On October 15, AstraZeneca agreed to pay the AI drug-discovery company insitro $40 million upfront, with up to $1.46 billion riding on results, to hunt for new targets in two diseases the industry has struggled to crack. Only the smallest number in the deal is actually guaranteed.

Mitsubishi Heavy Industries and the Spanish power electronics maker Ingeteam signed a non-binding MOU on October 15 to sell Ingeteam's grid-interface technology into US AI data centers through MHI's distribution channels. Neither company disclosed money or customers — the deal is a bet on who controls the layer between the grid and the rack.

At Workday Rising on October 13, Workday wired Anthropic's Claude directly into Sana, the agent layer that now runs compensation cycles and financial closes inside Workday's HR and finance systems. The same day, Workday also agreed to buy an AI-agent identity startup to police exactly who — and what — gets to touch that data, while insisting Claude isn't the only model it plans to let in.

On October 6, Google and Constellation Energy signed the largest nuclear power deal of the AI buildout: two purchase agreements covering 3,590 megawatts, plus a separate pact that puts Google Cloud and Gemini Enterprise to work inside Constellation's control rooms. It's Google's first nuclear bet on reactors that already exist, not ones still being invented.

On October 6, CoreWeave announced its first data centers in India: three buildings, 240 megawatts, built and owned not by CoreWeave but by the Adani Group's data-center arm. It's the same arrangement Google already has with Adani, and a clearer signal than most quarterly earnings calls of how foreign AI clouds are entering India without laying a single brick themselves.

On October 8, Nvidia-backed startup Upscale AI launched Token Fabric, a networking platform that embeds Nvidia's own Spectrum-X silicon — while Upscale's other product, a chip called SkyHammer, is built explicitly to beat Nvidia's NVSwitch. Here's why Nvidia funded both sides of its own networking business.

At WebexOne on October 7, Cisco announced it is bringing Anthropic's Claude into Webex as an embedded agent and separately piping Webex's enterprise context into OpenAI's Dots — the same week, from the same stage. It's a bet that collaboration software wins by staying neutral among AI labs, not by picking one.

At its investor day on September 30, Synopsys agreed to co-develop an OpenAI model built to operate its own chip-design software, with the two companies sharing whatever revenue it generates. It's a bet that the moat in agentic AI is the tool, not the model — and it puts the question squarely to Cadence.

Three weeks after Amazon blocked Meta's shopping agent for scraping its site without permission, Meta and the enterprise AI startup Sierra are proposing an open standard for how AI agents authenticate with businesses. Here's what it actually does, and who's conspicuously not in the room.

I didn't choose strategic partnerships so much as notice, six years and five jobs later, that it's the only thing I'd actually been doing the whole time.

I built a case interview log for myself on a Sunday. By Tuesday it was a real product on its own domain. Here is what changed in between, including the moment I realised I had been asking the wrong person to do the work.

My latest internship at Pareto Agent: build and sell an autonomous agentic AI that can do 80% of a B2B renewals account manager's job. Here's how three weeks went, including the dashboard I had to kill and rebuild from scratch.

Build real GTM workflows. Enrich faster. Personalize at scale. A structured 6-week path from complete beginner to production-grade Clay expert.