Amazon Invests $33 Billion in Anthropic. Here Is What Every Amazon Brand Should Understand.

24 Apr 2026

Amazon Invests $33 Billion in Anthropic. Here Is What Every Amazon Brand Should Understand.

On April 20, 2026, Amazon and Anthropic announced an expanded partnership that reframes how seriously Amazon is treating artificial intelligence — not as a feature set, but as the foundational layer of its entire business going forward.

The numbers alone make the scale clear. Amazon will invest $5 billion in Anthropic immediately and up to an additional $20 billion in the future, tied to commercial milestones. This is on top of the $8 billion Amazon previously invested, bringing Amazon’s total potential commitment to $33 billion. The deal was struck at a valuation of $350 billion — notably more favorable terms than the $380 billion valuation Anthropic achieved in its February funding round.

But the investment figure is only part of the story. The structure of the deal reveals something more significant about where Amazon is heading.

What the Deal Actually Involves

Anthropic has committed to spending more than $100 billion on AWS technologies over the next decade. This encompasses current and future generations of Trainium — Amazon’s custom AI silicon — including Trainium2, Trainium3, Trainium4, and subsequent iterations, alongside tens of millions of Graviton CPU cores. Anthropic will secure up to 5 gigawatts of capacity to train and power its advanced AI models.

With frontier data centers costing $40 to $50 billion per gigawatt, Anthropic has achieved competitive silicon pricing of around $20 billion per gigawatt through this arrangement — optimized specifically for its models. The deal gives AWS a demand anchor that de-risks its multi-generational silicon investment, while Anthropic gains supply certainty in a market where accelerator availability remains constrained.

On the product side, AWS customers will now be able to access the full Anthropic-native Claude console directly from within their existing AWS accounts, with no additional credentials, contracts, or billing relationships required. The integration brings Claude closer to the infrastructure layer rather than keeping it as a third-party model accessed through an API.

Why This Is Not Just a Cloud Deal

It is tempting to read this as a straightforward infrastructure arrangement — Amazon gets cloud revenue, Anthropic gets compute. But the deeper implication is about control of the AI development stack at a moment when that stack is becoming decisive.

Amazon is reportedly planning to spend around $200 billion on capital expenditures in 2026, largely focused on AI infrastructure. The Anthropic partnership is one piece of a much larger build. AWS revenues accelerated 24% year over year in Q4 2025 — its fastest pace in 13 quarters — reaching a $142 billion annual run rate, with Trainium cited as a multi-billion-dollar annualized business growing triple digits. Amazon is not just investing in AI to keep pace. It is investing in owning the infrastructure that other AI companies depend on.

Anthropic’s annualized revenue crossed $30 billion in 2026, up significantly from the prior year, with both enterprise adoption and consumer usage increasing across subscription tiers. However, this rapid growth strained the infrastructure, affecting performance during peak usage, which is partly what makes this compute commitment so timely.

The competitive context is also worth noting. The deal echoes an arrangement Amazon struck with OpenAI just two months prior, when it joined a $110 billion funding round and committed $50 billion, structured partly as cloud infrastructure services rather than straight cash. Amazon is placing major bets on multiple frontier AI labs simultaneously while ensuring both run on AWS infrastructure.

What Project Rainier Tells Us About Amazon’s AI Ambitions

One of the clearest signals of how seriously Amazon is treating this partnership is Project Rainier — the compute cluster built jointly with Anthropic. Project Rainier houses nearly half a million Trainium2 chips and was, at launch, larger than any AI compute cluster in the world. Anthropic actively uses it to train and deploy Claude models, and it is now a template for deploying the kind of raw computational power that will allow AI to tackle the hardest human challenges.

With meaningful Trainium2 capacity already online and nearly 1 gigawatt of combined Trainium2 and Trainium3 capacity expected by year-end 2026, the infrastructure timeline is moving fast. This is not a roadmap — it is already being built.

What This Means for Amazon Sellers and Brands

Most brands on Amazon are focused on campaigns, listings, and fees. That is the right focus for day-to-day execution. But the AI infrastructure Amazon is building right now will determine what the platform looks like in two, three, and five years — and that has direct consequences for how products are discovered, recommended, and purchased.

Rufus, Amazon’s AI shopping assistant already used by 300 million customers and driving over $12 billion in incremental annualized sales, runs on this infrastructure. So does the pricing intelligence, the demand forecasting, the automated replenishment signals, and the personalization layer that increasingly mediates what a shopper sees before they ever run a keyword search.

Over 100,000 organizations are already running Anthropic’s Claude models through Amazon Bedrock. As Claude becomes more deeply embedded in AWS — and as AWS becomes more deeply embedded in every layer of Amazon’s retail, logistics, and advertising stack — the AI layer becomes less of a feature and more of the operating environment.

