A Shenzhen-based artificial intelligence infrastructure company has accidentally handed US prosecutors what could become the smoking gun of the Nvidia chip smuggling era. On April 10, 2026, invoice records surfaced in filings to China’s State Taxation Administration showing that Sharetronic Data Technology Co. procured $92 million worth of Super Micro and Dell servers compatible with banned Nvidia H100 and H200 GPUs – equipment that has been prohibited from export to China without a US license since October 2022.
The disclosure, first reported by Bloomberg News and confirmed by multiple outlets the same day, lands at the worst possible moment for the AI hardware supply chain. It arrived less than four weeks after federal prosecutors in Manhattan unsealed a sweeping $2.5 billion smuggling indictment against Super Micro co-founder Yih-Shyan “Wally” Liaw – and it provides the first publicly documented trail of where banned servers actually landed inside China.
Sharetronic’s Shenzhen-listed shares dropped roughly 10% on the news, while Super Micro Computer Inc. (NASDAQ: SMCI) slipped 12% in the trading session that followed. The Sharetronic filings detail 276 Super Micro SYS-821GE-TNHR servers and 32 Dell PowerEdge XE9680 systems – both platforms purpose-built to host the very Nvidia HGX accelerator boards that US export controls are designed to keep out of Beijing’s hands. Tom’s Hardware independently corroborated the tally, describing Sharetronic, an Nvidia Cloud Partner, as having procured “almost 300” banned-GPU-compatible servers worth $92 million in the May-June 2025 window – a count that lines up almost exactly with Bloomberg’s own 308-unit total.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Sharetronic’s April 10 Disclosure Actually Says
The records uncovered by Bloomberg are not a press release or a corporate filing. They are tax invoices, lodged with China’s State Taxation Administration between May and June 2025, that document intra-company transfers from Sharetronic Data Technology to its own wholly owned Shenzhen subsidiary. According to the documents, the transaction value tied to the banned-compatible servers was 632 million yuan, equivalent to roughly $92 million at April 2026 exchange rates – though Bloomberg’s review found that figure sits inside a much larger paper trail. The filing set runs to 17 separate invoice documents, which together point to roughly 6 billion yuan in total planned Nvidia-related server procurement as of May 2025, suggesting the disclosed H100/H200-compatible purchase is only a fraction of what Sharetronic intended to buy.
The platform inventory is striking. The Super Micro SYS-821GE-TNHR is an 8U air-cooled chassis explicitly designed to host the Nvidia HGX H100 and HGX H200 8-GPU baseboards. The Dell PowerEdge XE9680 supports the same Nvidia HGX H100 and H200 modules, as well as AMD MI300X accelerators. Neither chassis ships to China without a Commerce Department license under the rules tightened in October 2023 and again in December 2024 – and by the time the invoices were dated, in May and June 2025, Nvidia’s H100 and H200 lines were already unambiguously barred from Chinese export under those rules, a timeline later underscored by 3DNews’ review of the filings, which noted the servers were booked well after the ban had taken full effect.
Sharetronic insists that its purchases complied with all applicable Chinese hardware procurement regulations and that it had no direct business relationship with Super Micro Computer Inc. or Dell Technologies. The invoices, the company says, reflect equipment obtained through “compliant channels” – language that has become a euphemism for the gray-market broker networks that have flourished across Southeast Asia since the 2022 rules took effect.
Both Super Micro and Dell told reporters they have no record of selling the equipment in question to Sharetronic. That denial, however, only deepens the mystery: if the servers reached Shenzhen, and the OEMs say they did not sell them there, then somebody else moved them – exactly the conduct alleged in the Super Micro indictment unsealed on March 19, 2026.
