NVIDIA Plays Trillion-Dollar Chess Game | Rewire News Morning Edition

Publication Date: 2026.03.17 · BlockBeats

Huang Renxun previewed a number at GTC 2026 that the Blackwell plus Vera Rubin purchase order will break through 1 trillion dollars by 2027. In the same week, Meta laid off 20,000 people and signed a 27 billion dollar computing power contract. People out, racks in.

1｜Huang Renxun announced Nvidia's next era at GTC

The Vera Rubin platform was officially launched on Monday, with a 7-chip architecture, achieving a 10x improvement in inferencing performance, a 10x improvement in performance per watt, and reducing the cost per token to one-tenth. But the most critical aspect is not these numbers, but NemoClaw simultaneously released by Huang Renxun. This is Nvidia's open-source Enterprise AI Agent Build platform, with 17 companies including Adobe, Salesforce, SAP, ServiceNow, Siemens, Palantir announcing integration.

This is the moment Nvidia redefines itself from a "GPU supplier" to an "AI infrastructure operating system." Hardware lock-in was the previous war, and software ecosystem is the endgame of this war. Today's 17 companies are building walls on Nvidia's platform with code.

Huang Renxun also casually released the DGX Station, a desktop workstation capable of running trillion-parameter models, with 748GB of unified memory, 20 petaflops. The emergence of this machine means sovereign AI no longer needs data centers.

(Source: VentureBeat / Reuters / TechCrunch)

2｜OpenAI is using PE money to buy a ticket to the corporate market

According to Bloomberg and Reuters reports, OpenAI is in negotiations with TPG, Bain Capital, Advent International, Brookfield, planning to jointly establish a $10 billion joint venture company dedicated to deploying AI to PE-owned enterprises. PE firms will receive preferred stock and board seats, while OpenAI will gain access to a distribution channel into the $13 trillion PE managed assets.

OpenAI Applied Business CEO Fidji Simo calls it the "deployment arm." This term is accurate. OpenAI has an annualized revenue of $25 billion, with $10 billion coming from the enterprise side, and this joint venture is a bet on where the next $100 billion will come from. PE firms are not only buying preferred shares with their money but are purchasing a "not replaceable by AI" ticket for their own portfolio companies.

Anthropic uses common stock, OpenAI uses preferred stock, and their PE-raising and security-seeking pricing are completely different.

(Source: Bloomberg / Reuters)

3｜The Iran War is Hitting Two Weaknesses of the Chip

According to Bloomberg, the Iran War has begun to raise substantial concerns about the global chip supply chain. The chokepoint of the Strait of Hormuz affects not only oil but also the scarce chemical materials needed for chip manufacturing. On the same day, Micron announced the construction of a second memory chip factory in Taiwan, expanding production against the trend.

According to the MIT Technology Review, OpenAI has signed an agreement with the Pentagon, and its technology can now appear in classified environments. Sam Altman stated it will not be used for autonomous weapons, but the specific boundaries of the agreement remain opaque.

Laying these two signals side by side reveals the logic. AI chips are both a tool of this war and the most vulnerable supply node of this war. The war is consuming it and also threatening the ability to produce it.

(Source: Bloomberg / MIT Technology Review / WSJ)

4｜Meta Lays Off 20,000 Employees, Signs $27 Billion Computing Power Contract the Same Week

According to Reuters, Meta signed a 5-year, $27 billion AI infrastructure agreement with Nebius, and Meta's stock price rose after the layoff announcement. The timing of these two events in the same week is not a coincidence.

What Meta is doing is replacing labor costs with computing resources. The 20% reduction in workforce, about 20,000 people, is transforming the corresponding human expense into GPU lease contracts. This is not a contraction but a restructuring. Human positions are being taken over by machines, and the heads on the balance sheet are turning into racks.

Several years ago, Zuckerberg announced that 2023 would be the "Year of Efficiency." Today's round of layoffs is the most thorough execution of that statement. The difference is that year was a passive cut, while this time it is an active replacement.

(Source: Reuters / WSJ)

5｜xAI Faces Child Pornography Lawsuit and Pentagon Inquiry in the Same Week

Three minors have filed a class-action lawsuit against xAI, accusing Grok of generating their real photos into sexual content, and the plaintiffs are seeking to represent all similar victims. In the same week, Elizabeth Warren issued an inquiry to the Department of Defense, demanding an explanation of why xAI was granted access to a classified network, citing multiple harmful output records by Grok posing a potential national security risk.

Viewed individually, both of these things can be PR-ed away. But side by side, the logic breaks down. A company that can't even keep its product's security perimeter intact having access to national secrets concurrently is the question Warren asked, far sharper than she perhaps realized.

(Source: TechCrunch)

Also worth knowing ↓

PhonePe pauses IPO plans, as Tiger Global and others exit once again. Walmart's Indian mobile payment arm postpones IPO due to global market volatility from the Iran conflict. This is the second Asian unicorn this year to delay listing due to geopolitical events, extending the exit cycle for Tiger Global and Microsoft. (Source: TechCrunch / Finextra)

Frore becomes a liquid-cooled chip unicorn, valued at $16.4 billion. The 8-year-old company originally focused on solid-state cooling solutions, but, as reported by TechCrunch, a suggestion by Jensen Huang led it to pivot to liquid cooling, securing a $143 million Series D. On the day of the GTC announcement, Frore's valuation just crossed into unicorn territory, not a coincidence. (Source: TechCrunch)

LLMs self-train LLMs, now performing at human efficiency of 23.2%, up from just 9.9% nine months ago. Jack Clark's Import AI reviewed the latest results from PostTrainBench, showing AI's ability to autonomously fine-tune LLMs rapidly growing. Researchers are not concerned with the current numbers but with the pace of acceleration. (Source: Import AI / Jack Clark)

Z.ai launches GLM-5-Turbo, priced 80% lower than GPT-4o, specializing in Agent workflows. Not open-source, not vying for performance leaderboards, directly targeting the enterprise workflow market with an OpenClaw style. Chinese model makers are shifting their strategy from "score comparison" to "scenario comparison." (Source: VentureBeat)

OpenClaw evades EDR, DLP, IAM, with zero alerts throughout. Six independent security teams validated the same vulnerability, exfiltrating credentials through a legitimate API call, firewall logging HTTP 200, with no signature trigger. AI Agent's security perimeter issue is now entering mainstream security awareness in a reproducible manner. (Source: VentureBeat)

Apple acquires video effects software company MotionVFX. Apple is making a strong push against Adobe Creative Cloud, bringing professional visual effects tools to the Final Cut Pro ecosystem. The acquisition price was not disclosed. (Source: TechCrunch)

Argentinian investigation finds $5 million payment linked directly to Milei's LIBRA Memecoin event. Leaked documents show that Hayden Davis structured this payment around the time of Milei's promotion of LIBRA, marking a key piece of the crypto community's largest political scandal. (Source: Bankless)