News related to
Nvidia

Nvidia is set to begin production of a new, lower‑spec Blackwell‑architecture AI chip for the Chinese market as early as June, aimed at complying with US export controls while maintaining its presence in the nation’s booming data centre sector. Priced between $6,500 and $8,000—significantly cheaper than the $10,000–$12,000 H20—the chip trades performance for accessibility under current regulatory constraints.

The streamlined variant is based on the RTX Pro 6000D, utilising conventional GDDR7 memory instead of high‑bandwidth memory and eschewing TSMC’s advanced CoWoS packaging to stay within US‑mandated memory bandwidth limits of 1.7–1.8 TB/s. Industry analysts note this is markedly below the 4 TB/s capability of the H20.

Yet Nvidia is not stopping there. As part of a dual‑chip strategy, it is reportedly developing a second Blackwell model scheduled to enter production by September. The specifications remain undisclosed but are expected to similarly align with US export regulations.

China accounts for around 13 per cent of Nvidia’s revenue, a slice that has shrunk sharply from a pre‑2022 peak due to export restrictions, which caused market share to decline from roughly 95 per cent to near 50 per cent today. A ban on the H20 in April prompted the company to write off about $5.5 billion in inventory and forgo an estimated $15 billion in sales.

CEO Jensen Huang has acknowledged the company’s “limited” options, stating that until a compliant design is approved by US authorities, Nvidia remains locked out of China’s $50 billion data centre market. Without it, his comments suggest many customers may migrate to domestic alternatives like Huawei’s Ascend 910B.

Domestic competition in China has intensified. Huawei’s Ascend 910B and others have gained momentum, closing the gap to Nvidia’s downgraded models within one to two years, according to semiconductor analysts. Nvidia counters this with its CUDA programming ecosystem — a key competitive advantage. With CUDA deeply integrated into AI development workflows globally, it may anchor developer loyalty even as hardware capabilities shift.

Further complicating the picture are whispers of a third chip internally known as “B30” or “B40”, which appears designed to operate in clusters and align with new export rules by focusing restrictions on individual chip rather than entire systems. Reports indicate interest from China’s leading tech companies—ByteDance, Alibaba and Tencent—with ambitions to produce more than 1 million units this year. Nvidia has not directly confirmed the chip, and one Chinese supplier, ZJK Industrial, briefly announced production ramp‑up for the B40 before retracting the statement, triggering increased market speculation.

US officials have shown concern that widespread availability of such chips—even within controlled clusters—could bolster China’s capabilities in foundational model training and inference tasks. Commerce Department Under‑Secretary Jeffrey Kessler recently told lawmakers that AI chip smuggling “is happening”, echoing longstanding export fears. Nvidia’s Huang maintains there is “no evidence” of diversion, but a consistent black‑market narrative continues to shadow the sector.

Chinese firms have outlined plans to deploy thousands of H100 and H200 chips in new data centres, particularly in western provinces such as Xinjiang and Qinghai, where abundant renewable power and cool climates offer logistical advantages. Target installations indicate upwards of 115,000 such processors in aggregated plans—most of which remain aspirational due to export prohibitions and logistical challenges.

Analysts suggest that while Nvidia’s tiered Blackwell chips may temporarily appease China’s AI infrastructure demands, the deeper impact will depend on regulatory outcomes and the strength of the CUDA ecosystem. If export restrictions ease, Nvidia stands to regain momentum; if they tighten, domestic chipmakers like Huawei will likely secure a larger foothold.

This unfolding scenario underscores a shift in semiconductor geopolitics: global tech leaders are adapting designs to fragmented access zones while local players race to fill the vacuum. The episode illustrates how export policy, market demand and strategic software assets intersect in shaping tomorrow’s AI hardware landscape.

Kraken has launched xStocks, a platform enabling users outside the United States to trade 60 tokenised U.S. stocks and ETFs around the clock, five days a week. Accessible through the Kraken app, these equities are issued on the Solana blockchain and are fully backed one‑for‑one by genuine shares held via Backed and Dinari, offering both price parity and economic rights.

The new service opens over 60 blue‑chip names—including Apple, Tesla, Nvidia and Amazon—to non‑U.S. investors, granting fractional ownership starting at just US$1, with no purchase fee in USD or Kraken’s native USDG. Trading is available 24/5 via the exchange, while withdrawal to self‑custodial wallets allows for 24/7 on‑chain usage, such as collateral in DeFi.

