Paperspace
Winter 2015AcquiredPaperspace is a cloud platform for building and scaling AI…
Pressure-test this opportunity
Turn this teardown into a decision-ready prompt for ChatGPT, Claude, or your agent.
Paperspace is a cloud platform for building and scaling AI…
Turn this teardown into a decision-ready prompt for ChatGPT, Claude, or your agent.
If you only have a few minutes to spare, here’s what investors, operators, and founders should know about Paperspace (W15).
Paperspace began in 2014 with an unfashionable premise: cloud computing had become powerful before it became pleasant. Architecture graduates Dillon Erb and Daniel Kobran wrapped GPU-backed virtual machines in a browser-based desktop, followed their users from CAD and rendering into machine learning, and turned that wedge into Gradient, a development platform spanning notebooks, training jobs, and deployment.[1]
DigitalOcean bought Paperspace for $111 million in cash in July 2023 and folded the brand into DigitalOcean in 2024.[2][3] This was not a shutdown disguised as an acquisition. Paperspace built valuable software and scarce GPU capacity. Its strategic limit was that simplicity alone could not remove the capital, supply-chain, and distribution advantages of a larger cloud. The buyer supplied all three.
Erb and Kobran met in the University of Michigan's architecture program, where both earned Master of Architecture degrees in 2014. Their studio work mixed design with computation. Structural simulations and complex CAD workflows required powerful machines, yet the tools for reaching that compute were difficult to configure. They saw access as a design problem.[1][4]
Their training also gave them unusual confidence in each other. Erb later said: "The best way to find a co-founder is to stay up all night in an architecture studio." The pair began Paperspace in fall 2014, only months after graduation, and entered Y Combinator's Winter 2015 batch. Advisors included Garry Tan, Alexis Ohanian, and Justin Kan.[1]
The first product was a personal computer in the cloud. Users opened a Windows or Ubuntu desktop in a browser, retained files and settings, and upgraded the remote hardware without replacing the device in front of them. A $50 puck called Paperweight acted as a zero client for monitors that did not already have a suitable computer. Service pricing was expected to begin around $10 a month.[5]
Paperspace did not invent virtual desktops. It removed installation and management work from a category dominated by Citrix, VMware, and Amazon. That distinction came directly from the founders' design background. Kobran later described their broader goal in one sentence: "You shouldn't need decades of experience to build intelligent applications."[6]
Paperspace's first system combined virtual machines, dedicated GPUs, storage, and a low-latency remote display. The browser client used technologies including WebSockets, WebGL, and JavaScript rendering to stream media-rich desktops. Enterprise features later included machine cloning, Active Directory and VPN integration, backups, monitoring, templates, and shared drives.[5][9]
Performance forced an important architectural decision. The founders found public-cloud VMs too slow for their target experience, so Paperspace operated GPU servers in colocation facilities, initially in California and New York. That gave the company control over latency and hardware but made growth capital intensive.[9]
Gradient translated the same design philosophy from desktops to machine learning. Developers could start Jupyter notebooks from templates, attach GPUs, run containerized jobs, store datasets, automate workflows, and deploy models. Community Notebooks made public projects forkable and supplied free GPU tiers. Gradient could also run as a managed service or on Kubernetes outside Paperspace's own cloud.[11][14]
The product line broadened across Core virtual machines, Gradient notebooks and workflows, and Workstream cloud desktops. That reach created useful cross-pollination, but it also produced navigation and abstraction friction for expert customers who wanted a GPU, SSH access, and transparent storage behavior.
Paperspace entered through architects and designers with expensive workstation needs, then expanded toward teams in regulated industries, rendering, gaming, scientific computing, and machine learning. The decisive shift came from observed demand: by 2016-2017, deep-learning developers had become its largest user group.[1]
Gradient targeted individual learners with free notebooks while selling paid GPU time and team tooling to startups and enterprises. By 2021, a university profile listed HBO, Deloitte, and Dropbox among customers, alongside a team of more than 30.[4]
The company did not publish a defensible standalone market-size figure. Its outcome gives a better demand signal. DigitalOcean paid $111 million specifically to add GPU infrastructure and AI software, then made Paperspace the foundation of its AI/ML offering.[2][15]
Current pricing shows a mature, segmented market rather than a single commodity. DigitalOcean lists on-demand GPUs from $0.76 an hour for RTX 4000 Ada to $3.44 for H200, with reserved and multi-GPU configurations above that.[16] Buyers compare chip, memory, capacity, storage, network, software environment, and billing behavior, not merely hourly price.
