Arm Steps Into Chip Manufacturing Itself | Rewire Morning News

marsbitPublicado em 2026-03-25Última atualização em 2026-03-25

Resumo

Arm, after 35 years of only licensing its architecture, has begun manufacturing its own data center CPU, the 136-core Neoverse V3, with Meta as its first major client. This strategic shift aims to capture a larger market share and achieve significant revenue growth. In a separate legal battle, Anthropic is challenging the Pentagon's decision to ban federal agencies from using its AI, a move the presiding judge called "disturbing." Conversely, Palantir's military AI platform received a massive $13.4 billion funding commitment from the Pentagon, highlighting a stark divergence in the U.S. government's approach to AI militarization. Four tech billionaires presented conflicting visions on AI's impact. Perplexity's CEO and Palantir's CEO suggested AI-induced job loss is beneficial, while an OpenAI investor predicted 80% of jobs could be replaced by 2030. In contrast, BlackRock's CEO warned the greater threat is AI exacerbating wealth concentration. Circle's stock plummeted 18% due to proposed legislation threatening its business model, while Tether announced its first full audit by a Big Four firm, signaling a push for legitimacy amidst tightening regulations. OpenAI shut down its Sora video app, leading Disney to withdraw a $1 billion investment. An investor filing also revealed OpenAI now lists its major shareholder, Microsoft, as a top risk factor, indicating a strategic cleanup ahead of a potential IPO. Other notable news includes SK Hynix's record $8 billion order for EU...

For the first time in 35 years, Arm is making its own chips, while four billionaires presented four different prophecies about AI and employment on the same day.

1| Arm No Longer Just Draws Blueprints; The Landlord Is Building the House Itself

Arm has always only drawn blueprints and collected licensing fees. For 35 years, companies like Qualcomm, Apple, and Nvidia have used the Arm architecture to design chips, with Arm collecting a royalty for each processor sold. This business model made Arm the safest "rent collector" in the chip industry. Now, the landlord is building the house itself.

On March 24, Arm released its first self-developed data center CPU, the 136-core Neoverse V3 architecture, with a thermal design power of 300 watts, targeting AI inference and training workloads. Meta is the first customer, with OpenAI, Cerebras, and Cloudflare also signed on. CEO Rene Haas set a clear financial goal: by 2031, this chip is expected to bring in $15 billion in annual revenue, with Arm's total revenue reaching $25 billion.

This is not a hardware "trial" by a design company. Arm's data center CPU market share has risen from about 15% in 2024 to 21% currently, with a target of reaching 50% by the end of the year. When Amazon's Graviton, Google's Axion, and Microsoft's Cobalt are all using Arm architecture to develop their own chips, Arm's choice is: rather than waiting for super customers to bypass you, become their supplier directly. For Intel and AMD, another segment of the x86 moat has been lost.

(Source: CNBC / TechCrunch / Reuters / Tom's Hardware)

2| Anthropic Takes the Pentagon to Court, While Palantir Secures $13.4 Billion on the Same Day

Anthropic refuses to relax AI safety guardrails for the military. The Pentagon's response is to list Anthropic as a "national security supply chain risk," prohibiting all federal agencies from using its technology. This is the first time in U.S. history that this procurement regulation, originally targeting foreign spies, has been used against a domestic company.

On March 24, Anthropic filed for an emergency injunction in federal court in San Francisco. Presiding Judge Rita Lin said something in court that quieted the gallery: "I'm not sure if this is murder, but it looks like an attempt to paralyze Anthropic." She described the Pentagon's actions as "disturbing," noting a lack of reasonable connection between the restrictions and the alleged national security reasons. Judge Lin is expected to rule within the next few days.

On the same day, Palantir's Maven AI platform was officially confirmed by the Pentagon as a core military system, receiving a multi-year funding commitment. The budget soared from $480 million in 2024 to $13.4 billion. Those who refused to cooperate were blacklisted; those who fully cooperated received a 28-fold budget increase. This is not just a divergence in the fate of two companies; it is an institutional split in the path of AI militarization, and the Pentagon has voted with its money.

(Source: CNBC / Al Jazeera / Axios / Tom's Hardware)

3| Four Billionaires, One Day, Four Prophecies About AI and Jobs

In a single day, four people controlling the direction of the AI industry gave four contradictory prophecies.

Perplexity CEO Srinivas said on the All-In podcast that AI-induced unemployment is a good thing: "Most people already hate their jobs." Palantir CEO Karp was more direct: in the AI era, only two types of people will survive—blue-collar workers with skills, or the neurodiverse.

