Covariant pricing

About

Covariant is an Emeryville, California warehouse-AI lab founded in 2017 by four researchers with deep reinforcement-learning roots: Pieter Abbeel (the prominent UC Berkeley robotics-learning professor), Peter Chen, Rocky Duan, and Tianhao Zhang — the first three all ex-OpenAI. Its product is the “Covariant Brain,” an AI platform that gives industrial robot arms the ability to autonomously pick, place, and manipulate a near-infinite variety of items in warehouses — apparel, health and beauty, grocery, and pharmaceuticals — without per-item programming. In March 2024 the company unveiled RFM-1 (Robotics Foundation Model 1), an 8-billion-parameter multimodal model trained on text, images, video, robot actions, and physical measurements, reportedly running on 100+ warehouse arms and learning from tens of millions of real pick trajectories.

Covariant raised roughly $222M over its life, including an $80M Series C in 2021 and a further ~$75M Series C extension in 2023 co-led by Radical Ventures and Index Ventures (with CPP Investments, Amplify Partners, Gates Frontier, AIX Ventures, and Northgate Capital). It rarely sold direct: the Covariant Brain typically shipped inside an integrator’s robotic cell — most visibly KNAPP’s “Pick-it-Easy Robot,” part of a multi-year partnership.

Crucially for a pricing blueprint, Covariant never published a price. The site exposes only marketing and a sales/“Contact us” motion; covariant.ai/pricing returns a 404. There is no plan grid, no per-pick rate, no robot-hour rate, and no self-serve tier — every deployment was scoped and quoted. And the bigger story is a consolidation post-mortem: in August 2024 Amazon hired the three founders plus about a quarter of staff and took a non-exclusive license to Covariant’s foundation models — a “reverse acqui-hire” that moved the team and core IP to Amazon while leaving Covariant to continue in a diminished form under new CEO Ted Stinson. So this page documents what is honestly known — the company, its funding, its product, its sales-led posture, and the Amazon event — rather than inventing prices it never published.

Pricing summary : a sales-only warehouse-AI deployment, now largely absorbed by Amazon

Covariant runs a sales-led, no-public-price commercial model. There is no subscription, no published per-pick or per-robot-hour rate, and no self-serve tier to evaluate. The dimensions, as far as they can be observed, are:

• The Covariant Brain (enterprise/integrator deployment) — quoted bespoke via “Contact us.” No floor price, packaging, or per-unit rate is disclosed. In practice it was usually sold embedded in a partner’s robotic picking cell (e.g. KNAPP’s Pick-it-Easy Robot) rather than as a standalone metered SKU.
• RFM-1 (the model, not a product) — the 8B robotics foundation model is the engine inside the Brain, not a separately priced item. Amazon’s non-exclusive license to these models in 2024 is the only “transaction” around RFM-1 that is publicly known, and its terms are undisclosed.
• No free tier, no self-serve, no public meter — robotics manipulation is a high-touch, integration-heavy sale; there is nothing to sign up for.

What makes this different: unlike a software lab, Covariant sold a physical-world outcome (robots reliably picking goods) bundled with hardware integration — the kind of deal that resists a public unit price. And unlike peers that are still scaling, Covariant’s “pricing story” effectively ended as a standalone one in August 2024, when its founders and models moved to Amazon in a structure that has become a recurring AI-consolidation pattern.

Pricing by product

Sales motions across products: fully sales-led for the only revenue-bearing surface (the Covariant Brain, quoted per deployment and frequently delivered through an integrator partner). There is no self-serve or developer motion.

Hidden costs : What Covariant buyers actually pay

Because Covariant publishes no price, the “real bill” cannot be read in advance — and for a physical-robotics deployment it is far larger than any software line item.

The deployment is the cost. A Covariant Brain rollout is not a SaaS subscription; it is an integrated robotic picking cell. The real spend includes the robot hardware and end-effectors, integration and commissioning at the warehouse, the Covariant Brain software/AI itself (quoted, not listed), and ongoing support and maintenance. Because the Brain typically shipped through an integrator like KNAPP, the buyer’s contract is usually with that integrator, and Covariant’s economics sit inside that bundle — invisible as a standalone price.

Want to model what an autonomous-picking deployment might cost? There’s no published Covariant rate to plug in, but you can sketch scenarios with the Covariant pricing calculator, and read how to choose a usage metric when the deliverable is a physical outcome (picks completed) rather than a software unit.

