$110 BILLION, A PENTAGON DEAL, AND A WHOLE LOT OF HEADACHES TO COME
An incredibly busy week for OpenAI, both fiscally and spiritually. From closing a $110 billion round (mad) to creating a firestorm with massive government announcements. A lot is about to change rapidly.
What all this means for them, control, margins and who ends up owning the upside sits below for paid members. Sign up and understand it before everyone else does.
OpenAI raised $110 billion at a $730 billion pre-money valuation, blowing past the $500 billion secondary round from last autumn. Amazon committed up to $50 billion (with $15 billion upfront and $35 billion contingent on IPO or broadly defined AGI milestones), while NVIDIA and SoftBank each invested $30 billion. In return, OpenAI committed to consume roughly 2 gigawatts of Trainium capacity on AWS. Amazon and OpenAI also agreed to co-build a stateful runtime environment designed for persistent agent capability, while Sam Altman confirmed stateless API access remained exclusive to Azure and underlined NVIDIA systems as foundational to AI computing even as workloads expanded within AWS. An additional $10 billion from financial investors is expected before the end of March, which could push post-round valuation toward $850 billion.
ChatGPT usage is accelerating toward one billion weekly active users, roughly 350 per cent growth over eighteen months. More than 50 million of those weekly users are paying subscribers, and developer engagement via Codex has more than tripled since the start of the year to 1.6 million. Capital scale and user momentum are advancing together, but scale alone does not settle the question. Investors are underwriting future cash flows that have yet to fully materialise. One billion weekly active users creates surface area; fifty million paying users creates signal; Codex tripling to 1.6 million suggests genuine developer pull. Burn rate at multi-gigawatt compute scale is real, and OpenAI's revenue must now compound fast enough to justify a $730 billion pre-money anchor. Capital has already voted, but cash flow has not.
Hours before global markets reacted to geopolitical shock, Sam Altman announced a deal to deploy OpenAI models on the Department of War's classified network under explicit safety commitments, including bans on domestic mass surveillance and human responsibility for use of force. Context matters enormously here. Anthropic, the only commercial AI maker whose models had been approved for classified Pentagon use, was blacklisted by the Trump administration the same week for refusing to remove two contractual guardrails: no mass domestic surveillance, and no fully autonomous weapons without human oversight. Defence Secretary Pete Hegseth designated Anthropic a "supply chain risk to national security", a label previously reserved for hostile foreign entities like Huawei. OpenAI stepped in hours later, claiming its agreement preserves the same red lines Anthropic fought for, framed in language the Pentagon could accept: referencing existing law rather than writing explicit prohibitions. Whether the substance differs or merely the salesmanship remains an open question, and more than 300 Google employees and 60 OpenAI staff signed open letters demanding their employers refuse the Pentagon's terms.
SO WHAT?
Every major deal OpenAI struck this week came with strings. Amazon's $50 billion commitment is contingent on IPO or AGI milestones and locks OpenAI into 2GW of AWS compute. Azure keeps exclusive control of stateless API access. NVIDIA's $30 billion buys continued influence over the hardware layer. SoftBank returns as conviction capital. No single player owns OpenAI, but no player walked away without securing something in return. $110 billion did not just extend the runway; $110 billion bought a lot of people a seat at the table.
Now add the Pentagon. OpenAI says its classified deployment preserves the same red lines Anthropic held: no mass surveillance, no autonomous weapons, human oversight on use of force. Anthropic held those lines and got blacklisted inside a week. OpenAI framed the same commitments differently and got a contract. Whether that framing holds when the next political pressure arrives is the question nobody can answer yet, and more than 360 employees across Google and OpenAI clearly have doubts.
Here is what to watch next. Consumer conversion has to sustain at scale, because one billion weekly users means nothing if the paying share plateaus. Enterprise agents have to generate real revenue, not just pilot programmes. Defence deployment has to survive the next political cycle without reputational fallout or forced renegotiation. And OpenAI's revenue must compound fast enough to justify a $730 billion pre-money valuation at a burn rate measured in gigawatts, not just dollars.
Private markets priced OpenAI like a pre-IPO mega-cap infrastructure company this week. Capital came with milestone triggers, not blind belief. Pentagon entry came with safety commitments that mirror the ones that just got a competitor expelled from government work entirely. OpenAI has never been more powerful, more capitalized, or more exposed. Certainty is not the word for any of this.
FIELD NOTES
USERS
Do: Understand you are now using system-level infrastructure, not just a clever tool. Lock in workflows while pricing and access are still relatively generous. Know which provider sits behind the tools you rely on, because a government decision two tiers up your supply chain could force a switch you did not plan for.
Don't: Assume free or low-cost access remains the long-term default. Scale changes economics, and political exposure now changes availability.
ANALYSTS
Do: Track margin structure across Azure, AWS and NVIDIA relationships. Watch milestone triggers tied to IPO or AGI language. Factor in defence contract risk as a new variable: OpenAI's Pentagon commitments rest on political framing that could shift with the next administration or the next news cycle.
