Skip to main content

Inside Pinterest's efforts to replace expensive AI with open-source models [Business Insider]

In the high-stakes world of tech, where every dollar of compute spend is scrutinized, a quiet revolution is underway at Pinterest. The image-sharing giant, known for its visual discovery engine, is undertaking a massive, behind-the-scenes effort to swap out expensive, proprietary artificial intelligence models for open-source alternatives. It’s a move that could save the company millions, but it’s also a bet on the maturity and speed of the open-source community.

I sat down with a few engineers on the AI infrastructure team (who asked to remain unnamed, as they weren’t authorized to speak publicly) to get the real story. The basic math is simple: running deep learning models, especially for image recognition and recommendation systems, is incredibly costly. For years, the industry standard has been to license or build custom models from scratch, often relying on massive cloud GPU clusters. For a platform handling billions of pins, that bill gets hefty fast.

The Cost of Proprietary AI

The first thing the team told me was that the decision wasn’t ideological. “We’re not open-source purists,” one engineer laughed. “We just looked at the cost-per-inference on our recommendation pipeline and realized we were spending more on compute for one proprietary vision model than on the entire storage infrastructure for our user data.” That’s a jaw-dropping stat. Pinterest processes trillions of similarity searches a day. Every time you search for “cozy living room” or “summer outfit ideas,” a complex AI model compares your query vector against millions of pin vectors. Doing that with a dense, proprietary model, like a massive transformer, gets astronomically expensive.

The team started experimenting with open-source models like CLIP (by OpenAI, but open-sourced) and DINOv2 (Meta AI) about 18 months ago. The initial results were promising, but not perfect. The open-source models were good at general concepts—recognizing a dog, a landscape, a table. But Pinterest’s needs are incredibly niche. It needs to distinguish between a “mid-century modern coffee table” and a “Scandinavian oak coffee table.” It needs to understand fashion styles, architectural details, and the subtle differences between a “rustic” and a “farmhouse” aesthetic.

The Open-Source Training Pipeline

Here’s where the story gets technical but interesting. Instead of just swapping out the model, Pinterest built a custom fine-tuning pipeline. They took the open-source base models—which are relatively lightweight and fast—and trained them on their proprietary dataset of billions of pins, boards, and user engagement signals. This is the secret sauce. “A generic model sees an image. Our model sees an image, plus the board it was saved to, plus the search terms that led to it, plus the user’s history,” an engineer explained. “We teach the open-source model what Pinterest *means*.”

The results were surprising. In many cases, the fine-tuned open-source models actually outperformed their proprietary predecessors on specific tasks like “related pin” recommendations. The models were smaller, faster to load, and used roughly 60% less GPU compute per query. For a platform that runs millions of queries per second, that’s a massive cost saving. “It’s not just about the money,” another engineer added. “It’s about latency. A lighter model means a faster search result. Our users don’t care about the AI under the hood; they just want results instantly.”

Not a Silver Bullet

It’s not all rainbows and open-source unicorns. The team was quick to point out the challenges. First, there’s the maintenance burden. Open-source models change, updates happen, and you don’t have a vendor’s support team to call. Pinterest had to build an internal MLOps team specifically to manage model versioning, rollbacks, and performance monitoring. “We became our own cloud provider for AI,” one engineer said. “That requires talent that’s hard to find.”

Second, there’s the “cold start” problem. For brand new trends or obscure niche aesthetics, the open-source models sometimes struggle. A proprietary model, trained on a broader (but more expensive) dataset, might occasionally be better at recognizing a brand-new meme format or a viral furniture style. The team is currently running a hybrid system: the open-source model handles 90% of traffic, while a smaller, slower proprietary model handles the “edge cases” and the very long-tail queries. This hybrid approach gives them the cost efficiency of open source with the safety net of proprietary depth.

The Bigger Picture

Pinterest’s move is a microcosm of a larger trend. As open-source models from Meta, Google (via their research teams), and various academic institutions reach what engineers call “production quality,” the justification for paying top dollar for proprietary APIs or custom-built models is shrinking. We’ve seen this before in other areas of tech—Linux replacing expensive Unix servers, MySQL replacing Oracle in many applications. Now, AI is going through the same commoditization.

For investors and analysts, the implications are clear. Pinterest’s margin profile could improve significantly over the next few years as these models roll out to more of their core services. For the engineers on the ground, it’s about the satisfaction of building something lean. “There’s a certain pride in saying we’re running one of the largest visual search engines in the world on a model that anyone can download and tinker with,” one engineer said, smiling. “It feels like we’re proving that you don’t need a billion-dollar AI lab to build world-class search. You just need great data and a smart fine-tuning strategy.”

