Huang, identified in chip-industry coverage as the head of NVIDIA, has committed to roughly double chip output through a 37% reservation of high-bandwidth memory (HBM) supply running through 2026, according to a shattered.io trade analysis.

The pledge — framed in the report as Huang's "2x chip vow" — converts a doubled chip-volume target into a fixed share of stacked DRAM capacity across calendar year 2026. It is the memory side of the commitment, rather than the chip-volume target, that quantifies the lockup.

What 37% of HBM means in procurement terms

HBM is the vertically stacked DRAM technology used in data-center AI accelerators. Memory bandwidth commonly sets the training-time ceiling on large models; HBM stacks therefore sit on the critical path for every shipped GPU in the accelerator class.

A 37% allocation corresponds to a multi-year reservation of HBM wafers across the 2026 delivery window. HBM contracts typically run on pre-negotiated wafer bookings negotiated well ahead of delivery, so a reservation of this size removes a comparable slice of qualified supply from competing buyers for the contract period.

How the "2x" framing maps to chip output

The doubled volume element anchors the supply position to a stated chip-output target. The analysis does not break out:

• The NVIDIA product line covered by the doubled volume
• The unit count or wafer equivalent behind the 37% memory allocation
• The pricing terms or option structure of the reservation

The report frames the commitment as a single package — chip volume and memory share together — without segmenting the two components.

Why 2026 is the relevant endpoint

The 2026 calendar year is the explicit endpoint of the reservation. HBM capacity planning operates on multi-quarter lead times, so a booking that runs through end-2026 ties up a matching share of qualified capacity for the 12 months it covers.

Any reservation booked against 2026 output would draw on production schedules already in place at the memory vendors, leaving limited room to flex the allocation mid-year.

What the trade note does not specify

The shattered.io analysis summarises the commitment at headline level. It does not name the HBM vendors supplying the reserved capacity, identify the specific NVIDIA product line tied to the doubled volume, or quote the parties to the agreement. Readers should treat the 37% figure as an allocation summary rather than a confirmed contract count until the underlying agreements are disclosed.

The note also leaves unanswered whether the 37% applies to HBM volume, HBM revenue, or HBM wafer starts — three different denominators that produce different headline percentages from the same physical capacity.