Pinterest is putting $4 billion behind a bet that Amazon Web Services can keep its AI ambitions fed with enough chips. The visual discovery platform confirmed on June 4, 2026 that it will commit that sum to AWS cloud services through 2031, the largest infrastructure agreement in the company’s 16-year history. The timing is notable: AWS parent Amazon reports second-quarter 2026 earnings today, and Wall Street will be watching whether deals like Pinterest’s are translating into the growth hyperscalers keep promising investors.
The Pinterest AWS cloud deal extends a relationship that dates back to 2010, but the scope has changed. Instead of renting generic compute, Pinterest is locking in AWS Trainium chips and Graviton processors to run the large language models and vision-language models behind its AI-powered visual search, according to Sahm Capital’s reporting on the agreement. It’s a small deal by the standards of 2026’s AI infrastructure arms race, but it says a lot about how mid-size tech platforms are securing capacity before it runs out.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Pinterest Bets $4 Billion on AWS Through 2031
CNBC first reported the agreement on June 4, 2026, describing it as Pinterest’s largest agreement to date. Law firm Sidley, which represented AWS in the negotiations, called it a “long-term strategic cloud services agreement” in its own announcement, and confirmed that AWS and Pinterest have worked together since 2010, according to Sidley’s deal summary.
Pinterest serves more than 600 million monthly users worldwide, and the company is framing this AWS cloud deal as the infrastructure backbone for its next generation of AI-assisted discovery features, according to AWS’s own announcement of the partnership. That announcement states plainly that the planned $4 billion commitment for cloud services through 2031 is the largest infrastructure investment in Pinterest’s history, and that Pinterest will use AWS Trainium and Graviton to train and run AI models at scale.
“This enhanced partnership with AWS provides us with the computational flexibility, hardware options, and infrastructure efficiency needed to expedite our AI ambitions.”
Matt Madrigal, Chief Technology Officer, Pinterest — via CNBC
Madrigal’s framing matters because it signals the deal isn’t just about capacity. It’s about the specific mix of custom silicon, hardware options and pricing flexibility Pinterest needs to run inference at the scale a 600-million-user platform demands.
Inside the Deal: Trainium, Graviton and the Move to EKS
Three technical shifts define what Pinterest is actually buying. First, AWS Trainium accelerators will handle training and inference for the large language models and vision-language models that power Pinterest’s visual search and personalized discovery. Second, Pinterest is expanding its use of AWS Graviton, the custom Arm-based processor line that already covers roughly one-third of the company’s compute footprint. Third, and perhaps most consequential for Pinterest’s engineering org, the company plans to move from EC2-based environments toward Amazon EKS, AWS’s managed Kubernetes service, a shift Sahm Capital detailed in its coverage of the agreement.
That EKS migration is worth pausing on. Moving a platform the size of Pinterest off EC2-based deployment and onto Kubernetes orchestration isn’t a weekend project. It typically means re-architecting how workloads get scheduled, scaled and billed. A simplified version of what that transition looks like at the infrastructure layer, using Graviton-based nodes for AI inference, resembles this:
# Illustrative example: a Graviton-based EKS node group for AI inference
eksctl create nodegroup \
--cluster=visual-search-prod \
--name=graviton-inference-pool \
--node-type=c7g.4xlarge \
--nodes-min=10 \
--nodes-max=200 \
--node-ami-family=AmazonLinux2023 \
--asg-accessThat’s a generic illustration, not Pinterest’s actual configuration, but it captures the shape of the work: swapping fixed EC2 fleets for an auto-scaling Kubernetes node pool built on Graviton silicon. For a platform running AI inference against hundreds of millions of users, that kind of elasticity directly affects the bill.
Why Pinterest Needs More Compute Right Now
Pinterest has spent the past two years repositioning itself around AI-driven discovery rather than simple image search. Visual search, personalized recommendations and AI-assisted shopping features all lean on models that get more expensive to run as they get more accurate. Serving those models to 600 million monthly users, per AWS’s own numbers, requires the kind of dedicated accelerator capacity that’s been in short supply across the industry through 2026.
That scarcity is precisely why multi-year commitments like this one have become the norm rather than the exception. Cloud providers increasingly want customers to commit capital up front in exchange for guaranteed capacity, and customers like Pinterest are willing to trade flexibility for certainty that the chips they need will actually be available when they need them. It’s the same dynamic driving GPU and memory price spikes across the hardware market this year.
