AMD's XCZU47DR RFSoC appears in a Shanghai-built SDR on Crowd Supply
The product at the center of the dispute is the PZSDR P047, which Puzhi is selling through the crowdfunding platform Crowd Supply. The listing describes a single-chip software-defined radio (SDR) platform based on the ZYNQ UltraScale+ RFSOC XCZU47DR with support for 8T8R synchronized operation, an RF range of 1 MHz to 6 GHz, a 5 GSPS RF ADC, and a 9.85 GSPS RF DAC. The page says the unit is produced by Puzhi in Shanghai, China and sold and shipped by Crowd Supply, with Mouser Electronics, Crowd Supply's fulfillment partner, handling delivery to backers worldwide.
The complete system costs $8,749. When the page was captured, it showed $39,645 raised against a $1 goal from four backers, and an August 27, 2026 update said Puzhi had boards in hand and was inspecting and testing them. The promised ship date differs between captures of the page: one says orders ship November 6, 2026, and another says orders placed now ship Nov 05, 2026. Tom's Hardware notes that Avnet and DigiKey quote XCZU47DR lead times of 40 to 52 weeks. A Chinese boutique board maker promising delivery within weeks is a big part of why the story drew attention.
The P047 is not Puzhi's only product with this chip. The company's own site lists a PZ-ZU47DR-SOM system-on-module and a PZ-ZU47DR-KFB development board, both specified with the XCZU47DR-2FFVE1156I. A Puzhi storefront on eBay has offered the Puzhi FPGA Core Board Xilinx ZYNQ UltraScale+ RFSOC XCZU47DR Industrial Grade for US $3.999,00, shipped from Shanghai's Pudong New Area. The listing snapshot shows a last update in July 2025, which suggests these parts have been available from Puzhi for more than a year.
Puzhi also isn't the only company building on this silicon. RTL-SDR.com, which covered the campaign weeks before the controversy, noted that the AntSDR T510 is also based on the ZU47DR RFSoC and is also due to launch on Crowd Supply soon.
Dylan Patel's 'treason' charge rests on a price gap
Patel made the accusation on X. He wrote that the chip's U.S. list price is $36,000, that U.S. companies pay $4,000–$5,000 at volume, and that it was "getting quoted $1K in China to crowdfunding campaigns." He argued this amounted to "selling military end use components to an adversary for less than your home country." The treason framing belongs to Patel. No authority has alleged treason or announced an investigation.
The $1,000 figure is the weakest part of the claim. Tom's Hardware stressed that the $1,000 figure, however, is Patel's reported price for the chip rather than a publicly advertised retail offer. Nothing on the Crowd Supply page or Puzhi's product pages shows a price for the bare chip. The $8,749 system price and the $3,999 module price don't reveal what Puzhi paid for the RFSoC inside.
The general gap is harder to dismiss, and it was noticed before Patel spoke up. RTL-SDR.com wrote in early September that when purchased from Western parts distributors like Digikey, the same AMD XCZU47DR-2FFVE1156I chip used in the PZSDR costs US$30,000+ for the chip alone. The site concluded that AMD was giving Asian manufacturers much better pricing. The public numbers don't settle that conclusion either. A Western distributor's single-unit list price and a volume contract price can differ enormously, as Patel's own $4,000–$5,000 volume figure shows. The anomaly anyone can check is simpler: a Shanghai vendor sells a complete module, with memory, flash, power circuitry and connectors, for less than the Western single-unit price of the chip it carries.
AMD's statement denies direct sales but doesn't confirm a probe of this case
AMD's reply to Tom's Hardware followed a standard compliance script. A spokesperson said the company "is committed to full compliance with all U.S. and global export regulations" and requires distributors, resellers and customers to comply too. The spokesperson called any shipment into restricted regions "a direct violation of our policy" and said "this recent instance is unrelated to any direct AMD sales or shipments." AMD added that it investigates all suspected diversion cases, and that if it finds a legal violation it will report it to authorities and stop doing business with the company involved. According to Patel's follow-up post, AMD told Patel the same.
The statement is limited in three ways. It doesn't confirm that AMD has opened an investigation into the Puzhi case specifically. It doesn't name a suspected intermediary. And it doesn't explain the pricing Patel described. "Unrelated to any direct AMD sales or shipments" rules out AMD shipping to Puzhi itself. It says nothing about which authorized distributor or customer first bought the parts.
Tracing that path is hard. Xilinx, which AMD later acquired, introduced its 3rd Generation Zynq UltraScale+ RFSoC family, including the XCZU47DR, in February 2019, and the devices reached production status in 2020, so the parts have been on sale for nearly six years. Tom's Hardware notes that Xilinx and then AMD have sold the chip to tens or possibly hundreds of companies in many countries. Any one of them could be the leak.
One detail on the Crowd Supply page complicates the picture. It says the PZSDR P047 is part of "AMD FPGA Playground" and lists the XCZU47DR as a key component. What that program involves, and whether AMD reviews the projects placed under it, isn't documented anywhere we could find. Still, a product built on an allegedly diverted AMD chip is being marketed under AMD branding on a Western storefront, and AMD's statement doesn't address that.