Kraken’s co‑chief executive Arjun Sethi described xStocks as a “foundational” shift in investment accessibility, emphasising seamless, permissionless ownership without intermediaries or geo‑barriers. The tokens are structured as SPL assets on Solana, chosen for its speed and efficiency, and dividends are auto‑reinvested into additional tokens, though users forfeit direct voting rights.

Kraken has imposed geo‑restrictions, excluding U.S. persons as well as residents in Canada, EU, UK and Australia during the initial rollout—but plans to expand to more than 185 countries in the coming weeks and eventually extend to multiple blockchains. The exchange emphasizes adherence to regulatory frameworks, supported by its partnerships with Swiss issuer Backed and U.S. SEC‑registered transfer agent Dinari.

The advent of tokenised stocks follows earlier, less‑successful attempts by Binance in 2021, which encountered regulatory pushback in multiple jurisdictions. Coincidentally, Bybit unveiled access to the same xStocks pool two hours after Kraken, signalling an intensifying push within the crypto sector. Bybit’s version supports both Solana and Ethereum versions, aimed at users seeking a bridge between traditional finance and crypto.

The global tokenisation of equities is attracting attention across the industry. Robinhood has launched tokenised stock trading in the EU in partnership with Arbitrum, enabling EU residents to trade over 200 U.S. equities 24/5. Coinbase is also pursuing approval from the U.S. Securities and Exchange Commission to offer tokenised equities domestically, pending regulatory clearance. Several firms recognise tokenisation’s potential to reduce costs, improve settlement times, and introduce continuous trading, though they acknowledge challenges surrounding liquidity and regulatory harmonisation.

Market watchers suggest tokenised equities could reshape global finance. The linkage to blockchain allows cross‑border access, seamless custody and composability in DeFi ecosystems. Critics note limitations, such as absence of voting rights and uncertain secondary‑market liquidity. The World Economic Forum and legal analysts have underlined the need for clearer disclosure standards and global regulatory coordination.

Kraken’s move appears calibrated to capitalise on a regulatory shift favouring tokenisation. It actively engages with international regulators, avoiding pitfalls that stymied predecessors. Its strategy builds on existing infrastructure—such as its listing of more than 11,000 traditional U.S. securities to U.S. clients—melding established market access with on‑chain innovation.

The launch of xStocks marks a pivotal moment in mainstreaming tokenised finance. By undercutting geographical limitations, slashing fees and uniting traditional equities with crypto rails, Kraken and peers are edging global capital markets closer to seamless, inclusive accessibility. As the ecosystem evolves, attention will turn to how regulatory authorities respond and whether liquidity deepens to support this new asset class.

A Singapore court has postponed the trial of three men accused of illegally redirecting Nvidia AI chips to China until 22 August, after prosecutors stressed the need for additional time to analyse fresh documents and obtain international cooperation. The adjournment allows police to deepen their review of evidence and reach out to overseas authorities for responses.

Charged with fraud, the defendants—Singaporeans Aaron Woon Guo Jie, 41, and Alan Wei Zhaolun, 49, along with Chinese national Li Ming, 51—stand accused of falsifying end‑user information to secure servers during purchases in 2023 and 2024. Those servers, allegedly equipped with high-end Nvidia chips, were then shipped via Singapore to Malaysia before possibly continuing to China. Political pressure surrounds the case, as the United States banned exports of leading-edge chips to China in 2022 over military and intelligence concerns. A senior U.S. official has asserted that DeepSeek, the Chinese AI firm implicated, supports military and intelligence operations.

Home Affairs Minister K. Shanmugam confirmed that Singapore authorities pursued the investigation independently after an anonymous tip-off, and preliminary findings indicate the servers may indeed contain Nvidia’s chips. The equipment, originally sourced from Dell Technologies and Super Micro Computer via Singapore‑based firms, was rerouted to Malaysia, though the final destination remains uncertain.

This case forms part of a broader probe involving 22 individuals and companies alleged to have falsified end‑user data in order to bypass export restrictions. Singapore’s position as a regional invoicing hub—recording 18% of Nvidia’s fiscal year revenues despite accounting for less than 2% of physical shipments—underscores its vulnerability as a transit point in such schemes.

Observers note that policing such complex supply chains is increasingly difficult, especially when high‑performance AI hardware carries dual-use potential with applications in advanced military or surveillance systems. Singapore’s legal actions and multilateral engagements will be closely watched as the court reconvenes late in August.

Cloud computing giants AWS, Google, Microsoft, Meta and OpenAI are accelerating in-house development of custom application‑specific integrated circuits, aiming to erode NVIDIA’s dominance in high‑performance AI datacentres. Industry reports highlight a projected annual growth rate of around 50% for ASIC purchases by hyperscalers, marking a strategic pivot in the AI hardware landscape.