A 16-Year AWS Relationship Enters a New Phase
Pinterest and AWS have worked together since 2010, back when Pinterest was a scrappy startup and AWS was still primarily known for EC2 and S3 rather than custom AI silicon. Sixteen years later, that relationship has evolved into a $4 billion strategic commitment, which MediaPost reported is explicitly aimed at powering the next version of Pinterest’s AI-powered visual search.
That long history is part of why the deal reads as an expansion rather than a switch. Pinterest never ran a public competitive bake-off against Azure or Google Cloud for this contract. Instead, AWS extended and deepened an existing account, which is generally cheaper and faster for both sides than migrating a platform this size to a new provider. Data gravity, existing tooling and engineering familiarity all favor incumbency once a company reaches Pinterest’s scale.
How Wall Street Reacted to the Announcement
Markets responded favorably on announcement day. Pinterest shares rose nearly 5%, while Amazon shares gained about 1.7%, according to coverage aggregated by Yahoo Finance and Cloud Computing News. That’s a meaningfully positive reaction for a company signing away $4 billion in future spending, and it suggests investors read the deal as a credible AI growth signal rather than a cost overhang.
The muted but positive move in Amazon’s stock fits a broader pattern from 2026: individual customer commitments in the hundreds of millions to low billions barely move AMZN on their own anymore, since AWS’s order book has grown so large that any single deal is a rounding error. What moves the stock now is the aggregate trajectory of AWS’s backlog, which is exactly what investors will be parsing when Amazon reports earnings later today.
Pinterest’s $4 Billion in Context: Comparing 2025-2026 AI Cloud Megadeals
Set against the biggest cloud commitments of the past year, Pinterest’s agreement is modest in dollar terms but part of the same pattern: AI-hungry companies locking in multi-year capacity with a hyperscaler rather than buying compute on demand. OpenAI’s $38 billion, seven-year agreement with AWS, announced November 3, 2025, remains the largest of these deals, giving OpenAI access to hundreds of thousands of Nvidia GPUs and roughly 2 gigawatts of Trainium3 and Trainium4 capacity. Meta’s AI compute deal with Google Cloud, worth more than $10 billion over six years, and Snowflake’s $6 billion AWS commitment round out the field.
| Customer | Cloud Provider | Deal Value | Term | Announced |
|---|---|---|---|---|
| OpenAI | AWS | $38 billion | 7 years | Nov 3, 2025 |
| Meta | Google Cloud | $10+ billion | 6 years | 2025 |
| Snowflake | AWS | $6 billion | Multi-year | 2026 |
| Anthropic / Amazon | AWS | $5 billion investment, $100B pact | Multi-year | 2026 |
| AWS | $4 billion | Through 2031 | June 4, 2026 |
Every deal on that list shares a structure: a large upfront dollar commitment in exchange for guaranteed capacity, often tied to a specific chip family. Tech Insider covered the OpenAI-AWS Bedrock agreement and Amazon’s Anthropic investment in detail, and the Pinterest deal follows the same playbook at a smaller scale.
AWS’s Growing Backlog and Why Every Deal Counts
Pinterest’s $4 billion lands as Amazon prepares to report second-quarter 2026 results after markets close today. Analysts polled ahead of the release expect AWS revenue of roughly $40.5 billion for the quarter, up 31% from the $30.8 billion AWS reported in the second quarter of 2025, according to IndexBox’s earnings preview. Amazon has guided to roughly $200 billion in capital expenditures for full-year 2026, spending it first disclosed back in February.
The more telling figure is AWS’s remaining performance obligations, the contracts signed but not yet fulfilled, which stood at $364 billion heading into the most recent quarter, per that same IndexBox analysis, up sharply from about $200 billion at the end of the third quarter of 2025. Deals like Pinterest’s are a small slice of that backlog, but they’re also the kind of steady, recurring commercial revenue that balances out the more speculative mega-deals dominating headlines. AWS doesn’t need another OpenAI-sized contract every quarter. It needs hundreds of Pinterest-sized ones.
The Graviton and Trainium Economics Behind the Deal
Custom silicon is the quiet engine behind why deals like this pencil out for both sides. AWS designs Graviton and Trainium chips in-house, which lets it offer better price-to-performance than reselling Nvidia GPU capacity at market rates, particularly as GPU scarcity has pushed Nvidia’s own pricing higher through 2026. For Pinterest, expanding Graviton usage beyond the third of its compute footprint it already covers is a straightforward cost lever, since Graviton instances typically run cheaper per unit of throughput than comparable x86 instances.
