Is Nvidia Selling AI Chips to China Again?
Yes — and it's a big deal. Nvidia CEO Jensen Huang confirmed at GTC 2025 that the company is restarting production of AI chips for sale in China. Specifically, Nvidia has received licenses for multiple Chinese customers, collected purchase orders, and has its supply chain actively ramping back up. The chip in question is the H200 processor — not a nerfed, China-specific version, but the real deal, one generation behind Nvidia's most powerful current GPUs. This reverses years of uncertainty and marks a significant pivot in US semiconductor export policy toward China.
Jensen Huang put it plainly at GTC: "We have been licensed for many customers in China. We've received purchase orders from many customers, and we're in the process of restarting our manufacturing. Our supply chain is getting fired up." For a company valued at $4.44 trillion, that's not a footnote — that's a seismic shift.
Why Were Nvidia Chips Banned From China?
The chip ban on China originated in 2022 under President Biden's CHIPS Act. The logic was straightforward on the surface: AI is a critical technology, the US wants to maintain its lead in the AI race, and fewer advanced chips flowing to China means more competitive advantage for American companies and a faster domestic AI buildout.
At the same time, the CHIPS Act unlocked billions in incentives for US-based chip manufacturing — so it wasn't purely restrictive. It was a dual-track strategy: slow China down while accelerating America's own capacity.
The narrative, however, kept shifting. Nvidia designed a special China-market chip called the H20 — essentially a nerfed H200 — to comply with export rules while still serving Chinese customers. But even that got halted. Last April, the Commerce Department blocked H20 exports after reports emerged that the team behind DeepSeek had figured out how to use those chips nearly as effectively as the more powerful versions. The line between "safe to sell" and "dangerously capable" kept blurring.
Then, in December, the US government approved Nvidia to sell the H200 to China — as long as Nvidia shares 25% of its China sales revenue with the US government. Effectively, an export tariff. As of GTC, Nvidia had received that approval but hadn't yet generated any revenue from it. Now, that's changing.
What Is the H200 Chip Nvidia Is Shipping to China?
The H200 is one generation behind Nvidia's current flagship GPUs — the Blackwell B200 series. It's a serious piece of hardware used for training and running large AI models. Think of it as the previous generation of the chips powering today's most advanced AI systems.
The earlier China-specific H20 was deliberately limited in its memory bandwidth and compute to stay under export thresholds. The H200 carries no such restrictions — it's the full-power version. This is a significant upgrade for Chinese AI labs that had been working around limitations or relying on stockpiled hardware.
Cerebras CEO Andrew Feldman used GTC to fire shots at Nvidia's broader inference chip strategy, arguing that Nvidia's approach — connecting thousands of discrete chips over cables — creates unavoidable latency bottlenecks. Feldman claims Cerebras' wafer-scale architecture sidesteps this entirely, delivering dramatically higher tokens-per-second in production. Whether or not you take the competitive framing at face value, it underscores just how intense the inference chip race has become — and how much the Chinese market matters to everyone competing in it.
How Do Chip Exports Affect the Taiwan Conflict Risk?
This is where the geopolitics get genuinely interesting. One of the strongest arguments against a blanket chip ban on China was always the Taiwan problem — and it cuts in a counterintuitive direction.
The reasoning goes like this: China depends on TSMC, Taiwan's semiconductor giant, to manufacture advanced chips. As long as China needs TSMC's fabs running, it has a massive economic incentive not to invade Taiwan. TSMC's manufacturing is extraordinarily precise — a single rocket strike, or even a significant earthquake, could knock out production. The employees are so attuned to the fabs that they reportedly don't need a push notification during an earthquake; they just get up and go to the factory.
But here's the problem: if you cut China off from TSMC chips entirely, you remove that deterrent. If China can't access advanced chips anyway, the economic cost of a Taiwan invasion drops dramatically. The harder the chip ban, the less China has to lose.
With global conflicts expanding, the US military potentially stretched thin, and a chip shortage expected to last through at least 2030, keeping China economically entangled with Taiwan's semiconductor ecosystem may actually be the more stable play. The argument for selling some chips to China isn't just about Nvidia's revenue — it's about maintaining the interdependence that makes conflict irrational.
Are Vibe Coding Apps Getting Banned From the App Store?
Apple has quietly moved to block AI-powered vibe coding apps — including Replit and Vibe Code — from releasing updates on the App Store unless they make specific modifications. The issue, according to Apple, is that these apps violate long-standing rules prohibiting apps from running code that changes how they or other apps function after approval.
The timing is notable. Vibe coding apps have been fueling an explosion of new app submissions to the App Store, reportedly slowing down Apple's review process. More critically, these tools help developers create web apps that bypass the App Store entirely — a direct threat to one of Apple's most important revenue streams.
In the App Store right now, Replit ranks in the top three of DevTools with over 14,000 reviews. Vibe Code markets itself as a no-experience-needed website builder. The category is real, it's growing, and Apple clearly sees it as a structural threat. The core tension: developers can generate a functioning app inside Replit's mobile app and preview it in ways that Apple didn't formally approve — essentially using one approved app to run unapproved software.
The broader question is whether Apple will find a way to enable peer-to-peer app sharing (something TestFlight gestures at but doesn't fully solve) or whether the crackdown will push vibe coding entirely to the web and desktop.
Is Private Credit at Risk From AI Disrupting Software?
Private credit has grown into a $3 trillion asset class by some estimates — a parallel banking system that's been quietly funding software company buyouts and growth deals for over a decade. Now, with AI compressing software margins and disrupting SaaS business models, some of those deals are going underwater.
The doomer framing compares this to the lead-up to the 2008 financial crisis. But analyst Eric Sufer pushes back in Mobile Dev Memo, arguing the comparison is structurally wrong. Unlike mortgage-backed securities, private credit funds operate on longer time horizons and don't trigger the kind of simultaneous bank runs that cause systemic collapse. If anything, the productivity gains from AI may more than offset localized losses in software-focused credit portfolios.
The key distinction: this is more likely a rolling dislocation — deals going bad fund by fund over years — than a cliff-edge crisis. That's cold comfort if you're a GP whose 2019 vintage software deal is getting repriced, but it's a very different animal from a global financial meltdown.
Are Peptides Actually Safe? Martin Shkreli Weighs In
In a viral post, Martin Shkreli — yes, that Martin Shkreli — delivered a surprisingly rigorous takedown of the peptide optimization trend sweeping tech and wellness circles. His core argument: most people talking about peptides have no idea what they're actually talking about.
Peptides are small proteins that have been used as pharmaceuticals since the 1950s. They have extremely short half-lives — often seconds to minutes — and without knowing a drug's specific target, binding mechanism, and pharmacokinetics, you don't have a drug. You have a guess. Shkreli is particularly pointed about BPC-157, a popular peptide in biohacking communities, noting it has no meaningful clinical evidence. His concern — shared by others — is that growth-stimulating peptides could accelerate cancer growth in ways that won't be visible for years.
The counterargument from the community is that many popular peptides are in active clinical trials (like Retatrutide, currently in Phase 3) and that the evidence base is stronger than Shkreli acknowledges. A live debate between Shkreli and Max Marchetti from Superpower is scheduled for Monday at 12pm Pacific — which should be worth watching regardless of where you land on the science.
