Yazılar

Trump Suggests Allowing Scaled-Down Nvidia AI Chips Sales to China Amid Security Concerns

/in /tarafından

U.S. President Donald Trump signaled on Monday that he may permit Nvidia to sell a reduced-performance version of its next-generation Blackwell AI chips in China, raising alarms in Washington about Beijing’s potential access to cutting-edge computing power.

Trump told reporters that Nvidia CEO Jensen Huang had discussed a “somewhat enhanced-in-a-negative-way Blackwell,” explaining it would carry 30–50% less computing power than the U.S. flagship model. “That will be an unenhanced version of the big one,” Trump said, suggesting the proposal remains under review.

The Trump administration also confirmed an unprecedented deal with Nvidia and AMD that requires them to give the U.S. government 15% of revenue from China chip sales. The move follows last month’s approval for Nvidia’s H20 chips, a lower-performance variant developed under Biden-era export restrictions, to resume shipping to China.

Security experts warn the policy could still advance Beijing’s AI capabilities. Saif Khan, a former White House tech security director, cautioned that China could buy enough scaled-down Blackwell chips to build frontier-level AI supercomputers, potentially leapfrogging U.S. progress.

Currently, the H20 is the most advanced chip allowed for Chinese markets, though Trump called it “obsolete,” noting Beijing already has access. Nvidia unveiled its flagship Blackwell chip in March, boasting performance up to 30 times faster than its predecessor.

While Nvidia has not confirmed a China-only Blackwell variant, Reuters reported in May that a lower-cost, scaled-down version was being prepared. China’s foreign ministry did not immediately comment.

Meanwhile, critics highlight Trump’s unusual interventions in corporate strategy, from demanding revenue-sharing to pressuring Intel’s CEO Lip-Bu Tan to resign over his ties to Chinese firms. The Commerce Department has begun issuing licenses for H20 exports, insisting national security remains safeguarded.

Nvidia and AMD both stated they will comply with U.S. rules, while China has accused Washington of using tech restrictions to “maliciously contain and suppress” its development.

Tesla Refocuses AI Chip Development, Elon Musk Confirms Shift Away from Dojo Supercomputer Team

/in /tarafından

Tesla CEO Elon Musk announced that the company will streamline its AI chip research to concentrate primarily on developing inference chips designed to run AI models and enable real-time decision-making. This follows reports that Musk ordered the disbandment of the in-house Dojo supercomputer team, with its leader, Peter Bannon, leaving the company.

The Dojo supercomputer, built around custom training chips, was originally created to process vast data from Tesla electric vehicles to train its autonomous driving software. Musk stated on X that it no longer makes sense for Tesla to split resources between two distinct AI chip designs. Instead, all efforts will now focus on Tesla’s AI5, AI6, and subsequent chips, which are optimized for inference tasks and still capable of training AI models effectively.

Analysts, including Morgan Stanley’s Adam Jonas, had previously valued the Dojo supercomputer at $500 billion in 2023, viewing it as a key growth driver for Tesla beyond vehicle sales, comparable to Amazon’s cloud business. It remains unclear how this restructuring will impact Tesla’s valuation.

Industry-wide, tech companies are consolidating custom chip development to reduce latency, power consumption, and costs while focusing on fewer architectures. Tesla’s recent restructuring includes executive departures, job cuts, and a strategic pivot toward AI-driven self-driving technology and robotics, with Musk aiming for synergy across his technology ventures.

Musk has announced plans for next-generation AI5 chips targeted for production by the end of 2026 and revealed a $16.5 billion contract with Samsung Electronics to supply AI6 chips. These chips are expected to power Tesla’s autonomous vehicles and Optimus humanoid robots, with potential for broader AI applications due to their substantial compute capabilities.

According to Bloomberg, around 20 Dojo team members have already left to join the startup DensityAI, while remaining staff are being reassigned within Tesla to other compute and data center projects.

AMD and Super Micro Shares Slide as AI Growth Expectations Dim After Data Center Results Miss

/in /tarafından

Shares of Advanced Micro Devices (AMD.O) and server maker Super Micro Computer (SMCI.O) fell sharply in early trading Wednesday after both companies reported weaker-than-expected results in their data center segments, casting doubt on their AI growth prospects and competitive standing. AMD shares dropped 5.1%, while Super Micro plummeted 18.2%, with the latter potentially losing over $6 billion in market value.

AMD’s data center revenue, driven by Instinct AI chips and server CPUs, grew 14% to $3.2 billion in Q2—slightly below analyst forecasts—and lagged far behind rival Nvidia’s 73% jump to $39.11 billion in the same segment. Jefferies analysts said the AI outlook failed to deliver the strong upside some investors anticipated. CEO Lisa Su cited U.S. export restrictions on AMD’s MI308 AI chips to China as a factor in year-over-year AI revenue declines, with no clear timeline for lifting those limits. HSBC noted that expectations for revenue recovery from lifting export restrictions appear muted.

The chip sector faces additional risks from impending U.S. tariffs on semiconductor imports and supply chain vulnerabilities tied to Taiwan Semiconductor Manufacturing Company (TSMC), which produces AMD’s advanced 3-nanometer wafers. Analyst Michael Ashley Schulman warned any slowdown at TSMC could disproportionately impact AMD.

Super Micro missed Q4 estimates amid intense competition from larger server makers Dell and HP, suffering from execution issues and Nvidia chip supply delays. Analyst Gil Luria of D.A. Davidson noted signs of market share loss. While over 70% of Super Micro’s Q4 revenue is linked to AI platforms, the company’s margins remain pressured by high AI server production costs and fierce rivalry. Bank of America analysts highlighted ongoing uncertainty over the gross margin impact this quarter.

Super Micro CEO Charles Liang expressed optimism about improved chip availability driving better growth going forward. Dell’s shares also declined 1.7% on the news.

AMD trades at a forward price-to-earnings multiple of 32.39, compared to Super Micro’s 19.69.