80% of U.S. AI startups use Chinese open-source models, NVIDIA CEO claims AGI has been achieved

Huang Renxun announced on a podcast, "We have achieved AGI," and 80% of U.S. AI startups are running Chinese open-source models. The technological revolution is accelerating—and so are the fractures.

NVIDIA CEO Jensen Huang made a judgment on the Lex Fridman podcast that no one in the industry dares to voice lightly. When Fridman asked, “How far are we from an AI that can found and operate a billion-dollar company?” Huang replied, “I think it’s now. I believe we’ve achieved AGI.” He cited the rise of the open-source agent platform OpenClaw as evidence, while also acknowledging that such systems may generate short-term value rather than enduring enterprises.

This statement set him apart from peers who have spent the past several months distancing themselves from the term AGI. Just as OpenAI’s contract with Microsoft includes clauses triggered by the achievement of AGI, Huang chose this moment to bring up the term. Those who announce AGI are selling GPUs—the motive needs no speculation.

In the same episode, he also defended DLSS 5, a generative graphics enhancement technology collectively criticized by the gaming community as "AI garbage," which Huang called "artist-guided optional enhancement." The juxtaposition of announcing that AGI has arrived while explaining why AI-generated content isn't garbage is the most precise snapshot of today's AI narrative.

(Source: Lex Fridman Podcast / The Verge / Ars Technica / Tom's Hardware)

A report from the U.S.-China Economic and Security Review Commission presented a striking figure: approximately 80% of U.S. AI startups are using Chinese open-source models. Models from Alibaba, Moonshot AI, and MiniMax now dominate global rankings on HuggingFace and OpenRouter. The commission warned that this is creating a "self-reinforcing competitive advantage," with an open-source ecosystem and manufacturing data forming a closed loop that allows China to approach cutting-edge capabilities even under chip restrictions.

Last week, the Cursor incident was the most concrete example. When the AI programming tool, valued at $29.3 billion, launched Composer 2 and claimed proprietary breakthroughs, developers discovered within hours via API debugging that the underlying model was Moonshot’s Kimi K2.5. The co-founder admitted that not disclosing the base model was a mistake.

Meanwhile, the Pentagon has classified Anthropic as a “supply chain risk.” Senator Warren wrote to the Secretary of Defense, calling this a “retaliatory move,” noting that the contract could be terminated without resorting to punitive labeling. The real supply chain risk lies not in Anthropic’s contract terms, but in the fact that 80% of startups depend on these models.

(Source: U.S.-China Economic and Security Review Commission / VentureBeat / TechCrunch / Reuters)

IEA Director Birol provided quantitative comparisons at the National Press Club in Australia. The two oil crises in 1973 and 1979 combined caused a global loss of approximately 10 million barrels per day; the current crisis in Iran amounts to a loss of 11 million barrels per day. Natural gas losses of about 140 billion cubic meters are nearly double those of the Russia-Ukraine conflict. At least 40 energy assets in nine Middle Eastern countries have been severely damaged. Chevron’s CEO stated more directly at CERAWeek that oil prices have “not yet fully reflected” the actual shortage.

The biggest beneficiaries of this crisis are not in the Middle East. According to CREA data, Russia earned approximately $7 billion from fossil fuel exports in the first two weeks of March. Urals crude prices surged from around $57 per barrel to nearly $100, nearly matching Brent prices and eliminating the long-standing discount. Trump’s 30-day sanctions waiver, expiring on April 11, allows countries to purchase Russian oil in transit; Treasury Secretary Bessent claimed it would not bring “significant financial benefits,” but analysts say this restriction is “nearly impossible to enforce.”

(Source: IEA / Fortune / Al Jazeera / CNBC / Guardian / CREA)

BlackRock CEO Fink placed AI at the center of the inequality narrative in his annual letter to investors. His core argument is that the vast wealth created over recent generations has primarily flowed to those who already held financial assets, and the AI boom will accelerate this trend. Without expanding access to markets, the benefits will only make the rich richer.