The Numbers Behind the Disclosure
The scale of the Sharetronic transaction makes it impossible to dismiss as a clerical accident. Each Super Micro SYS-821GE-TNHR populated with eight H100 SXM5 modules has historically sold in the gray market for $290,000 to $360,000, with H200 configurations climbing past $400,000 per chassis. The Dell XE9680 commands similar pricing. The 308 systems Sharetronic disclosed easily reach the $92 million invoice figure even at the low end of those gray-market ranges.
| Metric | Value | Source / Date |
|---|---|---|
| Total invoice value | 632M yuan / ~$92M USD | State Taxation Administration filing, May-June 2025 |
| Super Micro SYS-821GE-TNHR units | 276 servers | Bloomberg invoice review, April 10, 2026 |
| Dell PowerEdge XE9680 units | 32 servers | Bloomberg invoice review, April 10, 2026 |
| Compatible Nvidia GPUs | HGX H100, HGX H200 (8-GPU) | Super Micro and Dell platform specs |
| Total GPU capacity (8 per chassis) | ~2,464 H100/H200-class GPUs | Tech Insider calculation |
| Sharetronic Shenzhen stock drop | ~10% on disclosure day | Shenzhen Stock Exchange, April 10, 2026 |
| Super Micro (SMCI) drop | ~12% intraday | NASDAQ, April 10, 2026 |
| Date filings surfaced | April 10, 2026 | Bloomberg News |
For perspective, 2,464 H100-class GPUs is enough silicon to train a frontier large language model from scratch. Meta’s original Llama 3 70B was trained on approximately 6,000 H100s; OpenAI’s GPT-4 used a cluster on the order of 25,000 A100s. The Sharetronic invoice on its own represents about 10% of a frontier-scale training cluster – and that is one disclosure, from one Shenzhen company, surfaced almost a year after the equipment was procured.
How the Super Micro Indictment Set the Stage
To understand why the Sharetronic disclosure matters, you have to start with the federal complaint unsealed in the Southern District of New York on March 19, 2026. That indictment named three individuals: Yih-Shyan “Wally” Liaw, the 71-year-old co-founder of Super Micro who served on its board and oversaw business development; Ruei-Tsang “Steven” Chang, the company’s general manager in Taiwan, who fled to an undisclosed Asian location ahead of charges; and Ting-Wei “Willy” Sun, described in court papers as a logistics fixer.
The DOJ alleges the trio conspired to ship approximately $2.5 billion in Super Micro servers equipped with restricted Nvidia H100, H200, and Blackwell B200 GPUs to Chinese buyers through front companies based in Southeast Asia. The indictment describes warehouses stocked with mislabeled chassis – labels reportedly stripped using ordinary hair dryers – to deceive customs officers and corporate auditors alike. Each conspiracy count carries up to 20 years; smuggling adds another 10, and fraud another five.
Super Micro Computer Inc. itself was not charged. CEO Charles Liang told investors in a March 26 letter that the company “is not a defendant” and described Super Micro as “a victim” of the scheme, noting that it had fully cooperated with prosecutors and appointed Deanna Luna as acting Chief Compliance Officer. On the company’s fiscal Q3 earnings call on May 5, 2026, Liang went further: “Based on what we know so far, no one from the company other than those named in the DOJ indictment was involved.” SMCI stock jumped 18% in after-hours trading on the reassurance.
The Sharetronic disclosure, surfacing between the indictment and the earnings call, complicates that narrative. It does not directly name Super Micro as a knowing party – and Super Micro denies any sales record matching the invoices – but it shows, for the first time publicly, exactly the kind of Chinese end-user that prosecutors say the conspirators were serving.
Why Sharetronic Is Not an Obscure Buyer
Sharetronic Data Technology is not a fly-by-night reseller. Through its Guangzhou Fcloud Technology joint venture, the company became one of only eight Nvidia Cloud Partners certified to operate inside mainland China, a status that authorizes it to deploy approved Nvidia hardware for domestic Chinese AI workloads. After receiving NCP certification, Sharetronic publicly announced plans for roughly 32.2 billion yuan ($4.6 billion) in total hardware procurement over the program’s life.
That makes the appearance of restricted H100- and H200-class platforms on its invoice ledger especially uncomfortable for Nvidia. The Santa Clara company has spent the better part of two years arguing that its export-controlled GPUs do not flow into China at meaningful scale, and that the China-compliant H20 and B30A variants it sells legally are a sufficient release valve. The Sharetronic filing punctures that narrative – and it does so using documents that the Chinese government’s own tax bureau collected.