Export controls on the XCZU47DR leave room for chips to move through third countries
Tom's Hardware classifies the XCZU47DR as a dual-use item under ECCN 3A001.a.14 on the U.S. Commerce Control List. Under that classification, export, re-export or transfer to non-allied destinations such as China normally requires a license from the Commerce Department's Bureau of Industry and Security (BIS). Tom's Hardware says such a license would almost certainly be denied. We couldn't find AMD's own classification document for this part, so the specific ECCN comes from Tom's Hardware.
BIS explains how these rules work in general. An Export Control Classification Number (ECCN) is a five-character code that sorts items by type and technical parameters. Exporters can get it from the manufacturer, classify the item themselves using the Commerce Control List's order of review, or ask BIS for an official ruling through its SNAP-R system. The Commerce Country Chart then combines the item's reasons for control with the destination to decide whether a license is needed.
Tom's Hardware argues that country groups explain how the chips may have left legal channels. BIS sorts foreign countries into groups A, B, D and E, and a country's group determines which license exceptions apply. Group B countries qualify for several exceptions and face less strict license reviews. According to Tom's Hardware, a license exception allows shipments of items like the XCZU47DR to commercial companies in Group B countries, so the parts can legally reach buyers in dozens of countries and then be moved on to China illegally. That route is plausible, but nobody has shown it's the route these parts took. No public evidence traces Puzhi's chips to a particular reseller or intermediary.
This fits a pattern familiar from the GPU export restrictions: rules written around the first shipment struggle to control where a part goes after that. Here the item isn't a data-center accelerator. It's a six-year-old programmable RF chip, bought by many companies for many purposes, and much harder to track.
What the XCZU47DR does, and why it's export-controlled
A radio-frequency system-on-chip (RFSoC) puts a processor, programmable logic and high-speed data converters on one die. Tom's Hardware's description of the XCZU47DR covers four general-purpose Arm cores, two Cortex-R5F real-time cores, an FPGA with 930,000 gates, DSP engines with 4,272 DSP slices, as well as eight 14-bit analog-to-digital converters (ADC) capable of up to 5 GSPS and eight 14-bit digital-to-analog converters (DAC) operating at up to 9.85 GSPS. Because the converters are on the chip, it can receive, process and generate RF signals in real time without separate converter chips.
That integration is why the part appeals to civilian and military engineers alike. Devices based on Zynq UltraScale+ RFSoCs are used for software-defined radios, radar and phased-array systems, wireless and satellite communications, electronic test equipment, and other applications that require high-bandwidth, low-latency signal processing. Puzhi's campaign targets professional engineers and university researchers. Its own copy also mentions phased arrays, advanced radar, beamforming and satellite ground transmissions, and a campaign update lists drone design among possible uses.
Puzhi's comparison table shows how far this is from hobbyist SDR hardware. It lists the P047 with 6 GHz of RF bandwidth, 14-bit sampling and eight transmit and eight receive channels. The Ettus B210 in the same table has 61.44 MHz of bandwidth and two channels each way, and the HackRF One has 20 MHz and one channel each way. Puzhi also advertises phase-coherent synchronization across multiple boards, which is the capability needed for scaled-up beamforming arrays.
Patel called the part a "military chip," which goes further than its product line supports. AMD sells XC-series RFSoCs for commercial and industrial use and a separate XQ-series with defense-grade qualification. According to Tom's Hardware, the XCZU47DR has no direct XQ counterpart, though the closely related ZU48DR comes in both versions, and the XQ part adds ruggedization and support rather than different processing capability. The XCZU47DR is a commercial part with obvious dual-use potential. Whether it's controlled comes down to its performance specifications, not whether it carries a military label.
What this means for buyers of RFSoC hardware
Few WindowsForum readers will buy an $8,749 eight-channel SDR. But labs, universities and engineering teams that buy FPGA and RF development hardware should check where their RFSoC boards come from, and not assume a Western storefront has done that checking. Being sold on Crowd Supply and shipped through Mouser doesn't show that the chip inside went through licensed channels. Buyers of third-party XCZU47DR boards, especially organizations with export compliance duties of their own, should ask vendors to document the chip's supply chain.
The pricing is also a practical signal. When a complete module costs less than the chip's Western list price and ships within weeks while distributors quote close to a year, procurement teams should look more closely before buying.
- Puzhi's PZSDR P047 is a real product on Crowd Supply, built in Shanghai around AMD's XCZU47DR, priced at $8,749 and promised to ship in early November 2026.
- Patel's $1,000 chip price and his treason accusation are his own claims. No public listing shows the $1,000 price, and no authority has alleged treason.
- AMD denies any direct sales or shipments to Puzhi and says it investigates suspected diversion, but it hasn't confirmed an investigation of this specific case.
- Tom's Hardware reports the XCZU47DR as ECCN 3A001.a.14, which normally requires a BIS license for China. AMD's own classification isn't public.
- Organizations buying third-party RFSoC boards should ask for documentation of the chip's origin, because the retail channel alone doesn't show the part was legally exported.
- The similar AntSDR T510, also built on the ZU47DR, is due on Crowd Supply, so this question will come up again.
Wherever Puzhi's chips came from, the public record shows an export-controlled AMD RFSoC being sold openly in finished Chinese hardware, and AMD hasn't publicly explained how the chips got there. If AMD's diversion review leads to a report to BIS or to cutting off a distributor, that will show whether the company's compliance statement has consequences. Until then, Puzhi's November ship date comes first. Units will go through Mouser to backers worldwide, and the question of where their chips came from remains unanswered.