Early Paperspace competed with Citrix, VMware, Amazon WorkSpaces, and local workstations. Gradient later faced AWS, Google Cloud, Azure, Google Colab, Lambda, CoreWeave, RunPod, Vast.ai, and notebook startups. Hyperscalers owned breadth and enterprise trust. Specialist clouds competed on GPU availability and price. Colab made notebooks familiar and cheap.
Paperspace's position was simplicity plus control of the underlying GPU fleet. The company also expanded supply through Nvidia's partner network and a Graphcore relationship. Its weakness was the seam between beginner-friendly abstraction and expert workflows. A 2023 Hacker News commenter complained about slow dataset transfers, confusing GUI layers, outdated documentation, and difficult SSH access. That is anecdotal, not a user survey, but it names the tension precisely: hiding infrastructure helps a newcomer until the abstraction hides something an expert needs.[17]
Paperspace earned usage revenue from GPU and CPU machines, storage, and higher-tier Gradient subscriptions. Free notebooks acted as education and acquisition. Enterprise contracts added private deployments, team controls, and support. The company also had to fund servers and colocation capacity before customers consumed it.
Funding helped bridge that mismatch. Paperspace raised $4 million in 2016 and a $13 million Series A in 2018, reaching $19 million of disclosed funding at that point; a 2021 university profile put the later total near $30 million.[9][10]
DigitalOcean's filing provides the clearest economics, though only after the acquisition. From July 6 through December 31, 2023, Paperspace contributed $6.35 million in revenue and an $18.914 million net loss.[15] Acquisition accounting and integration costs make that loss a poor proxy for standalone operations, but the figures show why software differentiation did not make GPU cloud cheap to operate.
The 2015 announcement produced more than 12,000 signups. Paperspace later said its users ranged from genomics and gaming to CAD. A 2021 Gradient update reported more than 10,000 notebook creators in the prior month and more than two million GPU-compute hours across the platform.[8][18]
The $111 million cash acquisition was the strongest validation. DigitalOcean described Paperspace as rapidly growing and expected the deal to improve revenue growth from 2024 onward. The buyer did not preserve the independent brand, but it retained the products and used the acquisition as the base of a broader GPU and AI portfolio.
Paperspace launched as a cloud desktop, yet its durable asset became AI infrastructure. That change was not a desperate final pivot. The founders observed deep-learning users becoming the largest group and built Gradient around their work. The original insight, simplify access to powerful GPUs, survived even as the interface changed from a desktop to notebooks and jobs.
The attempted remedy for a broad desktop market was product expansion: Teams, Core, Workstream, and Gradient. It worked well enough to create several demand channels, but Gradient became strategically more valuable than the generic desktop. The lesson is not “pivot to AI.” It is to preserve the customer problem while allowing the product category to change.
Paperspace's design advantage was removing cloud setup. Its user base eventually included engineers who wanted direct, predictable control over data, SSH, environments, and billing. More visual layers did not always make those jobs easier. The company added containers, custom images, Kubernetes, and self-hosted Gradient, but every abstraction created another contract to document and maintain.
This did not destroy the business. It limited how much software alone could distinguish rented GPUs. When expert users could switch providers for a lower price or better availability, the cloud underneath regained power over the interface above it.
Running colocation infrastructure gave Paperspace the performance public VMs could not provide in 2016. It also tied growth to GPU procurement, facilities, networking, and depreciation. The company joined Nvidia's programs and experimented with Graphcore to widen supply, but hardware access became more strategic as generative AI demand surged.
DigitalOcean's acquisition solved the scale problem through combination. Its CFO called Paperspace "a rapidly growing business with leading edge technology." The buyer had a larger customer base, self-serve distribution, broader cloud services, and a balance sheet for capacity.[2] Paperspace brought software, specialized infrastructure, and credibility with AI developers.
Folding Paperspace into DigitalOcean in 2024 could look like product failure. The evidence supports a different reading. Legacy services remained available, while new customers were routed to DigitalOcean GPU Droplets, bare-metal GPUs, inference, and agent products.[3] The independent brand was redundant once the buyer adopted its function.
The counterpoint is financial. A $111 million sale after years of fundraising was not a hyperscale outcome, and public records do not reveal proceeds to founders, employees, or each investor. It was still a real strategic acquisition with cash consideration, retained products, and a continuing role in the buyer's portfolio. Paperspace succeeded by becoming useful infrastructure inside a larger cloud, not by becoming the larger cloud itself.