Nvidia Vera CPU sales pitch to China signals AI infrastructure shift

SpaceX IPO raises record $75B at $1.77T valuation, closes up 19% to $2.1T; debuts AI1 orbital compute satellite and tokenized trading mirrors

SpaceX's IPO priced at $135/share and raised $75 billion pre-greenshoe—the largest IPO in history—closing at $160.95 and valuing the company at approximately $2.1 trillion, making Elon Musk the world's first trillionaire. The company simultaneously unveiled AI1, an orbital data center satellite with 150 kW of compute capacity (roughly equivalent to one Nvidia GB300 rack), with plans for a Gigasat factory in Texas for mass production by late 2027, signaling serious infrastructure buildout for distributed compute and representing a major competitive entrant to ground-based AI datacenter economics. The listing demonstrated institutional demand for onchain exposure: tokenized versions from Ondo, Backpack, and Dinari traded alongside the Nasdaq listing, with Hyperliquid's SPCX perpetual futures carrying over $200 million in open interest into the opening bell. However, xStocks failed to source underlying shares for centralized exchange allocation campaigns, resulting in Binance, Bybit, and Bitget canceling their tokenized SpaceX programs and refunding $557 million in subscriptions at Binance alone—a failure that cleanly mapped the architecture: every product holding the asset onchain or through a broker-dealer delivered, while the single route depending on an intermediary to procure shares broke.

OpenAI and Anthropic file confidential IPO prospectuses at $852B and $965B valuations

OpenAI reports 900M+ weekly ChatGPT users with Goldman Sachs and Morgan Stanley managing a potential fall listing, while Anthropic filed separately for its own public offering. Rumors suggest Sam Altman may delay OpenAI's IPO to maintain better control over compute scaling while private. This marks the beginning of major AI lab public market transitions and signals confidence in unit economics despite massive infrastructure spending.

Anthropic forced to suspend Fable and Mythos models following US government national security directive

Anthropic suspended its Fable and Mythos models—released just three days prior—from all customers worldwide following a government export control order citing potential jailbreak risks. Anthropic disputes the technical legitimacy of the claim and notes similar capabilities exist in competing models like GPT-5.5. This sets a significant precedent for geopolitical risk in frontier model deployment and reinforces the strategic importance of owning inference infrastructure rather than relying on single API vendors.

SpaceX rents Colossus 1 data center to Anthropic after internal teams struggled to use it for Grok development due to latency issues

This signals potential infrastructure constraints and competitive dynamics in AI compute allocation among Musk-affiliated entities.

Nvidia Begins Vera CPU Sales Pitch to Chinese Clients; August Availability Expected

Reuters reported on Nvidia's aggressive push to sell its new Vera CPU to Chinese clients, with availability targeted for August 2026. Main takeaways:

• Vera pricing set at $20,000+ per chip before bulk discounts; fully configured 256-chip rack costs ~$10M, with Nvidia projecting $20B annual revenue by end of fiscal 2027
• At least one major Chinese cloud company planning initial order of 300+ servers (600+ Vera CPUs) for testing before committing to larger deployments
• Vera represents Nvidia's pivot from GPU dominance to CPU market as global AI workloads shift from training to inference, where CPUs face less regulatory friction than GPUs under U.S. export controls
• Chinese deployment initially limited to overseas data centers due to regulatory constraints, though CPU sales face fewer U.S. export restrictions than GPU sales (H200 licensing stalled with zero deliveries to China)
• Vera based on Arm architecture, directly competing with Intel and AMD x86 dominance in server CPU market amid reported global CPU shortage and 6-month lead times

Statement on the US government directive to suspend access to Fable 5 and Mythos 5

Anthropic published a statement on a US government export control directive suspending all access to Fable 5 and Mythos 5 models. Main takeaways:

• US government issued verbal directive citing national security, claiming discovery of a jailbreak method; Anthropic must disable both models for all customers globally, including foreign national employees
• Government's concern centers on a narrow, non-universal jailbreak (asking model to read codebase and fix flaws); Anthropic validated this capability is widely available in GPT-5.5 and other deployed models
• Anthropic disputes the standard being applied, arguing that if narrow jailbreak discovery triggers model recalls across the industry, it would halt all new frontier model deployments
• Anthropic states it conducted thousands of hours of red-teaming pre-launch with US government, UK AISI, and third parties; found no universal jailbreak; maintains defense-in-depth strategy is sound and comparable to existing deployed models
• No actual harmful results from disclosed jailbreaks; government provided only verbal evidence without written technical details or formal disclosure process