NVIDIA’s premium-priced solutions—including Blackwell GPUs—have placed pressure on hyperscalers to secure more cost‑efficient, scalable systems. With single GPUs ranging from $70,000 to $80,000 and fully configured servers tallying up to $3 million, these companies are betting on internal design to manage costs and supply risks.

Amazon Web Services has notably moved ahead with its in‑house chips—Trainium for training and Inferentia for inference—reporting 30 – 40% greater cost efficiency compared with NVIDIA hardware. AWS is also collaborating with Marvell and Taiwan’s Alchip on next‑generation Trainium versions. Internal indications suggest AWS may deploy as many as half‑a‑million ASIC units in its data centres, an expansive scale‑up that could rival NVIDIA’s installed base.

Google, meanwhile, has scaled its TPU v6 Trillium chips, transitioning from single‑supplier to dual‑supplier design by partnering with MediaTek. With deployments reportedly hitting 100,000‑unit clusters to support Gemini 2.0 workloads, Google claims competitive cost-performance metrics relative to NVIDIA GPUs. Microsoft’s forthcoming Maia 200 chip, co‑designed with GUC using TSMC’s 3 nm process, is scheduled for commercial release in 2026.

Meta’s Meta Training and Inference Accelerator, developed alongside Broadcom, Socionext and GUC, is expected in early 2026 on TSMC’s 3 nm node, featuring HBM3e memory—another step towards self‑sufficiency in AI compute. OpenAI has also announced a proprietary training processor, with mass production anticipated at TSMC by 2026.

Market projections reflect this tectonic shift. ASICs are poised to claim between $100 billion and $130 billion of custom AI accelerator spend by 2030, with Broadcom estimating a market of $60 billion to $90 billion by 2027. Traditional ASIC powerhouses—Broadcom, Marvell, MediaTek, Alchip and GUC—are experiencing surging demand as they support hyperscaler transitions.

Despite these developments, hyperscalers continue to reserve capacity for NVIDIA chips, recognising the GPU giant’s entrenched ecosystem—especially its CUDA software stack—and the steep technical barriers to immediate elimination of GPU dependencies.

The trend resembles historical transitions in specialised compute. Just as cryptocurrency mining moved from GPUs to ASICs for lower costs and greater efficiency, hyperscalers now aim to fragment the AI compute supply chain and diversify their hardware portfolios.

TSMC stands to benefit significantly, serving as the foundry for both NVIDIA’s mass‑market GPUs and hyperscaler ASICs. Its chairman emphasises that the competition between NVIDIA and cloud‑designed chips is ultimately beneficial to TSMC, ensuring a broad customer base.

Broadcom has emerged as a frontrunner, with its ASIC and networking chipset revenues soaring 220% to $12.2 billion in 2024. Hyperscalers are investing in clusters featuring up to one million custom XPUs over open‑Ethernet networks—an architecture that places Broadcom and Marvell in strategic positions. Networking ASICs are expected to account for 15–20% of AI data‑centre silicon budgets, rising from the 5–10% range.

Revenue trends reflect these structural shifts. Marvell has secured a multi‑year AI chip deal with AWS and anticipates its AI silicon revenue jumping from $550 million in 2024 to over $2.5 billion in 2026. Broadcom, similarly, is redirecting significant investment toward hyperscaler ASIC demand.

Nevertheless, NVIDIA retains a commanding lead in AI training and general‑purpose GPU compute. Its end‑to‑end platform—from hardware to software—remains deeply embedded in the AI ecosystem. Custom ASICs, by contrast, offer task‑specific gains but lack the breadth of software compatibility that NVIDIA enables.

Analysts caution that the AI compute landscape is evolving toward a more fragmented, mixed‑architecture model combining GPUs and ASICs. Hyperscalers’ shift signals strategic recognition of rising costs, supply constraints, and performance demands. Yet, they also underscore persistent obstacles: software ecosystem maturity, long development cycles, and the complexity of large‑scale deployment.

Questions remain regarding the timeframe in which hyperscalers can meaningfully shift workloads away from NVIDIA GPUs. Industry roadmaps project new ASIC deployments through 2026–27. Analysts expect GPU market share erosion may begin toward the end of the decade, provided in-house ASICs deliver consistent performance and efficiency.

The stage is set for a multi‑year contest in datacentre compute. NVIDIA faces increasing pressure from hyperscalers building bespoke chips to optimise workloads and control supply. The next evolution of AI infrastructure may look less like a GPU‑centric world and more like a diverse ecosystem of specialised, interlocking processors.