The Agent Infrastructure Landscape: What You Need to Know About Fly.io, Render, Modal, E2B, Daytona, Cloudflare, and What Just Changed This Week
Prepared for: NK, Founder, CreateOS Context: In-flight follow-up after confirmation that router.nodeops.network (inference routing) and MPP (live agent payment system) are live Date: April 16, 2026 Urgency: High — landscape shifted materially this week
The news that changes your Phase 3 timeline
Three days ago, on April 13, 2026, Cloudflare launched "Agents Week 2026" and made a concentrated set of announcements that directly overlap with CreateOS's strategic territory. Matthew Prince publicly framed Cloudflare as "the definitive platform for the agentic web." They shipped Dynamic Workers (isolate-based runtime at 100x container speed), Artifacts (Git-compatible storage for agent-generated repositories at tens-of-millions scale), Sandboxes (persistent Linux environments for agents) into general availability, the Think framework for long-running agents, and announced integration with OpenAI's GPT-5.4 and Codex. OpenAI also just shipped the Agents SDK with built-in support for Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel as default sandbox providers.
CreateOS is not on that list. Neither is any agent-payment rail provider. This is both the most important threat and the most important opportunity in this addendum.
The threat is that if you wait until Phase 3 (months 7-9 of the main playbook) to establish your agent-economy positioning, Cloudflare, AWS, and the foundation model companies will have already absorbed the category. The opportunity is that the current moment — right now, April 2026 — is the window where being added to the OpenAI Agents SDK's default sandbox provider list could generate more distribution than your entire Phase 2 marketplace plan combined. This is not hypothetical. Getting onto Anthropic's official integrations list is the same kind of move. You are one BD conversation away from being the default agent deployment target in the most influential developer SDK in the world.
I am going to argue below that the agent-infrastructure integration wedge should be accelerated from Phase 3 into the next 30 days. But first, here is the research you asked for.
Fly.io: The edge-first cult favorite that plateaued
The founding and the bet
Fly.io was founded in 2017 (incorporated 2016) by Kurt Mackey, Thomas Ptacek, Jerome Gravel-Niquet, and Michael Dwan. Kurt had previously founded MongoHQ (later Compose), which sold to IBM in 2015 — so he came in with credibility and an exit. They went through Y Combinator Winter 2020 with the stated ambition of being a "Global Heroku." Their core technical bet was edge computing: run your app in 30+ regions simultaneously, close to users, using Firecracker microVMs.
The strategic miss they publicly admitted
Here is the part of Fly.io's story that is most instructive for you. In their 2023 "We Raised A Bunch Of Money" blog post, Kurt Mackey openly admitted that they had misread the market. They thought the wedge would be edgy real-time apps (WebRTC, video, game servers). What actually happened is that when they launched on Hacker News, everyone wanted to talk about databases — specifically, running Postgres close to users. Mackey wrote that the team "evolved, and got religion about a particular vision of what we're building." But the evolution took years, and the window for Fly.io to define the category closed while they were still finding their narrative.
The numbers
Fly.io has raised $110.5M to $115M across four rounds. The most recent Series C in June 2023 was $70M led by EQT Ventures with a. $467M valuation. Revenue was $882K in 2021, $7.6M in 2023, and $11.2M in 2024. They have 60 employees. The most recent raise was 3 years ago. They have not closed a round in the AI-wave era, and their valuation is now materially below Railway ($100M raise at undisclosed but higher valuation in Jan 2026), Render ($1.5B valuation Feb 2026), and other deployment platforms.
Why this matters for CreateOS
Fly.io had three factors going for it: great tech (WebContainers-equivalent microVMs years ahead of market), great team (exit-level founder), and strong developer sentiment (loved on Hacker News). They still ended up growing slowly by the standards of this era because they never nailed the narrative. The lesson is direct: technical excellence without a sharp narrative produces $11M ARR on $115M raised. Narrative discipline is not optional.
The secondary lesson is about positioning in the agent era. Fly.io's Firecracker microVMs are, technically, a perfect match for agent sandbox workloads — they are what E2B is built on. But Fly.io did not reposition for this market. E2B did. E2B is now on the OpenAI Agents SDK default list, positioned at 88% of Fortune 100 signup, and in the middle of everyone's agent stack conversation. Fly.io has essentially the same underlying technology and is nowhere in that conversation. Positioning beats technology.