Nvidia declined to comment specifically on Sharetronic but reiterated to reporters that its products include “export compliance features” and that the company “fully complies with US Commerce Department licensing requirements.” The company’s own quarterly disclosures show China data center revenue fell roughly 25% year over year in fiscal Q1 2026 to about $3.5 billion, largely because of the export tightening – a decline that the Sharetronic numbers suggest is being partially offset by gray-market flows that Nvidia is not booking. The picture is complicated further by Beijing’s own posture: a Reuters-cited analysis published in April 2026 reported that Chinese authorities had separately ordered customs agents to bar Nvidia H200 AI chips from entering the country altogether, meaning Sharetronic’s H200-compatible servers were technically unwelcome on both sides of the Pacific even as they moved through gray-market channels.
Expert Reaction: From Capitol Hill to Wall Street
The Sharetronic revelation drew immediate fire from lawmakers who have spent years pressing for tougher enforcement. House Select Committee on the Chinese Communist Party Chairman John Moolenaar (R-Mich.) told reporters: “If a Chinese AI infrastructure company can buy hundreds of restricted servers and book them through a state tax filing, our export controls are not controls – they are a paperwork exercise. We need real penalties for the middlemen and real audits for the OEMs.”
From the investor side, Stacy Rasgon, senior semiconductor analyst at Bernstein Research, wrote in a note to clients dated April 11: “The Sharetronic disclosure does not prove Super Micro knew where the equipment ended up. What it does prove is that the gray-market plumbing exists at industrial scale. For SMCI, the legal risk is bounded by the indictment; the reputational risk is unbounded as long as new invoices keep surfacing.”
Gregory Allen, director of the Wadhwani AI Center at the Center for Strategic and International Studies, has spent the past two years warning that the original 2022 rules were too lax to be enforced. “We have built an export control regime that depends on the honesty of every reseller in Southeast Asia,” Allen told the Financial Times. “Sharetronic is not exceptional. It is the first one where the receipts are public.”
Paul Triolo, partner for China and technology policy at DGA-Albright Stonebridge Group, framed the disclosure in geopolitical terms: “Beijing knows exactly what this filing looks like to Washington. The fact that the tax administration did not scrub it suggests either bureaucratic indifference or a deliberate signal that China can keep procuring even under tightened rules.”
Dan Ives of Wedbush Securities took the bull case: “This is a $0.50 stock move dressed up as a $50 narrative. Super Micro’s underlying AI server pipeline is still booked at $40 billion through fiscal 2027. The company has fired the bad actors and replaced the CFO. SMCI will trade through this in two quarters.”
The Export Control Timeline That Made This Possible
To understand how restricted servers ended up on a Chinese tax filing, you need the four-year arc of US export controls. The Bureau of Industry and Security (BIS) first issued the October 7, 2022 advanced-computing rule, which set performance thresholds that effectively blocked the H100 and A100 from sale to China. In October 2023, BIS plugged the loopholes that had allowed Nvidia’s H800 and A800 export-compliant variants, then in December 2024 added 140 Chinese entities – including chip toolmakers and AI design houses – to its Entity List.
Despite those moves, the smuggling cases have piled up. In March 2026 alone, US Customs and Border Protection seized roughly 500 Nvidia H100-class servers valued near $200 million at a Texas freight forwarder. Earlier in February, BIS opened a civil enforcement docket against an unnamed Singapore distributor accused of routing roughly $430 million in Nvidia-bearing systems through Malaysia. Every one of those cases relied on customs documents inside the United States – never before on Chinese tax records.