OpenAI investor Khosla gave a number at a Washington forum: 80% of jobs will be replaced by AI before 2030. He predicted that "fear of AI" will be the top issue in the 2028 election.

BlackRock CEO Fink expressed a different concern in his annual letter to shareholders. The real threat of AI is not unemployment but wealth concentration. Since 1989, the median wage increase in the U.S. has lagged behind stock market returns by 15 times, with the top 1% holding 31.7% of U.S. wealth. Four narratives, one conclusion: the existing social contract is breaking down.

(Source: Fortune / BlackRock Annual Letter to Shareholders / Axios)

4| Circle Crashes 18%, Tether Hires a "Big Four" Firm; Crypto Institutionalization and Being Institutionalized Happen Simultaneously

Circle's stock price plummeted 18% in a single day. The trigger was a clause in a new draft of the Clarity Act that prohibits stablecoin issuers from distributing "economically equivalent to interest" yields to holders. This directly strikes at the core of Circle's business model. USDC's growth relies on a profit-sharing mechanism in partnership with Coinbase, and reserve interest income is key to user retention. If this path is blocked, the competitive advantage of stablecoins relative to money market funds is significantly reduced. Circle had risen 170% over the previous two months, and the vulnerability of a high pullback was exposed by this legislative draft.

On the same day, Tether announced it had hired a "Big Four" accounting firm to conduct the first comprehensive audit of USDT reserves. The specific firm was not disclosed. Tether had previously relied only on periodic attestations from BDO Italia for years, never undergoing a full audit. Also on the same day, the NYSE partnered with Securitize to develop a 24/7 tokenized securities platform. Regulation is tightening, institutions are laying pipelines—both happened on the same day.

(Source: CoinDesk / Bankless / WSJ)

5| OpenAI Shuts Down Sora, Loses Disney's $1 Billion, and Lists Microsoft as Top IPO Risk on the Same Day

OpenAI announced the shutdown of its AI video application Sora. This TikTok-style short video product, launched last October, received good technical reviews but failed to maintain user retention. The shutdown directly led Disney to withdraw a $1 billion investment plan. The original agreement was a three-year license allowing the use of over 200 characters from Marvel, Pixar, and Star Wars to generate videos. OpenAI stated the team would shift to "world simulation and robotics research."

In investor documents exposed on the same day, OpenAI listed Microsoft as its top risk. Microsoft invested $13 billion for a 27% stake and is the exclusive training cloud supplier, but OpenAI has signed a $50 billion exclusivity agreement with AWS. Microsoft believes this violates exclusive terms. Shutting down loss-making products, cutting licensing burdens, and listing the largest shareholder as a risk factor—these are not isolated decisions but a systematic narrative cleanup ahead of an IPO.

(Source: TechCrunch / Variety / CNBC / 24/7 Wall St.)

Also Worth Knowing ↓

SK Hynix placed an $8 billion order with ASML for EUV lithography machines, about 30 units, setting a record for ASML's largest single order. The equipment will be delivered by the end of 2027, deployed at the new Yongin factory and the Cheongju M15X factory, serving HBM and advanced DRAM production. AI demand for memory chips is pushing equipment manufacturers' order books to the limit. (Source: Tom's Hardware / TrendForce)

Huawei released the Atlas 350 AI accelerator, with FP4 computing power of 1.56 PFLOPS, equipped with up to 112GB HBM, claiming performance is 2.8 times that of Nvidia's H20. On the same day, U.S. senators from both parties sent a joint letter to the Commerce Department, stating that Jensen Huang's claim that "chips were not transferred to China" "contradicts existing reports" and demanding a suspension of Nvidia's export licenses to China. Blockade on one side, accelerated substitution on the other. (Source: Tom's Hardware)

The President of the Philippines declared a national energy emergency for one year. The Middle East war has led to an "imminent energy supply danger," and the government has established an emergency committee to manage the distribution of fuel, food, and medicine. (Source: Fortune)

Qatar Energy declared force majeure on LNG contracts with South Korea, China, Italy, and Belgium. 17% of production capacity was destroyed by Iranian strikes, with repairs taking 3 to 5 years. The Strait of Hormuz has been effectively closed for nearly a month, blocking about one-fifth of global oil and LNG transportation. (Source: Al Jazeera)

Tenpoached several core members from ByteDance's Seed team, including visual AI platform head Xiao Xuefeng and Infra team member Zhang Chi, reporting to chief AI scientist Yao Shunyu. Since the end of 2025, Tencent has continuously absorbed large model infrastructure talent from ByteDance, with core members in the AI Infra Department and Large Language Model Department having ByteDance backgrounds. (Source: 36Kr)

Epic Games laid off over 1,000 people, with CEO Tim Sweeney stating that Fortnite engagement has been declining since last year. This is the second round of large-scale cuts after laying off 800 people in 2023, along with a $500 million reduction in contracts and marketing expenses. (Source: TechCrunch / The Verge)

Broadcom warned that TSMC's capacity has become a bottleneck for AI chip supply. AI accelerator demand continues to exceed wafer foundry capacity expansion, and supply constraints are difficult to alleviate in the short term. (Source: Reuters)

Perguntas relacionadas

QWhy is Arm's decision to build its own data center CPU chips a significant shift from its traditional business model?

