AI Daily Brief: Claude Sonnet 5 Levels Up, Micron Bets Big on Anthropic's Memory Needs

Two threads ran through AI this week: how much smarter the models got, and how hard it's become to secure the memory chips that run them. Both showed up in the same seven days.

Claude Sonnet 5 becomes Anthropic's most agentic model yet

Anthropic shipped Claude Sonnet 5 on June 30 and made it the default model for Free and Pro users immediately, its largest capability jump between two Sonnet generations according to the company. Early testers describe a model that finishes multi-step tasks Sonnet 4.6 would have stopped short of, checking its own output without being asked. It closes much of the gap with Opus 4.8 on agentic work when run at higher effort levels, supports a one-million-token context window, and can generate up to 128,000 output tokens. Anthropic is pricing it aggressively too: $2 per million input tokens and $10 per million output through August 31, before it settles back to Sonnet 4.6's old rate.

Fable 5 comes back from exile, and Anthropic sweetens the return

Claude Fable 5 and Mythos 5 spent 18 days offline after Commerce ordered Anthropic to cut off foreign access in mid-June, triggered by Amazon researchers finding a prompt that got Fable 5 to identify software vulnerabilities and, in one case, write exploit code for one. The fix that ended the standoff was narrow: a single safety filter tuned to block that specific technique, reviewed by Commerce's AI standards office before controls came off on July 1. Anthropic is treating the return as an occasion: Pro, Max, Team, and select Enterprise plans get up to 50% higher weekly usage limits through July 7. It's a reminder that even the biggest labs' release calendars now run partly through Washington.

Micron ties its future to Claude's training runs

The memory side of the story moved just as fast. On June 22, Micron and Anthropic announced an agreement that bundles several things into one deal: co-designed memory and storage subsystems optimized around what the companies call "token economics," a multi-year supply contract covering HBM, DRAM, and SSDs for Anthropic's data centers, an internal Claude deployment across Micron, and a stake in Anthropic's $65 billion Series H, which closed in May at a $965 billion valuation. With Micron in, all three global HBM suppliers, Samsung, SK Hynix, and now Micron, are Anthropic infrastructure partners. Micron says its entire 2026 HBM output is already sold out. The timing wasn't an accident: the deal landed two days before Micron's fiscal Q3 earnings call, where the company posted revenue of $41.46 billion against estimates of $35.69 billion and watched its stock jump nearly 15% in a single session, hitting an all-time high days later.

SanDisk locks in $42 billion before the memory squeeze gets worse

SanDisk isn't naming Anthropic or OpenAI as customers, but it's riding the same wave. The company has signed five-year deals with multiple hyperscale cloud customers worth at least $42 billion, locking in more than a third of its fiscal 2027 production capacity and converting what used to be a brutally cyclical memory business into something closer to a subscription. It's also working with SK Hynix on stacked High-Bandwidth Flash, a NAND-based answer to HBM aimed at AI inference workloads, with samples due by year end. The read-through is simple: whoever isn't locking in memory supply years in advance right now is betting against the AI buildout continuing, and almost nobody in the industry is making that bet.

What it means for India

India has skin in this specific game. Micron's Sanand, Gujarat facility, the country's first commercial-scale semiconductor plant, opened in February and is expected to assemble and test tens of millions of chips this year, scaling to hundreds of millions in 2027. As Micron's global HBM output gets spoken for years out by deals like the Anthropic agreement, Sanand's role converting DRAM and NAND wafers into finished product becomes more strategically load-bearing, not less. Separately, the same enterprise AI push reshaping memory demand is reshaping India's IT services sector: Infosys has partnered with both Anthropic and OpenAI as the two labs build out enterprise deployment arms, a dynamic Indian IT firms are betting will still route large-scale implementation work their way even as frontier labs capture the higher-value strategy layer. On the funding side, Sarvam AI's $234 million Series B in June, at a $1.5 billion valuation, is a reminder that India's own AI stack is being built in parallel, not just consumed.

Markets and AI money

Money is flowing to the picks and shovels as fast as it's flowing to the labs themselves. Micron is now up roughly 280% year to date, the second-best performer in the S&P 500, on the same story that's pushing Anthropic toward what could be one of the largest public listings ever. The chips underneath the models turned out to be just as good a trade as the models running on them.