Render: The un-sexy Heroku replacement that just hit $1.5B
The founding and the bet
Render was founded in 2019 by Anurag Goel, a former Stripe engineer. The positioning was deliberately boring: "Heroku, but modern." Goel did not try to invent a new category. He looked at Heroku's stagnation under Salesforce, the free-tier removal in November 2022, and the pile of frustrated developers looking for an alternative, and he built exactly what they wanted — a clean, developer-friendly PaaS with predictable pricing, native databases, cron jobs, background workers, and static site hosting.
The trajectory
Render grew from roughly 2 million developers in January 2025 to 4.5 million by early 2026. More than 250,000 developers sign up monthly. Revenue growth is "well above 100%" per their CEO. In February 2026 they raised $100M at a $1.5B valuation, bringing total funding to $258M. OpenAI's Codex coding app lets users deploy apps directly to Render as one of the default targets (alongside Cloudflare, Netlify, and Vercel). Shortcut, an AI spreadsheet startup, publicly moved from Vercel to Render. Salesforce just announced they are backing away from new features for Heroku, accelerating the exodus.
The specific thing Render did right
Render did not try to be innovative. They looked at a category with a massive pool of existing users (Heroku had hundreds of thousands of paying customers), identified every specific pain point (30-second request timeout, 24-hour forced restarts, expensive add-on marketplace, no persistent disks, 10x pricing vs. modern platforms), and built a product that fixed each one. They published a detailed comparison page called "Render vs Heroku" that became a significant acquisition channel. They did not need a viral tweet. They needed Salesforce to keep underinvesting in Heroku, which Salesforce obligingly did.
Why this matters for CreateOS
Render's story is the opposite of Bolt's. No hockey stick. No single detonation moment. A patient five-year build, capturing a displaced user base from a declining incumbent, compounded into a $1.5B valuation. The specific CreateOS lesson: you do not need a hockey stick to build a billion-dollar business. You do need a clear category exit narrative (who loses when you win) and a product that fixes every specific pain point of the incumbent. If CreateOS's eventual narrative is "we are the deployment platform for the post-Vercel, post-AWS agent era" then you need the specific, itemized list of what Vercel and AWS get wrong for agent workloads, and a public positioning document that enumerates them.
The secondary lesson is the OpenAI Codex deployment integration. Render is on it. Cloudflare is on it. Netlify is on it. Vercel is on it. CreateOS is not. This is a single-BD-conversation problem worth solving.
Modal Labs: The AI-native infrastructure rocket
The founding and the bet
Modal was founded in 2021 by Erik Bernhardsson (ex-Spotify data team lead, ex-Better.com CTO) and Akshat Bubna. Their bet was explicit and early: traditional cloud infrastructure (Lambda, Kubernetes, containers) was not designed for AI workloads. AI needs fast GPU cold starts, Python-first developer experience, and serverless scaling without YAML. Modal built "infrastructure from code" — you decorate a Python function and it runs on GPUs across clouds.
The trajectory
Modal raised $111M total across multiple rounds. $87M Series B in September 2025 at $1.1B valuation (unicorn). Currently in talks for a new round at $2.5B valuation (February 2026 reporting) led by General Catalyst. ARR of approximately $50M. Hundreds of enterprise customers, including Suno, Cursor, Anthropic, and others using Modal for evals, RL environments, and MCP servers. Sub-second cold starts, scale to thousands of GPUs, multi-cloud GPU pool.
The specific thing Modal did right
Modal's wedge is that they built the infrastructure that AI teams actually wanted — Python-first, no YAML, no Docker config, no Kubernetes. When an AI team needs to scale a training job or evaluate a model, Modal is the frictionless answer. This is the same lesson as Supabase in Addendum 1: they made themselves the easiest integration, and the fastest-growing companies in AI adopted them. Cursor uses Modal. Suno uses Modal. These customers themselves are growing at breakout rates, so Modal grows with them.
Why this matters for CreateOS
The most important thing Modal proves is that the agent-era infrastructure market is real and rewards specialization. Modal's $50M ARR and prospective $2.5B valuation came from serving a specific workload (AI inference and training) with a specific developer experience (Python-decorated functions) for a specific customer (AI teams). They did not try to be everything. They tried to be the best at one thing.
The CreateOS-specific question this forces: what is the one thing that CreateOS is unambiguously the best at? The main playbook recommendation was a keystone event around prompt-to-live-app. After seeing Modal's arc, I would argue harder that your one thing should be "the agent economy" — specifically, the combination of inference routing via router.nodeops.network plus agent payment rails via MPP plus marketplace distribution. No other platform in this list has all three. That is your Modal-equivalent specialization.