| Date | Action | Scope / Value |
|---|---|---|
| Oct 7, 2022 | BIS advanced-computing rule | Blocks H100/A100 sales to China |
| Oct 17, 2023 | BIS closes H800/A800 loophole | Adds performance density threshold |
| Dec 2, 2024 | BIS Entity List update | 140 Chinese tech entities restricted |
| Mar 19, 2026 | SDNY indictment unsealed | $2.5B in Super Micro servers, 3 defendants |
| Mar 2026 | CBP Texas seizure | ~500 H100 servers, ~$200M value |
| Apr 10, 2026 | Sharetronic disclosure | $92M in restricted-compatible servers |
| Apr 11, 2026 | Senate hearing scheduled | Banking Committee on enforcement gaps |
The pattern is unmistakable. Each new enforcement action publicly surfaces something that has plainly been going on for months in private. The Sharetronic invoice is dated May-June 2025; the public disclosure is April 10, 2026. That is roughly an 11-month lag between the underlying transaction and the regulatory daylight – a window in which several thousand restricted GPUs can be deployed, trained on, and depreciated.
The Market Impact: $1.5 Trillion in AI Hardware on Edge
The investor reaction to Sharetronic was sharper than the dollar figure alone would justify because of where the news lands in the AI infrastructure cycle. Nvidia’s revenue ran at $60.9 billion in fiscal Q4 2026, with data center the dominant segment. Super Micro’s AI server backlog stood near $40 billion through fiscal 2027, by management’s own characterization on the May 5 earnings call. Dell’s Infrastructure Solutions Group booked roughly $13.8 billion in orders in its most recent quarter, much of it AI server pipeline.
Each of those numbers depends on the assumption that US export controls remain enforceable and that the OEMs are not exposed to material secondary liability. A wave of additional Sharetronic-style disclosures – one in five Nvidia Cloud Partners is China-based, by one industry estimate – could force a re-rating of every server stock in the AI complex. Wedbush’s Ives notes that Super Micro now trades at roughly 0.6x EV/Sales, against an OEM peer group at 1.1x; the gap is the smuggling discount.
The pressure also extends to Beijing. Sharetronic shares are listed on the Shenzhen Stock Exchange under ticker 300857.SZ and have a tight retail shareholder base that is unusually sensitive to US policy news. The 10% drop on disclosure day represented the equivalent of a single-day decline of roughly 1.8 billion yuan ($250 million) in market value. If other Chinese NCPs face similar scrutiny, the chilling effect on China’s domestic AI capex could be material.
Historical Context: From ZTE to Huawei to Today
The Sharetronic episode sits inside a longer history of US-China technology enforcement. In 2017, ZTE Corp. was hit with a $1.19 billion penalty for shipping US-origin telecom equipment to Iran and North Korea – at the time, the largest export-control settlement ever. In 2019, the Trump administration added Huawei to the Entity List, eventually crippling its 5G handset business. In 2020, Fujian Jinhua was barred from buying US semiconductor equipment after a DRAM trade-secrets dispute.
None of those cases involved AI chips. The Sharetronic-Super Micro arc is the first big enforcement moment for the AI era, and its character is different. ZTE was punished after a fact-finding settlement; Huawei was constrained by Entity List exposure. Sharetronic, by contrast, is a Chinese buyer whose own filings provide the evidentiary trail. The shift signals that enforcement will increasingly rely on foreign government data, social-engineering of broker networks, and after-the-fact accounting forensics.
It also signals the limits of the OEM-cooperation model. After the March 19 indictment, Super Micro promoted a Chief Compliance Officer, fired implicated employees, and engaged independent auditors – every step that the post-ZTE playbook says you should. The Sharetronic disclosure surfaced anyway. That suggests the next iteration of enforcement will require either real-time end-user verification of every server shipment or much tighter pre-clearance on which Chinese cloud partners are allowed to buy anything at all.
Competitive Comparison: Where Sharetronic Fits in China’s AI Stack
Sharetronic is small relative to China’s top-tier AI buyers, but its certification status gives it outsized strategic weight. Alibaba’s Aliyun, Tencent Cloud, and ByteDance are widely reported to operate the largest Chinese clusters of restricted Nvidia hardware, much of it acquired before the 2023 rule tightening. DeepSeek and Moonshot AI, the model labs behind two of China’s most competitive open-source frontier models, lean on hybrid pools that include the export-compliant H20 alongside legacy A100s.