Fink’s solution carries a clear product logic. He proposed establishing a government retirement investment fund of approximately $1.5 trillion, operating alongside existing Social Security trust funds, while pointing to tokenization as a key tool for expanding market access. This aligns precisely with BlackRock’s most central business bet over the past two years. The manager of $11.6 trillion in assets says “get more people investing,” which translates to “get more people’s money flowing into the products I manage.”

The signals aren't just coming from BlackRock. On the same day, according to Bloomberg, JPMorgan Chase launched a new tool to help clients hedge against AI-related debt risks. When Wall Street begins pricing hedging products for the AI bubble, it won't be far from the threshold of "too big to short."

(Source: BlackRock Annual Letter / CNN / Reuters / Bloomberg)

Sam Altman has stepped down as chairman of the board at fusion company Helion Energy to allow OpenAI to negotiate a power purchase agreement as an independent buyer. According to TechCrunch, the proposed agreement would give OpenAI 12.5% of Helion’s total electricity generation, equivalent to 5 gigawatts in 2030 and 50 gigawatts in 2035.

A few weeks after Otman acknowledged that "data centers are hard" following a outage at a Texas data center, OpenAI retreated from building its own infrastructure. Now the signal is clearer: OpenAI does not intend to solve its energy bottleneck within the traditional power grid framework, but is betting on nuclear fusion that has not yet been commercialized. The 5-gigawatt target for 2030 means Helion must go from lab to large-scale power generation in less than four years—a feat no nuclear fusion company has ever accomplished.

Previously, Altman served as both chairman of Helion and CEO of OpenAI, meaning the same individual made decisions for both the seller and the buyer. His resignation was a prerequisite for advancing the deal, signaling that commercialization has progressed to the point where conflicts of interest must be addressed. With traditional energy losing 11 million barrels per day due to the Iran crisis, betting on nuclear fusion feels less like science fiction and more like a hedge.

(Source: TechCrunch / CNBC / Axios)

According to the New York Times, Trump established the "Pax Silica" fund to reduce dependence on foreign chips. This multilateral framework now includes eight countries, including the U.S., Japan, South Korea, and others, with India joining in February this year. On the same day, Musk announced that SpaceX and Tesla will build an advanced chip factory in Austin. (Source: NYT / Reuters)

The CEOs of Kalshi and Polymarket jointly invested in a $35 million venture capital fund for prediction markets, the same day the Senate introduced a bipartisan bill banning sports betting on prediction markets. The fund, named 5c(c) Capital, was founded by early employees of Kalshi. While the two companies compete fiercely on product, they agree on the赛道 bet. (Source: Fortune / TechCrunch / WSJ)

Luma AI has released the Uni-1 image generation model, outperforming Google and OpenAI on multiple benchmarks at 30% lower cost. Google’s Nano Banana series had been the undisputed leader for months, but Luma, after entering the image generation space from video generation tools, has immediately rewritten the rankings. (Source: VentureBeat)

Apple announced that WWDC 2026 will open on June 8, hinting at "AI advancements." Following last year's visual overhaul, Apple Intelligence needs to deliver on its long-delayed promise of a Siri upgrade. (Source: TechCrunch / The Verge)

The strategy has restored Bitcoin purchasing power to $4.2 billion, with $76 million in BTC purchased last week via common stock offerings. This follows a $1.6 billion BTC purchase the prior week funded by STRC preferred shares. The pace of purchases has not slowed despite market volatility. (Source: CoinDesk / Fortune)

According to Bankless, two U.S. senators reached a "principled agreement" with the White House on stablecoin yield provisions. This marks another step forward in the shift of crypto regulation from "enforcement first" to "rules first." (Source: Bankless)

Click to learn about the open positions at BlockBeats

Welcome to the official BlockBeats community:

Telegram subscription group: https://t.me/theblockbeats

Telegram group: https://t.me/BlockBeats_App

Twitter official account: https://twitter.com/BlockBeatsAsia