E2B and Daytona: The agent sandbox wedge you could have owned
The E2B trajectory
E2B was founded as an open-source cloud infrastructure for AI agent code execution. They use Firecracker microVMs (the same underlying tech as Fly.io and AWS Lambda), with approximately 150ms cold starts. Their specific wedge: give every AI agent a secure, isolated sandbox to execute untrusted code. They claim 88% of the Fortune 100 has signed up to use E2B for agentic workflows. They are on the OpenAI Agents SDK default list. Pricing: $100 hobby credit, $150/month Pro, usage-based.
The Daytona trajectory
Daytona pivoted in early 2025 from developer environment management to AI agent sandbox infrastructure — a narrow, well-timed repositioning. They raised $24M Series A in February 2026 led by FirstMark Capital, with Datadog and Figma Ventures as strategic investors. Matt Turck joined the board. They hit $1M forward revenue run rate in under 3 months and doubled it six weeks later. Customers include LangChain, Turing, Writer, SambaNova, and Fortune 100 enterprises. Use cases: code execution, computer use, reinforcement learning. Sub-90ms sandbox creation (some configurations 27ms). Persistent sandboxes with auto-stop, auto-archive, auto-delete policies.
What both did right
Both E2B and Daytona identified a narrow, rapidly emerging workload — AI agents need somewhere safe to run generated code — and built a purpose-built product for exactly that workload. They got onto the default list in OpenAI's Agents SDK. They got written up in every agent-infrastructure comparison article. They showed up wherever developers were searching for "where do I run my agent's code?" E2B in particular is now structural infrastructure: if you build an AI agent in 2026, you will use E2B or a direct competitor within your first month.
Why this matters for CreateOS — this is the most important paragraph in this addendum
You have an agent payment system (MPP) that is live. You have inference routing (router.nodeops.network) that is live. You have a marketplace with creator revenue share that is live. You have deployment infrastructure. You have templates. None of your competitors in the sandbox space have the payment layer. E2B has code execution but no monetization. Daytona has persistent environments but no payment rail. Modal has inference but no marketplace. Cloudflare has compute but is just now figuring out monetization. You have the complete agent economic loop — execution, persistence, payment, distribution — and nobody else has assembled it.
What you lack is the sandbox positioning. You are not currently framed as "where agents run code." You are framed as a unified workspace that happens to include agent deployment. That framing does not get you onto the OpenAI Agents SDK default list. Here is the reframe: CreateOS is the only AI agent sandbox with built-in payment rails and marketplace distribution. Lead with that. Every other agent sandbox (E2B, Daytona, Modal, Cloudflare, Vercel) is a subset of what you offer. Make the subset argument publicly.
Cloudflare: The threat you need to take seriously right now
What just happened on April 13, 2026
Cloudflare declared themselves "the definitive platform for the agentic web." They shipped Dynamic Workers (isolate-based compute for agent-generated code at 100x container speed), Artifacts (Git-compatible storage for agent repositories at tens of millions of repo scale), Sandboxes (persistent Linux environments) into general availability, and the Think framework for long-running agents. OpenAI's Rohan Varma, who runs Codex product, publicly endorsed them. They acquired Replicate and rolled that model catalog into Workers AI. They now have AI Gateway aggregating 14+ model providers. This is a coordinated, fully-funded, publicly-traded-company play to own the agent infrastructure layer.
Why this is a serious threat
Cloudflare has four structural advantages you cannot match: they own 20% of internet traffic (distribution at zero cost), they have 9 years of Workers infrastructure (a brand developers trust), they are publicly traded (access to capital and narrative platform), and they have genuine technical differentiation in V8 isolates (millisecond cold starts at container costs). Their Dynamic Workers pricing is aggressive — "100x the speed and a fraction of the cost of containers" — and they are explicitly targeting the "dozens of personal agents per user" future.
Why Cloudflare is also beatable
Cloudflare has two weaknesses you can exploit. First, they are horizontal — they want to serve every use case, which means their agent-specific features are less opinionated than a specialist could build. Second, they do not have a built-in monetization layer for the agents themselves. Cloudflare's model is "we charge for compute and storage." Your model can be "we charge for compute and storage AND let the agent earn money via MPP." An agent deployed on Cloudflare is a cost center. An agent deployed on CreateOS can be a revenue center. That is a structural product difference.