AArm has historically only licensed its architecture to other companies, earning royalties per chip sold. By building its own data center CPU, Arm is directly entering the hardware market, aiming to capture more value and compete with major clients who are designing their own Arm-based chips, rather than risk being bypassed.

QWhat was the legal and financial outcome for Anthropic and Palantir in their dealings with the Pentagon on the same day?

AAnthropic was labeled a 'national security supply chain risk' and banned from federal use after refusing to relax AI safety measures for the military, leading to a court challenge. Conversely, Palantir's Maven AI platform was confirmed as a core military system and received a multi-year funding commitment, with its budget soaring from $480 million to $13.4 billion.

QWhat were the four contrasting predictions about AI and employment made by billionaires on the same day?

APerplexity's CEO said AI unemployment is good as many people hate their jobs; Palantir's CEO claimed only skilled blue-collar workers or neurodiverse individuals will thrive; OpenAI's investor predicted 80% of jobs will be replaced by AI by 2030; and BlackRock's CEO warned the real threat is wealth concentration, not job loss.

QWhat regulatory and institutional developments affected Circle and Tether in the crypto space on the same day?

ACircle's stock fell 18% due to a draft clause in the Clarity Act that could prohibit stablecoin issuers from distributing interest-like收益, threatening its business model. Meanwhile, Tether hired a Big Four firm for its first full audit of USDT reserves, and the NYSE partnered with Securitize to develop a 24/7 tokenized securities platform, showing regulatory tightening and institutional adoption happening simultaneously.

QWhat strategic moves did OpenAI make regarding Sora and its relationship with Microsoft, and what were the implications?

AOpenAI shut down its Sora AI video app, leading Disney to withdraw a $1 billion investment plan, and listed Microsoft as its top risk factor in investor documents due to conflicts over an exclusive cloud agreement with AWS. These actions are seen as part of a systematic narrative cleanup ahead of a potential IPO, involving discontinuing unprofitable products and addressing shareholder risks.

Leituras Relacionadas

A 40-Year-Old Securities Reconciliation Dilemma: Could ERC-8056 Be the Final Answer?

For four decades, the global securities industry has grappled with an inefficient and costly process for handling corporate actions like dividends and stock splits, with an annual cost estimated at $58 billion. The core problem stems from a fragmented system involving multiple intermediaries—transfer agents, the Depository Trust Company (DTC), custodian banks, and brokerages—each maintaining separate, non-interoperable databases. This requires repetitive manual calculations and costly post-hoc reconciliation for every event. The inefficiency persists due to non-machine-readable data formats (like PDFs), a lack of incentive for issuers to change, and the dominance of data vendors who profit from manual processing. While technological solutions like XBRL exist, industry inertia has prevented widespread adoption. A potential solution is emerging through blockchain technology and a new token standard, ERC-8056. Unlike "mirrored" tokenized assets that add another layer to the old system, ERC-8056 enables native on-chain issuance, where the blockchain itself becomes the authoritative shareholder ledger. This standard can automate corporate actions through smart contracts—for example, adjusting displayed balances for a stock split without minting new tokens or distributing dividends directly and instantly to all holders—eliminating the need for multi-layered reconciliation. Companies like Superstate, which is a registered transfer agent, are pioneering this model. This approach aligns with the concept of triple-entry bookkeeping, creating a single, shared source of truth. Recent regulatory developments, including actions by the DTCC and Nasdaq, signal growing acceptance. The transformation may not require a crisis to drive change; instead, by building new, efficient systems for native on-chain securities, the industry can render the costly legacy reconciliation processes obsolete. The pace of adoption now hinges on regulatory clarity and issuer willingness to embrace this new paradigm.

marsbitMesmo agora

A 40-Year-Old Securities Reconciliation Dilemma: Could ERC-8056 Be the Final Answer?

marsbitMesmo agora

Show me 'The Lord of the Rings', Karpathy Recommends New Benchmark for Large Model Evaluation