Sharetronic occupies a middle layer: an infrastructure-as-a-service player that resells GPU capacity to Chinese enterprises and research institutes. That positions it as a discovery point for any enforcement audit, because the company necessarily issues invoices and tax filings that are visible to provincial regulators. A hyperscaler like Aliyun, by contrast, can paper over hardware origins inside its own consolidated accounts. The next-cycle question for Washington is whether the same forensics that exposed Sharetronic can be turned on the larger players.
The OEM side is similarly tiered. Super Micro and Dell are the most exposed to restricted-platform smuggling because their 8-GPU systems are the workhorses of frontier AI training. HPE, Lenovo, Inspur, and Sugon are less affected: HPE’s Cray division focuses on supercomputing, Lenovo has shifted increasingly to domestic Chinese GPU integrations, and Inspur and Sugon themselves were Entity-Listed by BIS in 2023 and 2024 respectively.
Five Predictions for the Next 12 Months
1. More Sharetronic-style disclosures will surface from Chinese provincial tax records. Bloomberg’s invoice review was not the product of a leak; it was the product of database reporting on filings that any qualified accountant in China could see. Expect at least three more comparable disclosures from Chinese NCPs or large infrastructure resellers by the end of 2026.
2. BIS will issue a new rule mandating end-user attestations for every advanced-computing shipment above a specified value threshold. The October 2022 and 2023 rules apply broad performance bans; the next rule will move to transaction-level verification, possibly tied to a digital ledger of HGX baseboard serial numbers. A draft is widely expected in the second half of 2026.
3. Super Micro will face a federal civil penalty even though the company is not a criminal defendant. The pattern from ZTE and Schlumberger is that OEMs settle civil cases for “failure to maintain effective compliance controls” without admitting wrongdoing. A penalty in the $300-$600 million range by the end of fiscal 2027 is plausible.
4. Nvidia will accelerate the rollout of its China-compliant B30A and B40 variants to drain demand from the gray market. The math is straightforward: every dollar Beijing’s buyers spend on a smuggled H200 is a dollar Nvidia does not book. Expect Nvidia to widen its compliant-China lineup and to lobby BIS for slightly higher performance thresholds in exchange for tighter end-user tracking.
5. China will respond by accelerating Huawei Ascend 910C and 920 deployments. The political cost of being publicly caught buying restricted Nvidia equipment now outweighs the performance penalty of using domestic alternatives. Huawei’s Ascend roadmap is already on a 12-month cadence, and Sharetronic-type incidents will push Chinese cloud partners to over-rotate to domestic silicon, accelerating the bifurcation of the global AI supply chain.
What Sharetronic Means for the Broader Hardware Market
The biggest unanswered question is whether Sharetronic is an isolated breach or the leading edge of a much larger documented flow. The Bloomberg review covered one tax filing window, May-June 2025, for one company. Roughly two dozen Chinese infrastructure firms were certified as Nvidia Cloud Partners during that period, and many more buy through second-tier reseller chains. If even one in three of those entities holds comparable invoices, the cumulative dollar volume of restricted equipment inside China could reach $5 billion to $9 billion.
That has knock-on implications for hyperscaler capex. Big Tech’s 2026 AI infrastructure spend now sits near $725 billion globally, up from $670 billion at the start of the year. If BIS responds to Sharetronic with shipment-level attestations, the friction added to every H200 and Blackwell B200 sale will compress OEM margins by an estimated 150-250 basis points. Vendors with stronger compliance pedigrees – HPE, Lenovo on the non-restricted side – will gain modest share.
For Beijing, the takeaway is more political than technical. The Sharetronic invoices show that the gray market works – and that Washington can find out about it. That dual reality will push Chinese policymakers to publicly champion domestic silicon while quietly funding the supply chains that keep restricted hardware flowing. Expect Huawei, Cambricon, and Biren to receive accelerated state subsidies through the rest of 2026.
Frequently Asked Questions
What did Sharetronic Data Technology actually disclose on April 10, 2026?
