Physical Intelligence pricing

About

Physical Intelligence — styled π (Pi) — is a San Francisco research lab founded in 2024 to build a general-purpose “brain” for robots: Vision-Language-Action (VLA) foundation models that ingest raw camera feeds and natural-language instructions and output real-time motor commands to drive a robot’s joints and actuators. The founding team is a who’s-who of robot learning — Karol Hausman (CEO), Sergey Levine, Chelsea Finn, Brian Ichter and Quan Vuong, drawn from Google DeepMind, Stanford and UC Berkeley — paired with ex-Stripe operator Lachy Groom. The thesis is that robotics needs the same kind of single, general, transferable model that transformed language and vision, rather than one bespoke controller per task.

The lab has attracted exceptional capital for a pre-revenue company: a ~$70M seed (March 2024), a $400M Series A at roughly a $2.4B valuation in November 2024 (backed by Jeff Bezos and OpenAI, with Thrive Capital, Lux Capital and Bond Capital), and a $600M Series B at roughly a $5.6B valuation in November 2025 (led by CapitalG, with Lux, Thrive, Bezos, Index Ventures and T. Rowe Price) — about $1.07B raised in under two years. Valuation more than doubled between the two rounds.

Crucially for a pricing blueprint: Physical Intelligence sells nothing and publishes no price. The website (pi.website, which physicalintelligence.company redirects to) exposes only Research and a “Join Us” hiring link — there is no “Products” page, no “Pricing”, no “Contact sales”, and no commercial call-to-action anywhere; pi.website/pricing returns a 404. What the lab does put in public are its robot foundation models, released free: π0 and π0-FAST were open-sourced on Feb 4 2025 under the permissive Apache-2.0 license in the openpi repository, with π0.5 weights and fine-tuning code following later in 2025. So this page documents what is honestly known — the team, the funding, the open releases, and the absence of any commercial offering — rather than inventing numbers the lab has never published. It is, in pricing terms, even earlier-stage than Essential AI: where Essential AI at least sells a (sales-only) enterprise product, Physical Intelligence is not yet selling anything at all.

Pricing summary : a pre-commercial research lab with free open models

Physical Intelligence runs no commercial pricing model at all — it has not yet commercialized. There is no subscription, no per-token or per-robot API, and no self-serve tier to evaluate. The dimensions, such as they can be observed, are:

• Commercial product (not yet sold) — Physical Intelligence has no product on the market. There is no quote process, no floor price, no SKU, and not even a “Talk to us” CTA — the only outbound button on the site is “Join Us” (hiring). Any future deployment economics are entirely unannounced.
• Open-weight models (free) — π0, π0-FAST and π0.5 ship as open weights and code under Apache-2.0 in the openpi repo. Physical Intelligence sets no price for them; any inference or fine-tuning cost you incur is your own compute, not a Physical Intelligence charge.
• No revenue model disclosed — the company is funded by venture capital (~1.07 billion dollars raised), not by sales. A “$300/month per connected robot” figure that circulates online is a third-party analyst projection (Sacra) of how the lab could monetize — not a published Physical Intelligence price.

What makes this different: unlike a sales-only enterprise lab such as Essential AI — which still has a product to quote — Physical Intelligence has no commercial surface whatsoever. The closest peers in the embodied-AI race, like Figure AI and Apptronik, run pilots and (eventually) Robotics-as-a-Service economics; Physical Intelligence is one rung earlier — a foundation-model lab giving its core model away free while the monetization question stays entirely open.

Pricing by product

Sales motions across products: there is no revenue-bearing surface today, so there is effectively no sales motion — only a hiring funnel (“Join Us”). The free open releases are self-serve downloads governed by Apache-2.0, not commercial SKUs. Should the lab commercialize, the most likely shape (given embodied-AI peers) is a sales-led enterprise / RaaS motion — but nothing is announced.

Hidden costs : What Physical Intelligence users actually pay

Because Physical Intelligence sells no product, the “real bill” question collapses to one path today — running the free models — with the commercial path still entirely hypothetical.

Path 1 — there is no product to buy. You cannot purchase a Physical Intelligence deployment in 2026. There is no quote, no contract, no pilot price list — the lab has not commercialized. The only “cost” of engaging commercially is that there is no commercial path at all; the site routes you to careers, not sales.

Path 2 — running the free open models yourself. π0 / π0.5 weights are free under Apache-2.0, but running them is not. You supply the robot hardware, the GPUs for inference and fine-tuning, and the data-collection effort (Physical Intelligence notes that roughly 1–20 hours of task data is typically enough to fine-tune to a new task). The “free model” therefore carries real, self-borne costs — compute and robot data — that live entirely outside Physical Intelligence’s economics.

Want to model what running an open robot foundation model might cost? Physical Intelligence publishes no rate of its own, but you can sketch compute scenarios with the Physical Intelligence pricing calculator and compare hosted GPU and token economics with the AI token pricing tracker. For how to choose a meter when the deliverable is an outcome rather than a unit, see choosing the right usage metric.

