China's Nvidia Reversal

The Kill Switch

The brief détente in the US-China chip war lasted about two weeks. After months of tension, the US made a significant concession, allowing Nvidia to resume sales of its H20 AI processor to Chinese customers. This was viewed as a major step toward stabilizing a volatile relationship. Beijing's response was, to put it mildly, not what anyone expected.

Instead of rolling out the welcome mat for Nvidia's newly approved chips, Chinese authorities summoned the company's representatives to discuss "serious security vulnerabilities." It seems the truce is over before it even began. The whole episode is a bit baffling. After fighting for access to American AI hardware, why would China immediately declare it a security risk? Bloomberg reported on the sudden reversal:

The Cyberspace Administration of China said in a statement on Thursday that the chips had a serious security vulnerability and asked Nvidia staff to explain the risks and provide relevant documents. At issue are location tracking and remote shutdown capabilities of the products, according to the statement.

The move marked an abrupt change in tone after an apparent improvement in trade relations between the US and China... The particular H20 product, which had in April been banned from sale to China by the Trump administration, had only recently been cleared to return to the market by US authorities.

Nvidia, for its part, issued a swift denial. "Cybersecurity is critically important to us," the company said in a statement. "Nvidia does not have 'backdoors' in our chips that would give anyone a remote way to access or control them."

One way to look at this is that China has a legitimate concern. The idea that critical infrastructure hardware could contain a hidden "kill switch" is a reasonable fear for any government in a geopolitical standoff, a fear made more tangible by recent proposals from US lawmakers to require exactly that kind of location-tracking capability in advanced AI chips.

But there is another, more strategic way to interpret this. The American policy on these chips has been, to use a technical term, a complete mess. The H20 was banned in April, costing Nvidia billions, only to be un-banned in July. For a Chinese company trying to plan a multi-year, multi-billion-dollar data center buildout, this makes Nvidia an unreliable partner. Why would you build your future on a chip that could be outlawed again by the next presidential tweet?

This policy whiplash gives Beijing the perfect pretext to do what it has wanted to do all along: accelerate the shift to its own national champion, Huawei. US officials have already acknowledged that Huawei's homegrown Ascend AI chips are more powerful than the downgraded H20 that Nvidia is allowed to sell. The security concern, then, becomes an incredibly convenient justification for mandating a "Buy Chinese" policy. It transforms a commercial choice into a matter of national security.

The battle is no longer just about whether the US will allow its technology into China. It's about whether China, now confident in its own alternatives, even wants it. The irony is that in its quest to control the supply of AI chips, the US may have finally convinced its biggest rival that the only safe option is to build its own. China isn't just worried about a kill switch in Nvidia's hardware; it seems to be activating a kill switch on its dependency on American technology altogether.

Metaverse Déjà Vu

Mark Zuckerberg is on an audacious, multi-billion-dollar hiring spree to build a top-secret "superintelligence lab" inside Meta Platforms. For investors, however, the strategy is triggering a powerful sense of déjà vu. This isn't the first time Zuckerberg has staked the company's future on a grand, capital-intensive vision with a nebulous path to profitability. The memory of the Metaverse—a costly and commercially unsuccessful pivot that sent the stock to a six-year low—is still fresh. Now, as Meta's capital expenditure forecasts soar to as much as $72 billion for the year, Wall Street is asking a critical question: is this time any different, or is Zuckerberg leading them down another expensive rabbit hole?

The parallels to the Metaverse era are striking. Once again, Zuckerberg is pouring billions into a far-off, futuristic goal—in this case, "superintelligence"—and housing it in a clandestine unit described by one investor to the Financial Times as being like the "Manhattan Project." And once again, there is no clear business model.

“They are throwing all their cash at AI and trying to work out what to do,” said Uday Cheruvu, a portfolio manager at Harding Loevner. “What are the products you are going to build? It can’t just be a moonshot . . . What is the business model? If there isn’t a commercial plan that we as investors can see . . . then investors will get impatient.”

To be fair, there are reasons to believe this time is different. Unlike the speculative Metaverse, AI is a tangible technology that is already boosting Meta's core advertising business. And unlike the Metaverse, where Meta was largely going it alone, the entire tech industry is in a full-blown AI arms race, which helps validate the massive spending as a necessary cost of competition. Zuckerberg also earned back significant credibility with his "Year of Efficiency," giving him more leeway with investors to now go on another spending spree.

And yet, the red flags are familiar. The lab's focus on "superintelligence" is a long-term moonshot, not a near-term product. The hiring blitz is adding billions in additional annual R&D spending, which will inevitably pressure the bottom line. Perhaps most concerning are the signs of internal turmoil. The aggressive hiring has reportedly "unsettled" existing AI staff, and confidence in Meta's homegrown Llama models has reportedly waned to the point that teams are being allowed to use models from other companies. This suggests that even as Meta spends billions to acquire outside talent, its internal efforts are struggling.

With Meta set to report earnings, the pressure is mounting on Zuckerberg to articulate a vision that is more than just a vague promise of "personal superintelligence." He must convince Wall Street that this multi-billion-dollar bet is grounded in a coherent strategy with a plausible business model, and that the ghost of the Metaverse will not be making a return appearance.

The Scoreboard

• Semiconductor: Samsung’s chip profit falls 94% on US export controls on China (Nikkei Asia)
• Semiconductor: Arm sinks as chip-making ambitions, muted forecast shake investor confidence (Reuters)
• Quantum Computing: Fujitsu's new quantum computer to vie for world's most powerful (Nikkei Asia)
• Cloud Infra: Amazon’s cloud business records 18% growth in second quarter (CNBC)
• EV: CATL posts 34% first-half profit growth, defying weak EV demand worldwide and geopolitics (SCMP)