The specific Cloudflare tactic to steal
Cloudflare's "Agents Week" launch pattern is worth studying. They did not ship features incrementally. They bundled seven product announcements into one coordinated week, with a keynote narrative ("definitive platform for the agentic web"), strategic partner quotes (OpenAI), and daily blog posts. This is the Capital-L Launch pattern from the main playbook and Addendum 1, executed by a $70B market cap company. If you plan your MPP + router + marketplace launch as a coordinated week with similar discipline, you can get 1-3% of the attention Cloudflare just got, which is still significant distribution for CreateOS.
The three other players you should know about
Blaxel
Blaxel is a relative newcomer with 25ms sandbox standby resume times — the fastest in the benchmark set. They are on the OpenAI Agents SDK default list. Small company, narrow wedge, likely a partnership or integration target rather than a competitor.
Runloop
Runloop is positioned as agent-native development infrastructure. Also on the OpenAI Agents SDK list. Narrow but growing.
Baseten / Fireworks AI / Together AI
These are specifically AI inference infrastructure providers, not general app deployment. Baseten recently raised $300M at $5B valuation. Fireworks at $3.4B. Together at $3.5B. This is the inference market specifically, which is adjacent to but distinct from your app deployment market. Worth monitoring because if you expand router.nodeops.network into a standalone product, these are your competitors.
The updated competitive map
Here is the cleanest way to see the landscape, organized by what each player primarily offers.
The classic deployment incumbents (Heroku, now declining) and the modern deployment platforms (Render, Railway, Fly.io, Vercel) compete on developer experience and pricing. Vercel is the leader via framework lock-in (Next.js); Render is the Heroku-replacement play; Railway is the anti-marketing product-led play; Fly.io is the edge-first cult favorite that plateaued.
The AI coding and app builders (Cursor, Lovable, Bolt, v0, Replit, Base44) compete on prompt-to-app experience. Cursor owns the professional developer segment. Lovable, Bolt, v0 own the vibe-coder segment. Replit bridges both with an enterprise wedge. Base44 proved a solo founder can build a $80M company in six months in this space.
The AI infrastructure and inference providers (Modal, Baseten, Fireworks, Together) compete on GPU efficiency and developer experience for AI workloads specifically.
The agent sandbox providers (E2B, Daytona, Modal's sandbox product, Runloop, Blaxel, Vercel Sandbox, Cloudflare Sandboxes) compete on sandbox cold-start times, isolation models, and sandbox pricing for AI-generated code execution.
The foundation provider platforms (Cloudflare, AWS, GCP, Azure) compete on scale and breadth — they offer everything but without agent-specific opinionation.
CreateOS currently occupies a unique position: you are the only platform that combines agent sandbox capability (AI Sandbox), deployment infrastructure (CreateOS core), inference routing (router.nodeops.network), payment rails (MPP), and marketplace distribution. Nobody else has assembled this full stack. This is your strategic moat if you articulate it clearly. If you do not articulate it clearly, Cloudflare will assemble an equivalent stack in the next 12 months and you will lose the positioning.
Updated three-factor formula for CreateOS
Revisiting the Addendum 1 formula with the new intelligence: (new primitive) × (narrow product decision) × (distribution event) = hockey stick.
Factor 1, your new primitive: No longer hypothetical. You have three distinct, live primitives that competitors do not have combined: MPP payment rails, router.nodeops.network inference routing, and the marketplace with revenue share. Any one of these alone is a narrow product. The combination is a category. The specific narrow product decision should be about how you present the combination.
Factor 2, your narrow product decision: I am now recommending a different narrowing than the main playbook. Instead of "prompt to live app in 60 seconds" (the Bolt positioning), narrow to "the only agent sandbox where the agent can earn." This is structurally defensible, references primitives only you have, and maps directly to the emerging agent infrastructure category. Every OpenAI Agents SDK conversation becomes "use E2B for execution, use CreateOS if you want the agent to earn." That is a single, sharp value proposition that no competitor currently offers.
Factor 3, your distribution event: The capital-L Launch should be coordinated around this single narrative, in a one-week Cloudflare-style blitz. Day one: MPP case study with a live agent earning money. Day two: router.nodeops.network positioning as the only inference-aware routing layer with payment awareness. Day three: marketplace refresh with public creator earnings leaderboard. Day four: integration announcements — getting onto the OpenAI Agents SDK, Anthropic integrations page, and at least one major AI coding tool as a default deployment target. Day five: founder-led narrative piece (Medium, founder blog) tying it together. This is 60-90 days from now.