Pricing evolution : Covariant pricing history and changes

Covariant never had a public price to change. Its “pricing evolution” is really a commercial-trajectory evolution: well-funded warehouse-AI lab → production robotics foundation model → quasi-acqui-hired by Amazon. The milestones below are reconstructed from primary announcements and a live 2026-06-14 site check.

Cadence

Tracked range: 2021 Q3–2026 Q2. Zero public price changes across the company’s life — there has never been a published price to revise. Quarters not listed had no relevant public pricing event.

Notable changes

• 2021-07 — $80M Series C to deploy the Covariant Brain across industries (total funding ultimately ~$222M); sales-led, no price card (PR Newswire).
• 2024-03 — RFM-1 (8B robotics foundation model) unveiled at MODEX 2024; engine of the Brain, not a separately priced SKU (BusinessWire).
• 2024-08 — Amazon reverse acqui-hire: founders Pieter Abbeel, Peter Chen, Rocky Duan + ~25% of staff join Amazon; Amazon takes a non-exclusive license to Covariant’s models; Covariant continues under new CEO Ted Stinson (TechCrunch, About Amazon).
• 2026-06-14 — Live check confirms no public pricing: covariant.ai/pricing 404s; the site is a minimal marketing presence with a sales/“Contact us” motion.

What’s unique : Covariant’s distinctive pricing mechanics

1. The price is the absence of a price — bundled inside a robot cell. Covariant never anchored on a public rate. The Covariant Brain was sold as part of an integrated, hardware-plus-AI picking deployment, usually through a partner like KNAPP, so the AI’s economics were folded into a larger robotics contract. For a physical-world outcome (robots picking real goods), opacity is the deliberate packaging — every deal is scoped to the site, the SKU mix, and the integration.

2. The model is licensed, not metered. RFM-1 is genuinely a foundation model, but Covariant never exposed it as a priced API the way an LLM lab exposes per-token pricing. The only publicly known “transaction” on the model is Amazon’s non-exclusive license — a one-off corporate deal, not a published per-use rate. The value metric was the deployed outcome, not the inference.

3. The pricing story ends in consolidation. Most blueprint pages track a living price. Covariant’s effectively froze in August 2024, when the founders and core IP moved to Amazon in a “reverse acqui-hire” — the same license-don’t-acquire structure seen across recent AI consolidation (Adept→Amazon, Inflection→Microsoft, Character.AI→Google). The residual company continues sales-led, but the strategic engine moved in-house at a hyperscaler.

Strengths & weaknesses

Billing UX : Covariant billing controls and transparency

• Billing controls — None are public. There is no self-serve dashboard, usage meter, or plan-management UI on the site; commercial terms are handled entirely through a sales (and often integrator) relationship.
• Usage visibility — Not applicable to a public surface. Any per-pick or throughput metering lives in the warehouse deployment and the integrator’s systems, not in a Covariant billing console.
• Payment options — Not disclosed. Deployments are presumably invoiced under custom enterprise/integrator contracts; there is no card checkout or public billing portal.
• Transparency — Low on the commercial side throughout: Covariant has been open about its research (RFM-1) and partnerships, but never about price.

Strategic wins : Why Covariant’s pricing decisions worked (and where they led)

1. Bundling the AI into a robot cell protected the deal

By selling the Covariant Brain through integrators rather than as a standalone metered SKU, Covariant tied its price to a delivered physical outcome — robots reliably picking goods — instead of a fungible software unit. That kept the AI from being line-item-compared against cheaper alternatives and let each deployment be quoted to the site. See usage-based pricing strategy for when metering an outcome beats metering a unit.

2. A research-credible foundation model as the moat

RFM-1 let Covariant lead with capability (a real, multimodal robotics foundation model in production) rather than a price-performance table. For an enterprise robotics buyer, demonstrated reliability across an apparel/grocery/pharma item mix is the value metric — exactly the kind of thing that resists a public unit price. Related: outcome-based pricing trends.

3. The Amazon exit monetized the team and IP

The 2024 reverse acqui-hire is, in one reading, the ultimate “pricing” outcome: rather than scaling a metered product, Covariant’s founders and a non-exclusive license to its models were absorbed by the world’s largest warehouse operator. For a capital-intensive physical-AI company, a hyperscaler deal can be the cleanest realization of value — even as it ends the independent commercial story. Compare how AI companies are restructuring their monetization.

