Kleiner Perkins Raises $3.5 Billion on a Handful of AI Bets
On Tuesday, Kleiner Perkins announced it raised $3.5 billion across two funds — $1 billion for early-stage ventures and $2.5 billion for late-stage growth. This is a 75% jump from its $2 billion fundraise less than two years ago. The firm, founded in 1972, now operates with just five partners after recent departures — Ev Randle left for Benchmark, and Annie Case moved to an advisory role. The lift comes from early positions in Together AI, Harvey, OpenEvidence, Anthropic, and SpaceX (the latter two expected to IPO this year). Kleiner also captured returns from Figma’s IPO last year, where it led the $25 million Series B in 2018, and reportedly profited when portfolio company Windsurf was acqui-hired by Google last summer.
The takeaway: lean teams with concentrated exposure to AI infrastructure are attracting outsize capital. Kleiner is betting the next decade belongs to picks-and-shovels plays around model builders and deployment platforms. For LPs, the question is whether five partners can deploy $3.5 billion without diluting returns — or whether this is a final repricing before generalist firms lose access to the best AI deals altogether.
Amazon Buys Two Robotics Startups in Two Weeks
Amazon confirmed it acquired Fauna Robotics, a two-year-old startup building kid-size humanoid robots for the home. The deal follows Amazon’s acquisition earlier this month of Rivr, a Zurich-based autonomous robotics firm known for its stair-climbing delivery robot. Terms of neither deal were disclosed. Fauna began shipping its first product, a 59-pound bipedal robot called Sprout, earlier this year to select R&D partners. Both founding teams — ex-Meta and ex-Google engineers in Fauna’s case — will join Amazon in New York City.
Amazon has historically preferred internal development or licensing over outright acquisition in robotics. The shift signals urgency. The company is racing to automate last-mile delivery and in-home services before Tesla, Figure, or Chinese manufacturers dominate the form factor. Fauna’s consumer-scale humanoids and Rivr’s navigation stack could merge into a unified platform for package delivery, elder care, or warehousePickPack 2.0. For operators, this is a reminder that Amazon rarely buys for revenue — it buys to collapse timelines and lock out competitors. If you’re building in robotics and Amazon hasn’t called, you’re either too early or already obsolete.
Arm Breaks Its Own Business Model and Starts Selling Chips
On Tuesday, Arm announced it is producing its own semiconductors — a break from its 30-year model of licensing chip designs to manufacturers. CEO Rene Haas unveiled the Arm AGI CPU at an event in San Francisco, describing it as the world’s most efficient agentic CPU. The chip is fabricated by Taiwan Semiconductor Manufacturing Corporation using its 3nm process and is designed to handle AI agent workloads in data centers. Meta has received samples; OpenAI, SAP, Cerebras, Cloudflare, SK Telecom, and Rebellions have committed to purchase. Arm projects full production availability in the second half of this year.
The move puts Arm in direct competition with Intel and AMD, as well as its own licensees — including Apple, Nvidia, Amazon, and Google, all of whom use Arm designs in their custom processors. Creative Strategies forecasts the data center CPU market will grow from $25 billion this year to $60 billion by 2030; when agentic AI workloads are included, that figure climbs to $100 billion. Arm’s pitch is power efficiency: the company claims its AGI CPU delivers better performance per watt than x86 alternatives, translating to billions in energy savings for hyperscalers.
The risk: Arm’s customers may perceive it as a competitor and accelerate their own internal chip programs. The upside: even a small share of a $100 billion market dwarfs Arm’s current licensing revenue. For investors, this is a bet that the AI buildout will reward vertical integration over modular ecosystems — and that Arm can execute on manufacturing, supply chain, and customer support at scale.
The Pentagon Tries to Punish Anthropic — and a Judge Calls It Out
On Tuesday, US District Judge Rita Lin said during a hearing in San Francisco that the Pentagon’s decision to designate Anthropic a supply-chain risk “looks like an attempt to cripple Anthropic” and “looks like punishment” for the company’s public pushback on military use of its AI tools. Anthropic filed two federal lawsuits alleging illegal retaliation after the Trump administration labeled it a security risk following the company’s push for restrictions on how its models could be used by the armed forces. The Department of Defense (now called the Department of War) argues it followed procedures and determined Anthropic’s tools could no longer be relied upon in critical moments. Defense Secretary Pete Hegseth posted on social media that no military contractor may conduct any commercial activity with Anthropic, though the department’s attorney later admitted in court that Hegseth has no legal authority to enforce such a ban beyond Pentagon contracts. The Pentagon is working to replace Anthropic with Google, OpenAI, and xAI over the coming months.
Lin’s ruling on Anthropic’s request for a temporary injunction is expected within days. The case has opened a broader question: can AI companies place ethical guardrails on government use without facing existential retaliation? For investors, Anthropic’s customer base is now in flux — some have already paused contracts pending legal clarity. The company expected to IPO this year alongside OpenAI, but the supply-chain designation could spook both retail and institutional buyers. If the injunction is denied, Anthropic loses credibility as a stable vendor. If granted, it sets precedent that Silicon Valley can negotiate with the Pentagon on use-case terms — a dynamic Washington has never tolerated in defense procurement.
—
**Editor’s Conclusion**
Four signals, one direction: capital is consolidating around platforms that control compute, deployment, and customer lock-in. Kleiner’s $3.5 billion raise rewards early AI exposure. Amazon’s robotics M&A sprint is a landgrab before form factors commoditize. Arm’s chip gambit is a bet that the next cycle belongs to power efficiency, not architecture flexibility. And Anthropic’s courtroom fight is a test of whether AI companies can impose use restrictions on their most powerful customer — or whether Washington will simply build around them.
The through-line: every company in today’s coverage is racing to own the next bottleneck. Chips, robots, models, and military contracts are no longer separate markets — they’re layers in the same stack. For investors, the question is whether you’re positioned in the companies building infrastructure or the ones renting it. The gap between those two will define returns for the next decade.
If this was useful, drop a like or comment below. More signal, less noise — every time.
Leave a Reply