For brands, that operating environment is not optional. You are already selling inside it. The question is whether your product content, pricing strategy, and listing structure are built for a platform where AI mediates an increasing share of discovery and purchase decisions — or whether they are optimized for a version of Amazon that is already being replaced.

The Bigger Picture

Amazon is making a generational bet. The deal shows how serious the AI infrastructure race has become. Amazon is not just investing money — it is locking in long-term control over compute, which is quickly becoming the most valuable resource in AI.

For sellers, the practical takeaway is not to study chip architecture. It is to recognize that every AI-powered feature Amazon has rolled out in the past eighteen months — and every one it will roll out in the next five years — is being built on infrastructure that Amazon is now cementing with a $33 billion commitment and a decade-long partnership.

The platform you are selling on today is the early version of something much more AI-driven. Planning accordingly, not just for the next campaign but for the kind of brand presence that an AI-mediated marketplace rewards, is one of the more important strategic decisions a brand on Amazon can make right now.

If you need support with PPC, DSP, AMC, analytics, or a long-term growth strategy, contact the ANavigator team at info@anavigator.co

🚀 Book a call to get a FREE AUDIT by the link below: 🚀
⬇️ ⬇️ ⬇️
Book a call – FREE AUDIT

 

Follow my Weekly Newsletter on LinkedIn:
 / amazon-digest-for-brands-7232361008185372672  
Follow me on LinkedIn:
 / ookovalov 
Follow ANavigator on social media:
 / anavigator  
 /@anavigator_official
 / anavigator7  
 / @anavigators  

 

LinkedIn page to contact us:

 