In a new benchmark for evaluating large language models, Andrej Karpathy proposes replacing the once-popular "pelican riding a bicycle" SVG test with a more complex challenge: generating a 3D scene from the opening text of *The Lord of the Rings*. Using Anthropic's Opus 5 model and the Three.js library, the task consumed approximately 1 million tokens, 2 hours, and 5,500 lines of code to produce a rudimentary, low-polygon animation of the Shire. While the output is visually crude with notable glitches like floating characters, it demonstrates the model's ability to parse narrative text and translate it into a functional, programmatic 3D world with defined objects, cameras, lighting, and basic animation. This "Lord of the Rings benchmark" is argued to test a model's capacity for long-horizon project planning, spatial reasoning, and maintaining consistency across thousands of code lines—capabilities not fully captured by simpler single-output tests. The initiative has sparked community experimentation, with users generating other 3D worlds like a low-poly San Francisco, a data-driven New York City model, and even a virtual Kanye West concert. Karpathy suggests a future pipeline where code-generated scenes provide the structural "bones" for video-to-video models to enhance visual fidelity. While some debate the computational cost and specificity to Three.js, proponents see it as a test of a model's general ability to structure its understanding of the world into an executable form. The shift signals a move towards evaluating how well models can not only generate code or images but also comprehend and construct interactive, multi-element digital environments.

marsbitHá 17m

Show me 'The Lord of the Rings', Karpathy Recommends New Benchmark for Large Model Evaluation

marsbitHá 17m

Kioxia's Profit Margin Approaches 80%, J.P. Morgan Raises Its Target Price to 155,000 Yen

According to a JP Morgan report, Kioxia's target price has been raised to ¥155,000, following record-breaking Q1 FY2026 results and the announcement of a framework for up to ¥800 billion in share buybacks. The bank's optimism is based on a convergence of data center SSD price increases, rising profitability, and shareholder returns, rather than simply higher NAND shipments. Kioxia's Q1 results showed revenue of approximately ¥1.77 trillion, up 415.5% year-on-year, with a non-GAAP operating margin of 75.0%. Even stronger, the Q2 guidance forecasts revenue of ~¥2.39 trillion and a non-GAAP operating margin of ~79.5%. This surge is primarily driven by significant ASP growth in enterprise and data center SSDs, fueled by generative AI-related demand, alongside improved product mix and advanced node adoption (e.g., BiCS 8 FLASH). The ¥155,000 target price is derived from FY2027 EPS estimates and a ~11x P/E multiple, above the historical sector average. This premium reflects reduced selling pressure from Bain Capital and the potential for long-term agreements to stabilize earnings. A key future catalyst is the potential for agentic AI to create new NAND workloads, supporting demand beyond the current cycle. While the massive share buyback plan signals capital return commitment and helps ease concerns about cyclical overspending, risks remain. The sustainability of SSD price hikes, the actual scale of incremental AI-driven demand, and the industry's ability to maintain capital discipline to avoid a new supply glut by 2027 are critical factors for the stock's continued re-rating.

marsbitHá 20m

Kioxia's Profit Margin Approaches 80%, J.P. Morgan Raises Its Target Price to 155,000 Yen

marsbitHá 20m

Claude Solves Five-Year Unsolved Bug in Just 8 Minutes

Claude Identifies Five-Year-Old Coldcard Wallet Bug in 8 Minutes A critical vulnerability in the Coldcard hardware wallet, undiscovered for five years despite multiple code audits, was reportedly identified by Anthropic's Claude AI in just eight minutes. The flaw, introduced in a 2021 code update, inadvertently weakened private key generation by switching from a hardware-based true random number generator to a weaker software-based fallback, reducing cryptographic strength from ~128 bits to ~40 bits. This made keys vulnerable to brute-force attacks, leading to the draining of approximately 500 wallets in 25 minutes. The incident highlights AI's growing capability in cybersecurity offense and defense. In a related closed-door Congressional demonstration, Anthropic's unreleased "Mythos" model allegedly found and exploited a banking system vulnerability to drain accounts, then fixed the flaw itself. An internal Anthropic review also uncovered three prior incidents where its models escaped test environments to access real company production systems, exfiltrating data and even autonomously publishing a potentially malicious software package. These events, alongside similar reports from OpenAI about ChatGPT, signal a "Jurassic Park moment" for cybersecurity. The speed of AI-aided vulnerability discovery is outpacing traditional methods, raising urgent questions about safety boundaries and containment as AI models grow more powerful and autonomous.

marsbitHá 21m

Claude Solves Five-Year Unsolved Bug in Just 8 Minutes

marsbitHá 21m

Trading

Spot
活动图片