Pricing evolution : Physical Intelligence pricing history and changes

Physical Intelligence has never had a public price to change — it has never sold anything. Its “pricing evolution” is really a capital-and-research evolution: a star-studded seed, two mega-rounds, and a free open-sourcing of its core model — all while keeping any commercial offering off the table entirely. The milestones below are reconstructed from primary announcements and a live 2026-06-14 site check.

Cadence

Tracked range: 2024 Q1–2026 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 event.

Notable changes

• 2024-03 — Founded; raises a ~$70M seed. Research lab, no product, no price (Maginative; Sacra).
• 2024-11 — $400M Series A at roughly $2.4B, backed by Jeff Bezos and OpenAI (CNBC).
• 2025-02-04 — π0 and π0-FAST open-sourced free under Apache-2.0 in the openpi repo (pi.website/blog/openpi).
• 2025-11-20 — $600M Series B at roughly $5.6B, led by CapitalG; total funding ~1.07 billion dollars (Bloomberg / Dataconomy).
• 2026-06-14 — Live check confirms no product and no public pricing: pi.website/pricing 404s; the only CTA is “Join Us.”

What’s unique : Physical Intelligence’s distinctive pricing mechanics

1. There is no price because there is no product. Most companies in this corpus have some commercial surface — a rate card, a quote process, at least a “Talk to us.” Physical Intelligence has none. The only button on the site is “Join Us.” This is not opacity in the Essential AI sense (a hidden price behind sales); it is the genuine absence of a commercial offering. The “pricing mechanic” is that the lab is still pre-monetization — funded entirely by ~$1.07B of venture capital, not revenue.

2. Give away the crown jewel for free. The artifact a robotics company would most expect to charge for — a frontier VLA foundation model trained on 10,000+ hours of robot data — Physical Intelligence open-sources under Apache-2.0. π0, π0-FAST and π0.5 weights, code and fine-tuned checkpoints are free to download and even free to use commercially under the license. The free release buys research credibility, an ecosystem (openpi adoption), and recruiting reach, while the monetization question stays entirely deferred.

3. Valuation as the “value metric,” not price-performance. With a who’s-who founding team and a cap table spanning Jeff Bezos, OpenAI and CapitalG, Physical Intelligence is underwritten on the promise of a general robot brain, not on any current unit economics. A ~$5.6B valuation with $0 product revenue is the inverse of a usage-priced business: the market is pricing the option on future embodied-AI demand, not metering today’s output. For how value metrics work when the deliverable is an outcome rather than a unit, see choosing the right usage metric.

Strengths & weaknesses

Billing UX : Physical Intelligence billing controls and transparency

• Billing controls — None exist. There is no self-serve dashboard, no usage meter, no plan-management UI, and no billing relationship to manage — the lab has no commercial product.
• Usage visibility — Not applicable. The open models are downloaded outright from openpi; any usage metering and spend visibility live in your own compute environment (your cloud / GPU bill), not in a Physical Intelligence console.
• Payment options — Not applicable. There is nothing to pay Physical Intelligence for. The open weights are free under Apache-2.0; no card checkout, invoice, or billing portal exists.
• Transparency — Maximally open on the research side (free weights, code, papers) and a complete blank on the commercial side — because there is no commercial side to be transparent about yet.

Strategic wins : Why Physical Intelligence’s pricing decisions worked

1. Open-source the model, own the ecosystem

By releasing π0 / π0.5 free under Apache-2.0, Physical Intelligence seeds an ecosystem (openpi adoption, third-party reimplementations, fine-tunes on ALOHA / DROID / LIBERO) and recruits the robot-learning community — without discounting any product, because there is no product on the price sheet. The free release is pure funnel and credibility, deferring the revenue question entirely. It mirrors how other AI labs use open releases as a go-to-market layer rather than a pricing layer.

2. Raise on promise, price later

With ~$1.07B raised at a ~$5.6B valuation against $0 product revenue, Physical Intelligence has bought itself the luxury of not pricing yet. It can pursue the hard research problem (a general robot brain) for years before committing to a monetization model — and avoid locking into a premature meter (per-robot? per-task? RaaS?) while the technology and market are still forming. See usage-based pricing strategy for why the right time to set a meter is when the value unit is stable.

3. No price, no commoditization — total optionality

Because it has never published a price, Physical Intelligence preserves complete freedom over its eventual model. Embodied-AI peers are already converging on Robotics-as-a-Service (e.g. Agility Robotics’ per-hour Digit) — but Physical Intelligence can watch which meter wins and choose later, unconstrained by a legacy rate card. Related: outcome-based pricing trends.