Trainium plays a different role. It’s AWS’s answer to Nvidia’s grip on AI training and inference hardware, and getting large customers to build on Trainium rather than defaulting to Nvidia GPUs is central to AWS’s long-term margin strategy. Every workload Pinterest shifts onto Trainium is a workload AWS doesn’t have to source Nvidia silicon for, at a time when Nvidia allocation remains tight across the industry.
| Layer | Previous Setup | Target Under 2026 Deal |
|---|---|---|
| Orchestration | EC2-based environments | Amazon EKS (Kubernetes) |
| General compute | Mixed x86 instances | Expanded Graviton (from ~1/3 of footprint) |
| AI training/inference | Limited custom silicon | AWS Trainium for LLMs and vision-language models |
| Contract structure | Standard cloud spend | $4B strategic commitment through 2031 |
From EC2 to EKS: What the Kubernetes Shift Signals
Pinterest’s planned move to Amazon EKS fits a broader industry trend that Tech Insider has tracked through 2026: large-scale AI workloads increasingly run on Kubernetes rather than raw virtual machines, because container orchestration makes it far easier to scale inference clusters up and down as demand fluctuates. Readers wanting the mechanics of that migration can walk through our own guide to setting up Amazon EKS.
For a platform like Pinterest, where visual search traffic spikes around shopping seasons and viral trends, Kubernetes-based autoscaling is less about novelty and more about not paying for idle GPU and Trainium capacity during quiet periods. That’s the same cost logic driving FinOps teams across the industry to scrutinize AI spend more aggressively this year.
AWS vs Azure vs Google Cloud in the AI Infrastructure Race
Pinterest’s decision to deepen its AWS relationship rather than split workloads across providers stands out against a cloud market that remains a three-way race. AWS held roughly 30% of global cloud infrastructure market share in the first quarter of 2026, ahead of Microsoft Azure at 24% and Google Cloud at 13%, according to figures compiled by CommandLinux’s market share tracker and corroborated by CloudZero’s provider analysis. Growth rates tell a different story: Google Cloud grew 63% year over year, Azure grew 40%, and AWS grew a comparatively modest 19% in the same period, meaning AWS is defending share against faster-growing rivals even as it wins deals like Pinterest’s.
| Provider | Market Share | YoY Growth | Notable 2026 AI Commitment |
|---|---|---|---|
| AWS | 30% | 19% | Pinterest ($4B), OpenAI ($38B) |
| Microsoft Azure | 24% | 40% | OpenAI capacity buildout |
| Google Cloud | 13% | 63% | Meta compute deal ($10B+) |
| Other providers | 33% | Varies | Oracle, IBM, Alibaba and others |
Enterprise cloud infrastructure spending hit $129 billion in the first quarter of 2026 alone, up 35% year over year, and AI workloads now account for roughly 19% of total cloud spending, up from just 8% in 2023. Pinterest’s commitment is a small but representative data point inside that much larger shift. Readers who want the full head-to-head can see our AWS vs Azure vs Google Cloud comparison.
What the Deal Means for Pinterest’s Product Roadmap
For Pinterest’s product team, guaranteed Trainium and Graviton capacity through 2031 removes one of the biggest constraints on shipping AI features: not knowing whether the compute will be there when a feature is ready to launch. Visual search, AI-assisted shopping recommendations and personalized discovery feeds all depend on running increasingly large models against enormous image and video catalogs in near real time.
Locking in capacity years in advance also lets Pinterest plan its roadmap around what it can actually build, rather than what compute happens to be available on the spot market in a given quarter. That’s a meaningful advantage in a year when GPU and memory shortages have forced other companies to delay or scale back AI features.
The Broader Cloud AI Spending Boom
Pinterest’s $4 billion sits inside a much larger wave of hyperscaler capital commitments. Tech Insider has tracked this trend through several data points this year, including data center leases topping $850 billion as Meta and Microsoft lead a global buildout, and cloud waste climbing to 29% of AI cloud spend as budgets strain to keep pace with demand for accelerator capacity.
What distinguishes Pinterest’s deal from the headline-grabbing $38 billion and $100 billion commitments is scale, not structure. Mid-size platforms with hundreds of millions of users, not just frontier AI labs, are now signing multi-year capacity contracts, which suggests the AI infrastructure buildout has moved well past its early, lab-only phase and into a broader commercial one.