Author: Oleksandr Kovalov
Role: Founder & CEO @ ANavigator
— The ANavigator Team

our activity

News and podcasts

LATEST UPDATES

Amazon Just Removed Custom Return Instructions for FBM Sellers
Blog
August 20, 2026
Amazon Just Removed Custom Return Instructions for FBM Sellers
Starting August 2026, a field that seller-fulfilled brands have quietly relied on for years disappeared from Seller Central. The option to add custom return instructions in Return Settings and the Manage Seller-Fulfilled Returns screen is gone, confirmed by Amazon moderator Billy_Amazon in the Seller Central forums when sellers began reporting the change around August 13. Amazon's notice was a single sentence: "Starting August 2026, the option to add return instructions in your return settings will no longer be available." No replacement was announced. No equivalent field was introduced. For brands selling bulky, expensive, fragile, or condition-sensitive products through FBM, the implications are worth thinking through carefully, particularly in the context of five other FBM return changes that have already taken effect in 2026. What Sellers Actually Lost The custom return instructions field allowed FBM sellers to communicate specific requirements directly to buyers when a return was authorized. Common uses included telling customers to return items unused, include original packaging, use specific packaging materials, or understand that return shipping was their responsibility when the item was not defective. For products where condition on return directly affects resale value- outdoor equipment, electronics accessories, custom items, fragile goods, anything with hygiene or safety considerations- that field was doing real operational work. It set expectations before the package was shipped back, which reduced the frequency of returns arriving damaged, used, incomplete, or improperly packaged. Without it, buyers now receive a return label with Amazon's standard messaging. The seller has no dedicated channel to communicate product-specific return requirements at the moment of return authorization. Amazon's stated alternative is Buyer-Seller Messaging; FBM sellers can still reach buyers through the messaging system to communicate return requirements. That is technically true, but it is a manual, reactive step rather than a systematic one. It requires the seller to initiate a message after every return authorization, and it is subject to Amazon's messaging policies, which restrict commercial communication. The Broader FBM Return Picture in 2026 The removal of custom return instructions is one piece of a larger pattern of FBM return policy changes that have accumulated throughout 2026. Taken together, they represent a significant tightening of how seller-fulfilled returns work, and a consistent shift toward Amazon controlling more of the return experience while sellers absorb more of the cost. Effective February 8, 2026, all FBM sellers must use Amazon prepaid return labels for every order, regardless of item value. The previous high-value exemption, which allowed sellers of electronics, jewelry, cameras, and other expensive items to opt out of prepaid labels, was eliminated entirely. Sellers of high-ticket items now absorb prepaid return shipping costs on every return. Effective January 26, 2026, the FBM refund processing window was extended from two business days to four calendar days. If a seller does not process a refund within four calendar days of receiving a returned item, Amazon may issue an automatic refund, and in most cases, the seller loses eligibility for SAFE-T claim reimbursement. The four-day window sounds generous but requires tight operational discipline. Returns that arrive on Friday need a decision by Monday. Returns that arrive around holidays compress the inspection timeline further. Sellers previously had two business days, which excluded weekends. The new four calendar-day window sounds longer but is operationally shorter in practice for sellers without weekend fulfillment operations. What FBM Sellers Should Do Now The absence of a custom instructions field does not mean you have no options. It means the options require more proactive effort. The most effective replacement for custom return instructions is proactive messaging. When you authorize a return, immediately send a Buyer-Seller message that covers your key return requirements: unused condition, original packaging, return shipping responsibility if applicable. This does not happen automatically, so it needs to be built into your return authorization workflow as a standard step, not an occasional practice. For products where specific handling is genuinely important- fragile items, products with hygiene implications, items where incomplete packaging significantly affects resale value- review whether your product detail page, packaging inserts, and any post-purchase communications address return expectations proactively. Reducing the information gap before a return is initiated is more reliable than bridging it at the point of authorization. On the SAFE-T claim side, the four-day inspection window is now the critical deadline. A seller generally has four days after receiving the returned product to inspect it and submit a SAFE-T claim. Eligible claims may cover the return label cost plus up to 50% of the item's price, depending on circumstances and product category. If returned inventory arrives damaged, used, or incomplete, document the condition immediately upon receipt and submit the SAFE-T claim within the window, not after. For brands with high-value SKUs where the combination of mandatory prepaid labels, four-day refund windows, and now removed return instructions creates compounding risk, the economics of FBA versus FBM are worth revisiting by ASIN. FBA fees are higher, but Amazon takes on the return risk and associated costs for FBA inventory. For high-value items where damage or fraud is a real concern, those extra fees may now be cheaper than the combined cost of prepaid labels, damaged returns, and missed SAFE-T claims. The Broader Pattern 2026 has been the most active year for FBM return policy changes in recent memory. Mandatory prepaid labels, a revised refund window, high-value exemption removal, and now the elimination of custom return instructions have all landed within eight months. Each change individually is manageable. Together, they represent a consistent shift: Amazon is standardizing the return experience for buyers at the cost of seller control and flexibility. For brands running significant FBM volume, the practical response is not to fight the direction; it is to rebuild operational workflows around the current reality, document returns thoroughly, use SAFE-T claims consistently, and evaluate whether FBM remains the right fulfillment model for each SKU in your catalog. The custom return instructions field is gone. The work it was doing still needs to happen; it just requires more deliberate effort than it did before August 2026. If you want to stay updated on Amazon changes, subscribe to our blog. If you need support with PPC, DSP, AMC, analytics, or a long-term growth strategy, contact the ANavigator team at info@anavigator.co  Book a call to get a FREE AUDIT using the link below:     Book a call – FREE AUDIT   Follow my Weekly Newsletter on LinkedIn:  / amazon-digest-for-brands-7232361008185372672   Follow me on LinkedIn:  / ookovalov  Follow ANavigator on social media:  / anavigator    /@anavigator_official  / anavigator7    / @anavigators     LinkedIn page to contact us:   Author: Oleksandr Kovalov Role: Founder & CEO @ ANavigator — The ANavigator Team
