Nvidia Denies China LPU Chip; Broadcom Flagged as Better Bet Than

Nvidia is heading into a high-profile earnings week with a fresh distraction: a public denial over a China chip report. Meanwhile, the Broadcom-versus-Marvell debate is heating up again in the wake of last week’s big Google deal news.

Key points

• Nvidia denied a report by The Information that it planned to ship a China-specific language processing unit (LPU) in small batches by year-end, calling the claim inaccurate.
• CEO Jensen Huang previously acknowledged in May that Nvidia has “largely conceded” China’s AI chip market to Huawei, making the denial broadly consistent with its stated posture.
• Broadcom is being flagged as the stronger buy over Marvell, with a forward PEG of 0.66x and a FY3 PE under 14x, reflecting significant expected earnings expansion.
• Nvidia itself trades at $5.45 trillion in market cap, roughly 22x forward earnings, with annual revenue growth running at 85%.
• Fort Robotics, backed by Mark Cuban, announced a $557 million SPAC deal to become the first publicly traded company focused on physical AI safety.

What is Nvidia actually doing in China?

The denial is worth parsing carefully. Nvidia told reporters on August 20 that The Information’s report about a China-tailored LPU was incorrect. The company did not provide further detail on what, if anything, it is developing for the Chinese market.

The timing is awkward. Nvidia’s earnings are expected imminently, making any ambiguity about a major market particularly sensitive. Huang’s May comments about conceding ground to Huawei remain on the record, and the denial does nothing to change the underlying reality that U.S. export controls have largely cut Nvidia out of China’s most advanced AI infrastructure buildout. For investors, the practical takeaway is that China is not a near-term revenue catalyst for Nvidia, regardless of what any single report claims.

Is Broadcom the smarter custom-chip play right now?

Last week’s coverage focused on Marvell’s $12.2 billion Google chip win. That deal is real and meaningful, but analysts at Seeking Alpha argue that chasing Marvell after that headline is a mistake when Broadcom is sitting at a cheaper multiple with a comparable or superior earnings growth trajectory.

The core numbers: Broadcom at a forward PEG of 0.66x and a FY3 PE under 14x suggests the market is pricing in less optimism than the fundamentals may support. Marvell, by contrast, has likely had some of its Google upside priced in following the deal announcement. This is a relative-value argument, not an absolute one. Both companies are real beneficiaries of hyperscaler demand for custom AI silicon. The question is which one the market is paying less for that exposure.

Broadcom’s diversified revenue base, spanning networking, storage controllers, and its VMware software business, gives it a different risk profile than Marvell. Investors who already own Marvell and are sitting on gains from last week’s pop might weigh whether rotating into Broadcom captures more of the remaining upside.

Fort Robotics and the physical AI IPO wave

Away from chips, Fort Robotics is set to go public via a $557 million SPAC merger, positioning itself as a pure-play on physical AI safety. The company provides safety hardware and software for autonomous robots, with more than 600 customers including Google DeepMind and Zoox. Mark Cuban is among its backers.

The deal is notable because most AI investment discussions still center on software models, data infrastructure, and chips. Fort Robotics is betting that as autonomous machines move into warehouses, roads, and industrial sites, safety and trust frameworks become a chokepoint that commands real commercial value. The SPAC structure is familiar territory for speculative growth stories, and investors should treat the $557 million valuation as a starting point for scrutiny rather than a validated figure. Still, the customer list lends some credibility to the commercial traction claim.

SpaceX: unlock pressure meets spectrum ambition

SpaceX continues to generate investor headlines on multiple fronts. The stock has drifted back near its $135 IPO price as insider share unlocks proceed, creating the kind of technical overhang that often precedes short-term volatility. University endowments that received large SpaceX allocations pre-IPO now face a portfolio concentration problem, with some likely to trim regardless of their long-term view on the company.

At the same time, SpaceX and AST SpaceMobile are reportedly seeking Grain spectrum assets valued at $6 billion. Spectrum is the underlying resource for Starlink’s satellite broadband ambitions, so this pursuit is strategically coherent even if the price tag is large. SpaceX is growing fast but, as reporting notes, spending even faster, which means free cash flow discipline will be a key question for investors watching the stock near its IPO floor.

Nothing in this article constitutes investment advice. All figures are sourced from third-party reports published on or before August 21, 2026.