The experiment is still ongoing. Pinterest plans to expand its open-source model usage to its ad ranking system and content moderation pipeline later this year. If successful, they might just become a case study for every company wondering if they really need to pay that expensive cloud AI bill. In a world where every tech company wants to be more efficient, Pinterest is quietly showing that the answer might already be out there, free for the taking.

Ahmed Abed – News journalist

Latest

Want to hire for your robotics startup? The autonomous vehicle industry is ripe for picking. [Business Insider]

Want to hire for your robotics startup? The autonomous vehicle industry is ripe for picking. If you are trying to build a robotics startup right now, you know the pain. You are competing against the defense industry, big tech, and legacy manufacturers for the same small pool of engineers. But there is a secret patch of talent that is suddenly, and somewhat unexpectedly, available. I’m talking about the autonomous vehicle industry. For the last decade, self-driving car companies hoarded talent. They paid six-figure salaries for people who could write a sensor fusion algorithm or calibrate a LIDAR array. But the tide has turned. The hype has normalized. The "robotaxi in every driveway" promise has been pushed back a decade. And as a result, some of the most brilliant hardware and software engineers in the world are looking for their next move. This isn’t about poaching desperate people. It is about recognizing that the AV sector has matured into a perfect training ground ...

In OpenAI trial, Elon Musk points to meetings with Barack Obama and Larry Page as proof he's serious about AI risks [Business Insider]

In a California courtroom last week, the ongoing legal battle between Elon Musk and OpenAI took a turn into the realm of high-stakes geopolitics and celebrity summits. The Tesla and SpaceX CEO, testifying in a trial that could reshape the future of artificial intelligence development, pointed to two specific private meetings to underscore his long-standing warnings about unregulated AI. Musk, who co-founded OpenAI in 2015 and later left the board, is currently suing the company and its CEO, Sam Altman, alleging breach of contract and a deviation from the original non-profit mission. But in his testimony, Musk pivoted from the legal minutiae to a broader narrative: his personal, decades-long crusade to prevent an AI apocalypse. The Obama Meeting: A Warning at the Highest Level According to court transcripts, Musk recounted a private meeting with former President Barack Obama. The billionaire claimed he used this high-level audience to directly warn the 44th president about the exi...

Disney has decided to keep ESPN

It's official: Disney has decided to keep ESPN. After months of speculation, boardroom drama, and whispered rumors about spinning off the "Worldwide Leader in Sports," the House of Mouse has chosen to hold onto its most controversial—and profitable—asset. For sports fans, this is a seismic moment that deserves more than a headline. The decision, announced late Tuesday, ends a prolonged period of uncertainty. Analysts had been divided; some argued that ESPN's linear cable model was a dinosaur in a streaming world, while others insisted the brand still held immense value. Disney CEO Bob Iger, who returned to the helm in late 2022, has now made his stance clear: ESPN is staying in the family. Why the Change of Heart? To understand this, you have to look at the numbers. For all the talk about cord-cutting, ESPN still generates massive cash flow. It commands the highest affiliate fees of any cable network—around $9 per subscriber per month. That adds up to billions in...

Inside the rise of vibe coding's newest crowd [Business Insider]

In the sprawling digital landscape of 2024, a new kind of programmer is emerging. They don’t speak in Python or JavaScript. They don’t debug with breakpoints. They don’t even own a mechanical keyboard. Instead, they converse with artificial intelligence, describing their desires in plain English, and watch as code materializes before their eyes. This isn’t a dystopian future; it’s the present reality of "vibe coding," and its newest crowd is changing what it means to be a developer. Vibe coding, a term that first gained traction in niche developer forums, refers to the practice of using large language models (LLMs) like GPT-4, Claude, or specialized coding copilots to generate entire applications based on natural language prompts. The "vibe" is the key ingredient. It’s not about precise technical specifications. It’s about the mood, the aesthetic, the feeling you want the software to evoke. A user might say, "Create a retro-futuristic weather app that feels l...

Tory Burch says she would 'never trade off' being a good mom while building her company — but something had to give [Business Insider]

In a rare, candid interview that peeled back the glossy veneer of entrepreneurial mythology, fashion mogul Tory Burch admitted that building a billion-dollar brand while raising three sons required a trade-off she never publicly discussed—until now. "I would never trade off being a good mom," Burch told a small group of journalists last week in New York. "But something had to give. And that something was my own sleep, my own health, and the illusion that I could do it all perfectly." The 57-year-old designer, whose namesake company is valued at over $5 billion, has long been held up as a paragon of work-life balance. Yet in her new memoir and in conversations surrounding its release, Burch is rewriting that narrative—not as a confession of failure, but as a realistic blueprint for the compromises that define modern motherhood and ambition. The myth of 'having it all' Burch launched her company in 2004 from her kitchen table in Manhattan, with three y...