Areas to improve : Gaps in Covariant’s pricing approach

1. Publish something — even a deployment-shape anchor

A literal 404 price page maximizes evaluation friction. Even a one-line description of how deployments are priced (per picking cell? per site? per pick throughput?) would let warehouse operators self-qualify instead of starting cold with sales. The total opacity invites exactly the cost-unpredictability anxiety that slows capital-equipment buyers.

2. Clarify the post-Amazon offering

After the 2024 deal, it is unclear publicly what the diminished company now sells, to whom, and on what terms. A clear statement of the residual product and its commercial model would reduce uncertainty for existing customers and prospects weighing a multi-year robotics commitment.

3. Show RaaS economics like peers do

Robotics-as-a-service peers have at least gestured at observable structures (per-robot-hour, per-site). Surfacing even a high-level economic model — without giving up bespoke quoting — would let buyers benchmark Covariant against alternatives like Agility Robotics and Apptronik instead of treating it as a pure black box.

Key takeaways

1. No public price is itself a pricing decision. Covariant published zero commercial pricing and sold the Covariant Brain through sales and integrators — a deliberate, high-touch posture for a physical-world outcome, not an oversight.
2. The AI hid inside the hardware. By bundling the Brain into an integrator’s robotic cell, Covariant tied its economics to a delivered picking outcome and kept them invisible as a standalone price.
3. A foundation model isn’t always a metered product. RFM-1 was the engine, not a priced API; the only known transaction on it is Amazon’s non-exclusive license — a corporate deal, not a per-use rate.
4. Consolidation can be the real “exit price.” The 2024 reverse acqui-hire moved founders and IP to Amazon — for a capital-intensive physical-AI company, a hyperscaler deal can realize value that a metered product never would have.
5. Opacity plus a quasi-acqui-hire equals maximum uncertainty. Buyers face both no public price and a diminished, restructured vendor — the steepest possible due-diligence burden.

UBP implications

1. Physical-world outcomes resist public unit pricing. When the deliverable is “robots reliably pick your goods,” bundling the AI into an integrated, quoted deployment can beat a published per-unit rate. UBP practitioners should match transparency to fungibility — meter and publish commodities; quote and conceal integrated physical outcomes.
2. A model can be IP to license, not a meter to bill. Covariant shows that owning a foundation model doesn’t oblige you to expose a per-use rate. Sometimes the model’s value is realized through a one-off corporate license (here, to Amazon) rather than recurring metered revenue.
3. Consolidation is a pricing endpoint worth planning for. For capital-intensive physical-AI, the path to value may run through a hyperscaler acqui-hire rather than a scaled metered product. Founders should weigh that endgame when they choose not to build a transparent, self-serve commercial motion.

Sources

• Covariant — official site (no pricing page; sales/“Contact us” motion) (accessed 2026-06-14)
• Covariant Introduces RFM-1 to Give Robots the Human-like Ability to Reason (BusinessWire) (accessed 2026-06-14)
• Amazon hires the founders of AI robotics startup Covariant (TechCrunch) (accessed 2026-06-14)
• Amazon hires from AI robotics startup Covariant, licenses technology (About Amazon) (accessed 2026-06-14)
• Amazon hires founders of AI robotics startup Covariant (SiliconANGLE) (accessed 2026-06-14)
• Covariant Raises $80M in Series C Funding (PR Newswire) (accessed 2026-06-14)
• Congratulations to Covariant on their $75M Series C (Index Ventures) (accessed 2026-06-14)
• KNAPP and Covariant Extend Their Success Story (accessed 2026-06-14)
• Browse the pricing blueprint corpus

Bottom line

Covariant is the warehouse-AI corpus’s clearest consolidation post-mortem: an Emeryville lab founded by ex-OpenAI researchers (incl. Berkeley’s Pieter Abbeel), ~$222M raised, that built a real production robotics foundation model (RFM-1) — and never published a price for any of it. The Covariant Brain was sold sales-led, usually bundled inside an integrator’s robotic picking cell (KNAPP), so its economics stayed invisible. Then in August 2024 Amazon hired the three founders plus about a quarter of staff and took a non-exclusive license to the models — a “reverse acqui-hire” that moved team and core IP in-house at the world’s largest warehouse operator, leaving Covariant to continue in a diminished, still-sales-led form. covariant.ai/pricing is a literal 404; the real story isn’t a rate card, it’s how a well-funded physical-AI startup’s value was ultimately realized through consolidation rather than a metered product.

Want to compare Covariant against other embodied-AI and robotics peers? See Agility Robotics and Apptronik, or browse the full pricing blueprint.