OpenAI、コーディングツールとビジネス顧客に焦点を当てるため「サイドクエスト」戦略を廃止と報道

Skip to content

Read full article about: Google plans nearly two million new AI chips as it turns to Marvell for custom designs

Google is in talks with chip designer Marvell Technology to develop two new specialized chips for its data centers, according to The Information, citing two sources. One of the chips is a memory processing unit designed to run alongside Google's in-house TPUs (Tensor Processing Units) and split AI tasks based on compute and memory demands. The other is a new TPU built specifically for inference - running finished AI models.

Google plans to produce nearly two million memory processing units, with the design expected to be finalized by next year. Marvell is a natural partner for Google: the company already designed the first inference chip for the startup Groq. Nvidia licensed Groq's LPU technology (Language Processing Unit) in December 2025 for $20 billion and used it to unveil the Groq 3 LPU and the Groq 3 LPX rack system at GTC 2026. Groq founder Jonathan Ross previously worked at Google and is one of the original TPU engineers.

The move is also about reducing Google's dependence on its current chip design partner Broadcom, which charges high per-unit fees for each TPU produced. Broadcom remains an important partner for now, though, having signed a new contract with Google through 2031 in early April.

Salesforce bets on "Agent Albert" to prove AI won't kill enterprise software

Wall Street worries AI could make traditional enterprise software obsolete. Salesforce CEO Marc Benioff is pushing back with a new AI product and a homegrown metric.

Read full article about: The NSA is using Anthropic's most powerful AI model Mythos

The NSA, the US intelligence agency responsible for electronic surveillance, is using Anthropic's most powerful AI model, "Mythos Preview." That's according to Axios, citing two sources. The Pentagon has classified Anthropic as a security risk and has been trying since February to block the company as a vendor. The NSA, notably, falls under the Pentagon's authority. In court filings, the Defense Department has even argued that Anthropic's tools pose a threat to national security.

Anthropic has restricted access to Mythos to around 40 organizations as part of "Project Glasswing," arguing that the model's offensive cyber capabilities are too dangerous for a broad release. That framing has drawn some skepticism, though.

Anthropic CEO Dario Amodei met with White House officials last week to discuss deploying Mythos across government agencies. The Pentagon demanded that Anthropic make Claude available for "all legal purposes." Anthropic refused, drawing the line at mass surveillance and autonomous weapons.

The UK's intelligence services also have access to the model through the country's AI Security Institute, according to Axios.

Read full article about: Anthropic's revenue surge reportedly fuels talk of trillion-dollar valuation

Anthropic has dramatically accelerated its revenue growth. According to The Information, the company's annualized revenue topped $30 billion "earlier this month," more than triple what it was at the end of last year. That could put Anthropic slightly ahead of OpenAI, though the numbers aren't directly comparable. The growth is being driven by Claude Code and Cowork, likely along with rising token sales from reasoning models.

Margins have also improved sharply. Gross margins swung from minus 94 percent in 2024 to plus 40 percent the following year, while annualized revenue jumped from under $1 billion to $9 billion. That kind of momentum let Anthropic set tough terms for its September 2025 funding round, with investors expected to commit checks in the hundreds of millions.

Two people familiar with the company say no new funding rounds are planned for now. If one does happen, it won't be before a board meeting in May at the earliest, The Information reports. CFO Krishna Rao and his team have reportedly already fielded offers valuing the company at $800 billion, and some investors think a trillion-dollar valuation is within reach.

Read full article about: German court rules AI comic adaptation of copyrighted photo doesn't violate the original

Turning someone else's photo into a comic with AI doesn't automatically violate copyright. That's the ruling from a German Higher Regional Court on April 2, 2026 (case no. I-20 W 2/26). An animal photographer who sells underwater dog shots sued a former business partner who had fed her photo of a diving dog into AI software and posted the comic-style result on its website.

The original photo (left) and the AI comic adaptation (right), which the court ruled is legal. | Decision of April 2, 2026 (case no. I-20 W 2/26)

The court denied the appeal, finding that the AI image doesn't copy the protectable elements of the original, such as framing, perspective, lighting, and sharpness. The motif and subject themselves aren't protected. The judges cited a recent European Court of Justice ruling that focuses on the recognizable adoption of specific creative elements rather than the overall impression.

The court also clarified that AI-generated works only qualify for copyright protection when a human makes recognizably creative decisions; picking an AI suggestion or typing generic prompts isn't enough. The decision aligns with earlier rulings from other German courts and echoes the position of the US Copyright Office.

Read full article about: First token counts reveal Opus 4.7 costs significantly more than 4.6 despite Anthropic's flat pricing

Anthropic's Opus 4.7 carries the same sticker price as 4.6, but it burns through noticeably more tokens per request. That's according to measurements published by developer Abhishek Ray on Claude Code Camp.

Anthropic's own migration guide cites an increase of 1.0 to 1.35x. Ray's numbers somewhat line up in that range, while some content types push past it: 1.325x on average for real Claude Code content, 1.445x for a CLAUDE.md file, and 1.47x for technical documentation. A community evaluation on tokens.billchambers.me goes even higher, pointing to 37.4 percent more tokens and costs per request across 483 submissions.

Community data shows a 37.4 percent jump in both token usage and per-request costs when switching from Opus 4.6 to 4.7. | Image: Screenshot via Tokenomics

Code takes a bigger hit, Ray notes, while prose sees a smaller bump, and Chinese and Japanese texts are barely affected. For a sample session of 80 turns, he estimates an extra 20 to 30 percent in costs, pushing the bill from $6.65 to somewhere between $7.86 and $8.76.

In return, users get slightly better instruction following: a test using the IFEval benchmark across 20 prompts shows Opus 4.7 sticking to strict instructions five percentage points more reliably than its predecessor.

AI-generated influencers flood social media with pro-Trump content ahead of midterms

Hundreds of AI avatars are flooding TikTok, Instagram, and YouTube with pro-Trump messaging. Some accounts have pulled in more than 35,000 followers and millions of views, and Trump himself has already shared AI-generated content. It’s unclear whether this is the work of individual activists or a coordinated campaign.

Even the best AI models lose about half their performance when charts get complicated, new benchmark finds

The RealChart2Code benchmark puts 14 leading AI models to the test on complex visualizations built from real-world datasets. Even the top proprietary models lose nearly half their performance compared to simpler tests.