost 40 percent of its first-half revenue, and the first volume program is not booked before 2027.
Reality
The Dutch startup says craftwerk is designed to run AI agents, and the design details it has published so far are about ASICs and processor-memory co-design. Samsung, one of the biggest HBM suppliers, co-led the round.
Reality
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 Reno startup says its Asimov ASIC carries 288 GB of on-package LPDDR5X and realizes more than 90% of its memory bandwidth. Both figures come from the company, and no independent benchmark accompanies them.
Reality
Positron is now worth $5 billion on a design that swaps HBM for the memory that ships in phones, but Asimov has not taped out, mass production is set for the second half of 2027, and the 26x tokens-per-dollar figure comes from simulation.
Reality
The Verge expects Apple to raise iPhone prices this week, but the decision that set this up was Micron's in 2021, when it concluded only new plants could add capacity. Memory has stopped being a deflationary line item.
Reality
TrendForce has DDR2 up 55-60% this quarter and another 35-40% next, which multiplies to about 2.1 times, so the cheapest fallback part on the board is repricing faster than most bills of materials get revisited.
Reality
On TrendForce's numbers the memory bill grows 117% next year while the budget it sits inside grows 50%, which is why a headline capex print has stopped telling you how much compute actually got added.
Reality
DRAM revenue up 246.6% and NAND up 371.9% would double memory's share of the chip industry in a single year, though the same Gartner forecast has memory growing slower than everything else by 2027.
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
Taiwan and South Korea are splitting the AI server chain because demand exceeds what either can build. For 2026 supply forecasts, the binding step is the handoff, not any single fab.
Reality
Seoul Economic Daily calls glass substrates the next packaging battleground. The committed dollars it names belong to Intel; the only Korean number in the story is a share price.
Reality
A billion dollars a year buys Micron a US research base, not parity. Samsung's research line alone last year was larger than everything Micron spent on plant and R&D combined.
Reality
Micron's stock is up more than tenfold since end-2024 and it is spending $50 billion at home. Its employees and its host city now hold the same cyclical bet three ways.
Reality
IDC says memory tightness runs through 2027, driven by AI demand that compounds and producers who are managing supply on purpose. Treating DRAM as a spot commodity is now a two-year budgeting error.
Reality
TrendForce now expects domestic silicon to take about 90% of China's AI accelerator market in 2026, up from 45%. The arithmetic says the constraint is fab and packaging output, not Washington.
Reality
Chey Tae-won told CNBC he is willing to build a front-end fab in the U.S. but cannot find a site, and warns rising chip prices will reach finished goods. Memory is now a 2027 cost line.
Reality