AI Capital Race Intensifies as Anthropic, OpenAI, and Apple Make Major Moves

Anthropic has bet $200 billion over five years on Google Cloud, OpenAI spends $50 billion annually on computing power, and Apple paid $250 million to settle an AI-related dispute. Capital density overwhelms independent competitors.

Anthropic has signed a five-year, $200 billion cloud computing agreement with Google, becoming Google Cloud’s largest single customer commitment, accounting for over 40% of Google’s revenue backlog. A separate agreement secures multi-gigawatt TPU capacity, set to launch in 2027. Alphabet will make a reverse investment in Anthropic of up to $40 billion.

On the same day, Anthropic unveiled ten banking AI agents and Claude Opus 4.7 at its New York event, with JPMorgan Chase CEO Dimon endorsing the initiative, stating, “Every dollar spent on AI building is worth it.” Reuters reported that a joint venture between OpenAI and Anthropic is negotiating to acquire multiple AI service companies to accelerate enterprise adoption. The competition among labs has shifted from model capabilities to distribution channels, with infrastructure ties serving as entry tickets into the industry.

(Source: The Information / Reuters / Fortune / Axios)

Google, Microsoft, and xAI have signed an agreement with CAISI under the U.S. Department of Commerce’s NIST, agreeing to undergo federal security evaluations before releasing new AI models. The models are submitted for testing in a reduced defense state to better identify risks of cyberattacks and military misuse. OpenAI and Anthropic are simultaneously renegotiating existing terms to align with Trump’s AI initiative. All five leading labs are now纳入 the pre-release review system, and CAISI has completed 40 evaluations since 2024.

This has been a quiet policy reversal. After Trump took office and rescinded Biden’s executive order on AI, Silicon Valley believed federal regulation had ended. Now, regulation has returned under a new name—the framework is no longer “safety” but “national security,” and the mechanism is no longer mandatory but voluntary. The substance hasn’t changed much, but control has shifted from the academic safety community to the defense and intelligence apparatus.

(Source: NIST / Reuters / The Verge / Al Jazeera)

Coinbase laid off approximately 700 people, with CEO Armstrong attributing the decision to market downturn and increased efficiency from AI. Management layers have been reduced to five levels below the CEO, requiring all managers to also serve as individual contributors, with some teams consolidated into single-person roles combining engineering, design, and product. Severance costs amounted to $50 to $60 million, and the stock price rose that day.

Coinbase is not an isolated case. On the same day, Bloomberg reported that the crypto industry is undergoing a wave of AI-driven layoffs. PayPal announced a similar transformation, targeting $1.5 billion in annual savings. Polymarket traders are betting on more tech layoffs to come. Three-in-one single-team structures are emerging—this is not a cyclical adjustment, but a permanent shift in organizational structure.

(Source: CoinDesk / Reuters / Bloomberg / TechCrunch / CNBC)

Bloomberg reports that Apple will allow users to select third-party AI models to power Apple Intelligence features in iOS 18 this fall. Google and Anthropic are already undergoing internal testing. Third-party models will integrate into system settings as "extensions," enabling them to power Siri, writing tools, and image generation. ChatGPT’s exclusivity has ended, with actual usage falling below Apple and OpenAI’s expectations.

On the same day, Apple settled a class-action lawsuit over misleading AI features for $250 million, with iPhone 16 and 15 Pro users eligible for up to $95 per device. Apple is also considering involving Intel and Samsung in U.S. chip production to reduce reliance on TSMC, sending Intel’s stock up over 12% that day. Its AI strategy shift from in-house development to platformization essentially acknowledges its own model capabilities cannot stand alone.

(Source: Bloomberg / TechCrunch / NYT / Tom's Hardware)

Huawei expects AI processor revenue to reach approximately $12 billion in 2026, a 60% year-over-year increase. This growth is based on existing orders from Alibaba, ByteDance, and Tencent, with the Ascend 950PR entering mass production last month. Morgan Stanley estimates China’s AI chip market will reach $67 billion by 2030, with Huawei有望 to capture 60% of the share by year-end.

On the other end, Jensen Huang clearly stated that China should not be allowed to obtain Blackwell or Rubin series GPUs, asserting that the U.S. must maintain "first, most, best" in AI hardware. Export controls have reduced NVIDIA’s market share in China to zero, allowing Huawei to fill the vacuum and build a self-sustaining ecosystem; in turn, NVIDIA has turned "China shouldn’t have it" into an official stance. Decoupling is already a reality—the question now becomes what the respective efficiency ceilings are for each ecosystem.

(Source: Tom's Hardware / Digitimes / Morgan Stanley / WCCFTech)

OpenAI is spending $50 billion on computing power annually. During Musk’s lawsuit trial, co-founder Brockman testified that OpenAI’s computing expenditure will reach $50 billion by 2026, up from $30 million in 2017. The company has raised over $100 billion in funding and is preparing for an IPO that could value it at $1 trillion. (Source: Bloomberg / Quartz)

OpenAI has released GPT-5.5 Instant, reducing hallucination rates by over 50%. This model replaces GPT-3.5 Instant as the default model for ChatGPT. OpenAI states that hallucination rates have decreased by 52.5% in high-stakes domains such as medicine, law, and finance. (Source: TechCrunch / The Verge)

The SEC proposes ending the quarterly reporting requirement for public companies. This 55-year-old mandatory quarterly report would become an optional semiannual report. Trump raised this idea during his first term. Investors worry about reduced market transparency, while some companies welcome the reduced pressure to meet short-term performance targets. A 60-day public comment period has begun. (Source: SEC / WSJ / CNBC)

DeepMind’s UK employees voted to form a union, protesting military AI contracts. This will be the first union at a leading AI lab. The catalyst was Google’s agreement with the Pentagon allowing the U.S. Department of Defense to use the Gemini model within classified military networks. (Source: Fortune / Wired)

a16z Crypto raised $2.2 billion amid a downturn. This is a16z Crypto’s fifth fund, sized at half of its record-breaking fourth fund in 2022. While many crypto VCs shifted focus to AI, a16z Crypto chose to double down. CTO Lazzarin was promoted to general partner. (Source: Fortune / CoinDesk)

Meta has been collectively sued by five major publishers for copyright infringement. Macmillan, McGraw-Hill, Elsevier, Hachette, and Cengage have joined forces to sue Meta, alleging that Meta massively copied books and journal articles without authorization to train Llama. (Source: FT / The Verge)

Meta plans to conduct an internal test of its consumer AI agent, Hatch, within six months, and will launch an AI agent-based shopping tool on Instagram. The trend of AI agents moving from enterprise tools to consumer applications is accelerating. (Source: FT / Reuters / 36 Kr)

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

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