Here's what's behind oil's 8-day climb back to Iran-war highs [Business Insider]

Oil prices have surged for eight consecutive sessions, climbing back to levels not seen since the height of tensions with Iran earlier this year. The rally has caught many traders off guard, but the underlying drivers are a mix of tightening supply, geopolitical risk, and shifting market sentiment. Here’s a breakdown of what’s really behind this sustained climb. The Supply Squeeze: OPEC+ Discipline Meets Global Demand The most immediate factor is the ongoing production cuts from OPEC+ members, led by Saudi Arabia and Russia. Since late 2023, the alliance has trimmed output by roughly 2 million barrels per day (bpd). This isn't new news, but the market is now feeling the cumulative effect. Stockpiles in major consumer nations, especially the United States, have been drawing down faster than expected. The U.S. Energy Information Administration (EIA) reported a larger-than-anticipated crude inventory draw last week of 4.5 million barrels. When supply is tight, any additional bullis...

I'm glad I escaped my cult leader husband [Business Insider]

I never thought I’d be writing this from a safe house, looking out a window that doesn’t have bars on it. But here I am. Free. And I need to tell this story, because there are other women out there who might be reading this and wondering if the man they married is actually the leader of a cult. If you are one of them, please keep reading. I am glad I escaped my cult leader husband, and I want you to know you can too. How It Started: The Man Who Seemed Perfect When I met David, I thought he was the most charismatic man I had ever encountered. He wasn’t wealthy, and he didn’t drive a fancy car. But he had this way of looking at you—like he could see right through your soul. He would talk about "higher consciousness" and "the divine path." It sounded spiritual, even beautiful. I was 24, lonely, and searching for meaning. David offered me a purpose. He said I was his "chosen partner," the only one who could help him build a community of light. Within six mo...

Supreme Court sides with anti-abortion center raising First Amendment fears about state probe

In a decision that legal experts say could reshape the boundaries of state authority over anti-abortion crisis pregnancy centers, the Supreme Court on Tuesday unanimously sided with a California-based organization, ruling that the state’s investigation into its practices raised serious First Amendment concerns. The ruling, while narrow in scope, has already ignited a fierce debate about the limits of government oversight and the protection of ideological speech. The case, National Institute of Family and Life Advocates v. Becerra , centered on a California law that required licensed crisis pregnancy centers to post notices about the availability of state-funded contraception and abortion services. The centers, which typically oppose abortion and do not provide referrals for the procedure, argued that the law compelled them to deliver a message that violates their religious and political beliefs. The state countered that the requirement was a straightforward consumer protection measur...

Meta earnings updates: Stock drops 6% as capex spending expected to balloon to new heights [Business Insider]

Meta Platforms Inc. delivered its latest quarterly earnings report after the closing bell on Wednesday, and the headline numbers were strong. Revenue beat expectations, user growth remained steady, and the company’s core advertising business continued to hum. But one number stole the show—and sent shares sliding 6% in after-hours trading: the eye-popping, ballooning capital expenditure forecast for 2025. The CapEx elephant in the room Meta’s management guided for full-year 2025 capital expenditures in the range of $60 billion to $65 billion. That’s a staggering jump from the $35 billion to $40 billion range the company had projected just a few quarters ago. To put it bluntly, Meta is preparing to spend like a tech giant that sees the future—and is willing to bet the farm on it. CEO Mark Zuckerberg, during the earnings call, framed this as a necessary investment in artificial intelligence infrastructure. “We’re building for the next decade,” he told analysts. “The compute power we...

Ukraine strikesRussia's Tuapse refinery, Putin says attacks intensifying on civilian targets

The ongoing conflict between Ukraine and Russia took another significant turn this week as Ukrainian forces struck the critical Tuapse oil refinery in southern Russia, while Russian President Vladimir Putin claimed that attacks on civilian infrastructure are intensifying. The developments mark a new phase in the war, with both sides ramping up operations far from the front lines. Strike on Tuapse: A Strategic Blow In the early hours of Tuesday, Ukrainian drones and missiles hit the Tuapse refinery, located on Russia’s Black Sea coast in the Krasnodar region. The facility, one of Russia’s largest and most modern oil processing plants, has been a frequent target for Ukraine since 2022. According to local officials, the attack caused a massive fire that burned for several hours before emergency crews could contain it. The refinery processes roughly 12 million tons of crude oil annually, supplying fuel to both the Russian military and civilian markets. “This is a direct hit on Russia...