Risks and Open Questions Around the Commitment
A five-plus-year, $4 billion commitment is not without risk for Pinterest. Locking in spend that far out assumes AI-driven features keep growing engagement and revenue enough to justify the cost, an assumption that’s come under scrutiny industry-wide as investors question whether hundreds of billions in AI capex are translating into proportional returns. Amazon and Microsoft’s combined AI spending has already drawn public skepticism from investors watching for payoff, and Pinterest’s much smaller bet will face the same basic question on a smaller stage: does the AI actually move revenue?
There’s also a lock-in question. Deepening a single-vendor relationship through 2031 trades negotiating leverage and multi-cloud flexibility for pricing certainty and guaranteed capacity. If Trainium pricing or performance falls behind Nvidia-based alternatives over the life of the contract, Pinterest has limited room to pivot without disrupting a codebase increasingly built around AWS-specific tooling like EKS and Graviton.
What Comes Next: 5 Predictions Through 2031
- More mid-size platforms will follow. Expect other social, e-commerce and discovery platforms in the 100-million-to-1-billion-user range to sign similar multi-year AWS, Azure or Google Cloud AI commitments over the next 12 to 18 months as available capacity keeps tightening.
- AWS’s backlog keeps climbing past $364 billion. With deals like Pinterest’s stacking on top of megadeals like OpenAI’s, expect AWS’s remaining performance obligations to set new records in upcoming quarters.
- Graviton adoption accelerates industry-wide. As Nvidia GPU pricing stays elevated, more AWS customers will likely follow Pinterest’s lead in expanding custom-silicon usage as a cost-control measure.
- EKS becomes the default path for AI workload migrations. Pinterest’s EC2-to-Kubernetes shift will likely be cited by AWS as a reference case for other large customers making the same move.
- Investor scrutiny of AI capex intensifies before it eases. Expect at least one more earnings cycle of pointed questions about AI infrastructure ROI before spending commitments like this one are judged a clear success or a costly overreach.
Frequently Asked Questions
How much did Pinterest agree to pay AWS?
Pinterest committed $4 billion to AWS cloud services under an agreement running through 2031, making it the largest infrastructure deal in Pinterest’s history.
When was the Pinterest AWS cloud deal announced?
The deal was announced on June 4, 2026, and confirmed via reporting from CNBC and an official AWS announcement.
What AWS technology will Pinterest use under the deal?
Pinterest will use AWS Trainium chips for training and running large language models and vision-language models, expand its use of AWS Graviton processors, and migrate from EC2-based environments to Amazon EKS.
Is this Pinterest’s first cloud partnership with AWS?
No. Pinterest and AWS have worked together since 2010. The 2026 agreement is an expansion of that existing 16-year relationship, not a new partnership.
How did investors react to the announcement?
Pinterest shares rose nearly 5% and Amazon shares gained about 1.7% on the day the deal was reported, according to coverage from Yahoo Finance and Cloud Computing News.
How does Pinterest’s $4 billion compare to OpenAI’s AWS deal?
OpenAI’s AWS agreement, announced in November 2025, is worth $38 billion over seven years, nearly ten times the size of Pinterest’s commitment. Both deals rely on AWS Trainium chips, but OpenAI’s also includes access to hundreds of thousands of Nvidia GPUs.
What is AWS Graviton, and why does it matter here?
AWS Graviton is Amazon’s custom Arm-based processor line, designed to offer better price-to-performance than comparable x86 instances. It already powers about a third of Pinterest’s compute infrastructure, and the new deal expands that further.
Why are so many companies signing multi-year AI cloud commitments in 2026?
GPU and specialized AI accelerator capacity has been in short supply through 2026, pushing cloud providers to prioritize customers willing to commit capital and volume years in advance. Multi-year deals guarantee access to capacity that isn’t reliably available on demand.
Related Coverage
- Snowflake Commits $6B to AWS as Stock Jumps 36% [2026]
- OpenAI on AWS Bedrock: $38B Pact Ends Azure Lock-In [2026]
- Amazon’s $5B Anthropic Bet: 5GW Trainium and $100B AWS Pact [2026]
- How to Set Up Amazon EKS: 12 Steps, 90 Min [2026]
- AWS vs Azure vs Google Cloud 2026: The Definitive Cloud Platform Comparison
- Data Center Leases Hit $850B, Meta, Microsoft Lead [2026]
For more coverage of hyperscaler strategy and cloud infrastructure economics, visit Tech Insider’s Cloud Computing section.