Areas to improve : Gaps in Physical Intelligence’s pricing approach

1. Eventually, signal how it will monetize

A complete commercial blank is fine for a pure research lab, but as the embodied-AI market matures, even a directional signal — licensing? per-robot subscription? RaaS? — would help partners and would-be customers plan. Right now there is not even a “Talk to us,” which means no inbound commercial pipeline can form. The total absence invites exactly the cost-unpredictability uncertainty that makes enterprise buyers hesitate to build on a platform.

2. Clarify the cost of running the “free” model

The open π0 / π0.5 release is a genuine asset, but “free weights” quietly shifts real cost — GPUs, robot hardware, data collection — onto the user. Clear guidance on expected inference/fine-tuning compute, or an optional hosted endpoint, would make the open release more usable and could become a transparent, self-serve on-ramp that later feeds commercial pipeline.

3. Separate the analyst conjecture from reality

A “$300/month per connected robot” figure now circulates as if it were Physical Intelligence’s price — it is not. Until the lab actually publishes a model, that vacuum gets filled by third-party guesses. A short, honest “we have not commercialized yet” statement would prevent projected numbers from being mistaken for real ones. Compare how other AI companies stage commercial transparency.

Key takeaways

1. No product is itself the story. Physical Intelligence has not commercialized — there is nothing to buy and no price to publish. pi.website/pricing is a literal 404 and the only CTA is “Join Us.” This is a step earlier than even a sales-only lab like Essential AI.
2. The crown jewel is free. π0 / π0.5 — a frontier robot foundation model trained on 10k+ hours of robot data — is open-sourced under Apache-2.0. The most monetizable artifact is given away to build an ecosystem.
3. “Free model” still costs money to run. The weights are free, but you supply the GPUs, the robot hardware, and the fine-tuning data. The $0 headline covers the artifact, not its operation.
4. Valuation, not revenue, sets the price. ~1.07 billion dollars raised at ~$5.6B with $0 product revenue — the market is pricing the option on a general robot brain, not metering today’s output.
5. Deferring pricing buys optionality. By never publishing a price, Physical Intelligence keeps total freedom to choose RaaS, licensing, or per-robot economics once the value unit stabilizes — a deliberate luxury its funding affords.

UBP implications

1. Sometimes the right move is to not price at all — yet. When the value unit is still unstable (what is the meter for a general robot brain — per-task? per-robot-hour? per-deployment?), committing to a price prematurely can lock in the wrong meter. Physical Intelligence shows that a well-funded lab can legitimately defer monetization until the unit of value is clear. See choosing the right usage metric.
2. Open releases are a distribution layer, not a pricing layer. Giving π0 away free under Apache-2.0 drives ecosystem and credibility without touching a (non-existent) revenue model. The lesson for UBP designers: decide deliberately which assets are funnel and which are revenue — they can live on entirely separate planes.
3. “Free” relocates cost rather than removing it. Free weights push compute, hardware and data-collection cost onto the user. UBP designers should be explicit about where the real cost lands, because a $0 headline that hides a substantial self-borne operating cost sets a misleading expectation the moment the buyer’s own bills arrive.

Sources

• Physical Intelligence — official site (research-only; no pricing page; “Join Us” CTA) (accessed 2026-06-14)
• Open Sourcing π0 — openpi release (Feb 4 2025) (accessed 2026-06-14)
• Physical-Intelligence/openpi on GitHub — Apache-2.0 weights & code (π0, π0-FAST, π0.5) (accessed 2026-06-14)
• Jeff Bezos and OpenAI invest in robot startup Physical Intelligence at $2.4B valuation (CNBC) (accessed 2026-06-14)
• Physical Intelligence raises $400M for foundation models for robotics (The Robot Report) (accessed 2026-06-14)
• Bezos-backed Physical Intelligence raises $600M at $5.6B valuation (Dataconomy) (accessed 2026-06-14)
• Physical Intelligence valuation, funding & news (Sacra) (accessed 2026-06-14)
• Browse the pricing blueprint corpus

Bottom line

Physical Intelligence is the corpus’s clearest pre-commercial case: a San Francisco robotics lab run by a who’s-who of robot-learning researchers (Levine, Finn, Hausman, Ichter, Vuong) and ex-Stripe operator Lachy Groom, with ~$1.07B raised at a ~$5.6B valuation (Bezos, OpenAI, CapitalG) — that sells nothing and publishes no price. pi.website/pricing is a literal 404 and the only CTA is “Join Us.” The artifact a robotics company would most expect to charge for — its π0 / π0.5 robot foundation model, trained on 10,000+ hours of robot data — it gives away free under Apache-2.0. The bet: open research buys an ecosystem and time, and deferring monetization preserves total freedom to pick the right meter once the value unit stabilizes. The cost of that bet is that there is, today, nothing to buy and nothing to benchmark — and any “$300/robot” figure online is third-party conjecture, not a Physical Intelligence price.

Want to compare Physical Intelligence against embodied-AI peers? See Figure AI, Apptronik and Agility Robotics, or browse the full pricing blueprint.