Bloomberg News surfaced tax invoices filed with China’s State Taxation Administration showing that Sharetronic procured 276 Super Micro SYS-821GE-TNHR and 32 Dell PowerEdge XE9680 servers – both platforms compatible with banned Nvidia H100 and H200 GPUs – for a combined 632 million yuan, or about $92 million. The filings cover transactions from May and June 2025 routed through a Shenzhen subsidiary.
Is Sharetronic the same company involved in the Super Micro indictment?
No. The SDNY indictment unsealed on March 19, 2026 named three individuals associated with Super Micro – Wally Liaw, Steven Chang, and Willy Sun – but did not identify the Chinese buyers. Sharetronic’s disclosure is the first publicly documented account of a Chinese AI infrastructure firm holding hundreds of restricted-compatible Super Micro and Dell systems, but neither prosecutors nor Sharetronic itself has confirmed a direct link to the indictment.
Did Charles Liang face criminal charges?
No. Super Micro CEO Charles Liang has not been charged. Super Micro itself is not a defendant. Liang stated in the company’s March 26 investor letter that Super Micro is “a victim” of the alleged scheme and that the company is cooperating fully with the federal investigation.
How big is the smuggling problem in total?
Combining the $2.5 billion alleged in the Super Micro indictment, the $92 million on Sharetronic’s books, the $200 million CBP seizure in Texas, and the $430 million Singapore civil docket, publicly documented restricted-equipment flows already exceed $3.2 billion. That is before accounting for Sharetronic’s own broader tax trail: the 17 State Taxation Administration documents Bloomberg reviewed point to roughly 6 billion yuan (~$840 million) in total planned Nvidia-related server procurement, of which the $92 million in H100/H200-compatible chassis is just the confirmed slice. Industry estimates of the unreported gray-market total range from $5 billion to $9 billion since the 2022 rules took effect.
How did Super Micro stock react to the Sharetronic disclosure?
SMCI fell roughly 12% intraday on April 10, 2026, on the headlines, then partially recovered. The stock subsequently rose 18% in after-hours trading following the May 5 fiscal Q3 earnings call, where CEO Charles Liang reiterated that no one outside the three indicted individuals appeared to have been involved. As of mid-May the stock remained well below its 2024 peak.
What is Nvidia’s exposure?
Nvidia itself is not a defendant and has not been named in any active enforcement action stemming from the Sharetronic filings. The company has consistently said its products include compliance features and that it follows all BIS licensing rules. However, the disclosure puts pressure on Nvidia to demonstrate it has visibility into final end-users for HGX-class baseboards, which is the bottleneck for any future shipment-level attestation regime.
How could this affect future AI chip exports to China?
Expect BIS to move from performance-threshold rules toward transaction-level verification, possibly via a serial-number ledger for HGX baseboards. Chinese AI firms will face greater scrutiny when buying from any global OEM channel. In parallel, Beijing will accelerate domestic alternatives, particularly Huawei’s Ascend 910C and 920 lineups.
Conclusion: A Paper Trail That Changes the Conversation
For three years, the central question of US AI chip export controls has been whether the rules are enforceable. The Sharetronic Data Technology disclosure on April 10, 2026 changes the way that question can be answered. It is no longer a matter of inferring leakage from customs anomalies or shipment manifests; the receipts are now publicly visible inside the Chinese tax system. That is uncomfortable for Beijing, dangerous for Nvidia’s revenue narrative, and existential for any OEM whose compliance programs cannot prove a clean chain of custody.
The $92 million figure on its own is small relative to the $2.5 billion Super Micro indictment, the $725 billion in AI capex committed by big tech in 2026, or the $40 billion in Super Micro’s own AI server backlog. But it is a single data point in a curve that is just beginning. If even a fraction of China’s Nvidia Cloud Partners hold comparable invoices, the next 12 months will see export controls move decisively from threshold-based bans to transaction-level surveillance – and the AI hardware industry will rebuild itself around that new compliance overhead.
External references: Bureau of Industry and Security advanced computing rules · US Attorney’s Office, Southern District of New York · Nvidia Newsroom · Federal Register · Office of the US Trade Representative.