Learn more
Blog
August 18, 2026
Since September 2025, Amazon DSP advertisers have had access to Netflix's ad inventory. What changed on August 4, 2026, is practical: Amazon published the full technical documentation for how to actually set up a Netflix campaign through Amazon DSP, including deal types, vendor requirements, frequency capping, audience limitations, and reach reporting. The integration is no longer in early access. It is a live, documented advertising channel. For brands already running Amazon DSP, this is worth understanding in detail before adding it to a media plan. Why This Integration Exists Netflix reaches a unique and attentive audience - 44% of members who see an ad on Netflix never saw it on broadcast TV or other streamers. Campaigns on Netflix drive almost 2x the TV norm on long-term brand building and 23% above benchmarks on purchase intent compared to competitors. Amazon's interest in the partnership is straightforward. Amazon DSP's authenticated identity graph, built from purchase behavior, browsing signals, and retail data, can now be applied to Netflix's premium streaming environment. A retail brand that previously could use Amazon's purchase data only to target audiences on Amazon-owned properties can now extend those same audience definitions to Netflix inventory. Amazon's purchase signals reach Netflix's living room. For brands, this collapses a step that previously required separate negotiations, separate tech stacks, and separate measurement. Netflix inventory is now accessible through the same DSP interface, the same deal workflow, and the same reporting layer you use for Prime Video and other streaming placements. How Netflix Campaigns Work in Amazon DSP Netflix supply is available for private auction and programmatic guaranteed deals only, it is not available through open auction. Before creating a campaign, you need a Netflix deal with the correct configuration. Existing deals are not compatible with Amazon audience-enabled campaigns and will not work. There are two ways to secure a deal. You can work with Netflix or your Amazon Ads representative directly, in which case the deal appears automatically in your Amazon DSP manager account via API once it is ready. Or you can submit a deal proposal to Netflix directly inside Amazon DSP through the deal proposal workflow. Once a deal is in place, the campaign setup follows the standard Amazon DSP structure: campaign, ad groups, and ads. Activation is straightforward: enable the ad groups and campaign via the toggle in the Campaigns interface. Vendor requirements are strict. Netflix maintains firm third-party vendor requirements for all ad serving and measurement. Supported ad serving vendors are Google Campaign Manager 360 and Innovid. Supported measurement and verification vendors are DoubleVerify and Integral Ad Science. All other third-party vendors are not supported. If you are currently using a different vendor for your DSP campaigns, you will need to switch to one of the approved options or use Amazon DSP-hosted ad options before running on Netflix inventory. What Amazon Audiences Can and Cannot Do on Netflix Starting Q2 2026 in the US, advertisers can apply Amazon Audiences to Netflix campaigns, audience segments built from Amazon's purchase and behavioral data, applied to Netflix's streaming environment. That is the meaningful capability at the center of this integration. Your Amazon customer cohorts, category buyers, high-LTV segments, competitor brand shoppers, can now be used to target audiences on Netflix. The limitations are specific and worth knowing before building a campaign: Amazon audiences are available on private marketplace run-of-network deals only. They are not available on programmatic guaranteed deals. Do not layer third-party audiences, genre detargeting, or Netflix-side targeting on Amazon audience-enabled deals - this will stop delivery. Similar audiences and durable audiences are not available on Netflix inventory. Ad Exposure remarketing is not available. Audience-level reporting is not available. Brand safety settings through Amazon DSP do not apply to Netflix inventory; contact Netflix directly for brand safety options. Frequency caps are fully enforced for private auction and preferred deals. For programmatic guaranteed deals, impressions count toward caps but cannot be suppressed on the Netflix line item. What Is Actually Measurable De-duplicated reach and frequency metrics are now available for Netflix ad group lines and populate automatically, no setup, no opt-in, no additional cost. These metrics are available across all 11 Netflix ad-supported locales. Netflix and Spotify are now direct integrations available through Amazon DSP, with reach and frequency metrics that feed into cross-publisher measurement. That matters for brands running streaming TV alongside Netflix; you can measure unduplicated reach across Prime Video, Netflix, and other inventory sources from the same platform, without reconciling separate reports. The measurement piece is what has historically made streaming TV difficult to justify in a performance-oriented media plan. De-duplicated reach and frequency data, combined with Amazon's purchase attribution from the same DSP, changes that calculus meaningfully. What Brands Should Be Thinking About Before Running Netflix Campaigns Netflix inventory is premium, which means it is expensive relative to other streaming placements, and the minimum deal thresholds reflect that. This is not a format to test with a small budget expecting performance metrics comparable to Sponsored Products. The right frame for Netflix through Amazon DSP is brand building with measurable audience reach, not direct-response conversion. The brands for whom this makes sense have a few things in common. They are already running Amazon DSP with meaningful budget; Netflix is an addition to a working DSP strategy, not a standalone channel. They have a product or brand story that benefits from a lean-back, high-attention streaming context. And they have creative that is production-quality for the streaming environment, not repurposed Sponsored Brands video. A recent Dove campaign spanning consumer products and custom creative on Netflix saw an almost 60% increase in new shoppers for products. That result reflects both the reach quality of Netflix's audience and the creative investment Dove made for the placement. The format rewards brands that treat it like a premium channel, not a cheaper extension of search advertising. If you are running Amazon DSP and have not yet looked at Netflix as an inventory source, the August 2026 documentation update is a useful moment to evaluate it. The technical setup is now clearly documented, the audience integration is live, and the measurement tools are in place. Whether the economics make sense depends on your category, your brand objectives, and your existing DSP performance, but the access barrier that existed before the Amazon partnership is gone. If you want to stay updated on Amazon changes, subscribe to our blog. If you need support with PPC, DSP, AMC, analytics, or a long-term growth strategy, contact the ANavigator team at info@anavigator.co  Book a call to get a FREE AUDIT using the link below:     Book a call – FREE AUDIT   Follow my Weekly Newsletter on LinkedIn:  / amazon-digest-for-brands-7232361008185372672   Follow me on LinkedIn:  / ookovalov  Follow ANavigator on social media:  / anavigator    /@anavigator_official  / anavigator7    / @anavigators     LinkedIn page to contact us:   Author: Oleksandr Kovalov Role: Founder & CEO @ ANavigator — The ANavigator Team
Learn more