Revised Phase 1 priority shift: the OpenAI and Anthropic integration moves
The single highest-leverage BD action CreateOS can take in the next 30 days is getting added to the OpenAI Agents SDK's default list of supported sandbox providers. The current list is Blaxel, Cloudflare, Daytona, E2B, Modal, Runloop, and Vercel. CreateOS is not there. The process for getting added is:
Step 1: Reach out to the OpenAI Agents SDK team via developer relations (DevRel). OpenAI has a visible DevRel presence; the public-facing developer advocates are findable on Twitter and LinkedIn.
Step 2: Build a reference integration that implements the OpenAI Agents SDK sandbox provider interface against CreateOS. This is a non-trivial but bounded engineering effort — likely 2-4 weeks of work for an engineer familiar with both systems.
Step 3: Publish a case study or demo showing an agent running in CreateOS that earns money via MPP. This becomes the distinctive angle: "Every other provider is a cost center; CreateOS makes the agent a revenue center."
Step 4: Propose inclusion in the next SDK update.
A similar motion with Anthropic's SDK and integrations list should run in parallel. Your memory notes indicate you have a relationship with the Anthropic ecosystem via Claude Code users as your ICP. Leverage that.
This work should happen in Phase 1, not Phase 3. It is higher-leverage than almost anything else in the main playbook because success means you appear in SDK documentation that tens of thousands of developers will read in 2026, with zero ongoing marketing cost. One BD win here equals a year of content marketing.
The three updated recommendations
Recommendation 1: Narrow to "the only agent sandbox where the agent can earn." This is a sharper wedge than the original Phase 1 keystone event and uses primitives only you have. Your marketing, your homepage, your product documentation, and your founder narrative should all lead with this in the next 30 days.
Recommendation 2: Accelerate the integration-wedge initiative from Phase 3 into the next 30 days. The OpenAI Agents SDK default list, Anthropic integrations page, and at least one AI coding tool (Cursor, Lovable, Bolt, v0, or Claude Code) should all be BD targets for Phase 1. Getting onto two of these four surfaces in the next 90 days is a realistic target and is more valuable than most of Phase 2.
Recommendation 3: Plan a capital-L Launch for month 2 or 3, not month 10. The main playbook placed the big launch in Phase 4. Given what Cloudflare just announced and the pace of the agent infrastructure category, waiting 10 months is too long. Plan an "Agent Economy Week" for CreateOS — seven days of coordinated product announcements, customer case studies, integration news, and founder content — for month 2 or 3 of the playbook. This is a budget and execution question, not a strategic one.
What I got wrong in the main playbook
Two things in the original document need updating given this research.
Update 1: The original Phase 3 sequenced the MPP Gateway enterprise narrative for months 7-9. Given that Cloudflare just declared themselves the definitive agent platform and OpenAI shipped an Agents SDK with a default provider list, this timing is wrong. The MPP Gateway narrative needs to lead Phase 1, not Phase 3. The original playbook underweighted how fast the agent category was consolidating.
Update 2: The original playbook recommended picking one keystone event, with the default recommendation being the Bolt-style "prompt to live app in 60 seconds." Given that router.nodeops.network and MPP are live, I now recommend "the only agent sandbox where the agent can earn" as the keystone positioning. This uses strictly differentiated primitives and plays in a category that is structurally rewarding specialists right now. The Bolt-style positioning puts you in direct competition with Lovable and v0 where they have 18-24 months of head start. The agent-economy positioning puts you in a category that does not yet have a clear leader.
Final note on the landscape
I want to close with the most important observation in this addendum. In the past six weeks, the following has happened:
- Feb 5, 2026: Daytona raised $24M Series A for AI agent infrastructure.
- Feb 12, 2026: Modal Labs in talks for $2.5B valuation on $50M ARR.
- Feb 17, 2026: Render raised $100M at $1.5B valuation, revenue growth above 100%.
- April 13, 2026 (three days ago): Cloudflare declared themselves "the definitive platform for the agentic web" with seven coordinated product launches.
- April 2026 (this week): OpenAI shipped Agents SDK with default sandbox provider list including E2B, Modal, Daytona, Cloudflare, Vercel. CreateOS not on list.
The category is consolidating in real time. The companies with live agent infrastructure that is publicly visible are the ones getting onto default integration lists, the ones raising at premium valuations, and the ones that will own the next wave. You have the product — MPP and router.nodeops.network are genuinely differentiated — but you do not yet have the positioning, the integrations, or the launch moment that makes the product visible to the ecosystem.
The window to claim the "agent economy" position is measured in months, not quarters. Execute accordingly.
End of Addendum 2. The next research cycle should focus on specifically named BD targets within the OpenAI, Anthropic, and Cursor teams, plus a competitive response plan for Cloudflare's Agent Cloud.