Just a few days ago on August 26, 2026. A press release went out.
Salesforce, the company that spent ten years building its own AI, just handed the keys to Anthropic.
No fanfare about failure. No apology. Just Marc Benioff on stage, calling it the best of both worlds.
That’s the tell.
When a company spends a decade insisting it can build its own AI, and then quietly stops, that’s not a pivot. That’s a confession. And once you start looking, it’s not just one company. It’s a pattern, with a few different shapes.
Salesforce: give up the build, keep the business
Einstein. Then Einstein GPT. Ten years of Salesforce trying to own its own intelligence, baked right into the CRM.
Then Claudeforce. Claude becomes the default reasoning engine across Agentforce and Slack.
Marc Benioff, Salesforce’s Chair and CEO, put it plainly:
We are bringing together the world’s #1 AI and #1 CRM - the best of both worlds... By fusing Claude’s extraordinary reasoning with the trusted data, workflows, and governance every enterprise runs on, we are delivering a dynamic interface that thinks, reasons, and acts.
Dario Amodei, Anthropic’s CEO, added that:
The world’s leading businesses turn to Claude for their most important work.
Ten years of plumbing. One partnership to make it think.
Samsung: out-manufacture, don’t out-build
Samsung built its own model. Called it Gauss. Put it inside Galaxy AI.
Then, quietly, most of the real work moved to Google’s Gemini instead.
When the partnership was announced, Samsung’s own Janghyun Yoon said
The goal was to deliver the best Gemini-powered AI experience on Galaxy.
Samsung can out-manufacture almost anyone on the planet.
It still couldn’t out-build Google. It stopped pretending otherwise.
Apple: Paying its Biggest Rival
Apple spent years on its own Apple Foundation Models. Siri was supposed to get smarter. It kept slipping.
January 2026: Apple signs a multi-year deal with Google. \
Gemini now runs the next Siri.
Reported price tag: close to a billion dollars a year.
The joint statement from Apple and Google didn’t dress it up:
Google’s technology provides the most capable foundation for Apple Foundation Models.
Apple has more cash and more engineers than almost any company alive. It still chose to pay its biggest rival rather than keep trying alone.
Allianz: never even tried to build it
Not every company goes through the build-then-abandon arc. Some skip straight to partnership.
In January 2026, Allianz, one of the world’s largest insurers, announced a global partnership with Anthropic.
No prior attempt at a homegrown model, no walk-back.
Just a direct decision.
CEO Oliver Bäte called it
A decisive step to address critical AI challenges in insurance.
Amodei’s response was worth sitting with:
Insurance is an industry where the stakes of using AI are particularly high.
Allianz didn’t need to fail at building AI first to know it shouldn’t.
That’s arguably the more mature version of the same lesson.
Thomson Reuters: the middle path
Not everyone gives up entirely. Thomson Reuters spent roughly $40 million training its own model for legal and tax work, built on an existing open-source foundation rather than from scratch.
In its own words - “most of our investment went into further training on decades of proprietary content.”
Here’s the part that matters: they still use Anthropic’s models everywhere else.
They didn’t try to compete on general intelligence. They built narrow and deep on the one thing they own that nobody else has, their proprietary legal content, and bought the frontier reasoning underneath it.
TELUS: never bet on one horse
TELUS took a fourth path. Instead of building its own model, or picking a single AI partner, it built Fuel iX, an internal gateway that lets employees choose between 40-plus models, including Claude and Gemini, depending on the task.
One TELUS engineer summed up the appeal simply: “a win-win that makes life so much easier.”
TELUS never tried to be an AI lab.
It also never bet the company on one partner.
It just built the plumbing to stay flexible.
Microsoft: build small, buy big
Microsoft is doing something more selective. It’s building its own models, called MAI, but only for specific, narrower tasks: a coding model that’s now the default in GitHub Copilot for individual VS Code users, an image model generating visuals in PowerPoint, a transcription model rolling into Teams.
The heavy reasoning behind most of Microsoft 365 Copilot chat is still largely OpenAI and Anthropic.
Its AI chief was direct about the motive: “our goal is to reduce and ultimately eliminate that cost.”
Even one of the biggest AI research operations on the planet is only bringing the cheaper, narrower tasks in-house, and still buying the hardest reasoning from outside.
The Lesson
There isn’t one right way to do this. There are at least five.
Salesforce, Samsung, and Apple tried to build their own AI, and all three ended up partnering instead.
Allianz skipped the attempt altogether and went straight to a partner.
Thomson Reuters found a middle path: build narrow where your data is unique, buy the rest.
TELUS refused to bet on any single model, homegrown or not, and built a gateway instead.
Microsoft buys frontier reasoning for the hardest problems while quietly building its own models for the cheaper, narrower tasks.
The one thing none of them did was try to be a full frontier AI lab on the side of their actual business.
Know what you are actually good at.
Everything else is a rental, or at most, a narrow build on top of someone else’s foundation.
Stay curious,
AI Lady 💫
Sources
Salesforce Newsroom, “Salesforce and Anthropic Announce Claudeforce: The #1 AI Meets the #1 AI CRM,” August 26, 2026
Samsung Newsroom, “Samsung and Google Cloud Join Forces to Bring Generative AI to Samsung Galaxy S24 Series,” January 2024
TechCrunch, “Google’s Gemini to power Apple’s AI features like Siri,” January 12, 2026
TechCrunch, same article as above (joint Apple/Google statement)
Allianz SE press release, “Allianz and Anthropic Forge Global Partnership to Advance Responsible AI in Insurance,” January 9, 2026
The New Stack, “Thomson Reuters trained its own AI model. Then it kept using Anthropic’s anyway,” August 2026
Google Cloud customer story, “TELUS: Unleashing workplace innovation and engineering excellence with AI”
Reporting on Microsoft Build 2026 MAI model rollout
Yahoo Finance, reporting on comments from Mustafa Suleyman, 2026
About the Author
Priya Tahiliani has spent the last 15 years at the intersection of business and technology. She is a Copilot Deployment and Adoption Leader with a global reinsurance company living and breathing AI adoption every single minute of her day.
In her past life she has worked as a SAP HCM and SAP SuccessFactors consultant with Big Four firms and clients worldwide.
She built and launched her company’s first AI tool by forging a great partnership with IT, and today she continues to work with leaders to help shape the future of work and drive AI enablement.
Beyond work, she served as Vice President of Public Relations at Toastmasters. She is also the Founder of the AI Collective – Oakville Chapter in Canada, part of the world’s largest community for AI professionals - a network dedicated to learning and leading responsibly with AI.
And of course, she writes the AI Lady newsletter, where she shares her experiences, insights, and thoughts about how AI is reshaping our workplaces.
If this article sparked something for you, share it with someone else navigating this shift. These conversations matter more when we have them together.