Don't: Model this like a traditional SaaS growth curve. Infrastructure businesses with sovereign clients behave differently.
DEVELOPERS/PRODUCTS
Do: Build with the assumption that OpenAI will remain embedded across multiple clouds and enterprise layers. Design for persistence and agents, not prompts. Build in switching capability, because this week proved that provider access can be revoked for political reasons, not just technical ones.
Don't: Optimise only for today's API pricing or access rules. What your provider is permitted to do matters as much as what it can do.
INVESTORS/VC
Do: Read the $730 billion pre-money as infrastructure pricing, not startup exuberance. Follow capital conditions, not just headlines, and pay close attention to how OpenAI's Pentagon safety commitments compare to the ones that got Anthropic blacklisted, because the durability of that distinction is now a material risk factor.
Don't: Confuse growth with profit, because while distribution is being proven, neither margins nor political durability are anywhere close to settled.
COMPANY ↓
OpenAI, staff/leadership profiles, board & governance, mission & values, HQ & global offices, funding rounds & investors, financial health & revenue, partnerships & alliances, organisational structure & hierarchy, regional presence & expansion.
OpenAI announced announcing $110B in new investment at a $730B pre-money valuation with support from SoftBank, NVIDIA and Amazon, to scale the infrastructure needed to bring AI to everyone. /OpenAI
OpenAI issued a joint statement with Microsoft affirming no changes to the terms of the Microsoft and OpenAI relationship. /OpenAI
OpenAI announced a strategic partnership with Amazon. /OpenAI
Sam Altman announced an agreement with the Department of War to deploy our models in their classified network. /@sama
Sam Altman told employees that the DOD is willing to let OpenAI build its own “safety stack” and won't force OpenAI to comply if its model refuses a task. /Fortune
Employees of OpenAI asked their companies to join Anthropic in refusing DOD's demands. /Bloomberg
OpenAI welcomed Arvind KC, previously at Roblox, Google, Palantir Technologies, and Meta, as their Chief People Officer. /OpenAI
OpenAI said it will make London its largest research hub outside of the US. /OpenAI
OpenAI provided an update on its mental health-related work include an update on litigation. /OpenAI
OpenAI (amongst other tech companies) was planning to sign White House's initiative to build their own electricity supply for AI data centres. /Fox News
OpenAI announced ‘Frontier Alliances’, multi-year partnerships with Boston Consulting Group, McKinsey & Company, Accenture , and Capgemini to help them deploy AI coworkers across the enterprise. /OpenAI
OpenAI announced a partnership with Pacific Northwest National Laboratory to evaluate whether coding agents can help effectively accelerate federal permitting work. /OpenAI
OpenAI provided a progress update on its build at the Stargate site in Milam County, Texas. /OpenAI
Rohan Varma joined OpenAI Codex to work on the future of agentic development. /OpenAI
OpenAI hired Meta AI researcher Ruoming Pang, who joined Meta from Apple seven months ago. /The Information
Sam Altman announced that OpenAI is looking for research recruiters who can find people who will move the frontier forward, not just fill roles. /OpenAI
An OpenAI-powered voice chatbot featured in the launch of Burger King’s ‘Patty’ in headsets that offer coaching, meal prep help, and evaluation interactions for “friendliness.” /The Verge
TECHNOLOGY ↓
ChatGPT ecosystem, DALL·E tools, Whisper audio, product roadmap, prototypes, developer access, integrations, extensions, research, innovations
OpenAI shared a new report ‘Disrupting malicious uses of AI’ featuring case studies of how OpenAI is detecting and preventing malicious uses of AI. /OpenAI
OpenAI delivered a new Stateful Runtime Environment that runs natively in Amazon Bedrock. /OpenAI
OpenAI shared the article ‘Why SWE-bench Verified no longer measures frontier coding capabilities.’ /OpenAI
OpenAI confirmed that weekly Codex users have more than tripled since the start of the year to 1.6M users. /OpenAI
ChatGPT crossed 900M weekly users and 50M paying subscribers. /OpenAI
OpenAI shared the customer story of Figma: ‘OpenAI Codex and Figma launch seamless code-to-design experience.’ /OpenAI
OpenAI introduced WebSockets in the Responses API, yielding 30% faster rollouts in Codex. /OpenAI + OpenAI
GPT-5.3-codex got 86% on IBench, beating out all other models. /@adonis_singh
On the eyebench-v2 first leaderboard, Codex-5.3 came first. /@adonis_singh
OpenAI released GPT-realtime-1.5 with improved intelligence, instruction following, and voice quality. /OpenAI + OpenAI
GPT-5.3-Codex was made available for all developers in the Responses API. /OpenAI
OpenAI asked developers, ‘What did you build with Codex this weekend? /OpenAI
OpenAI shared the news that GPT-5.3 Codex was now available in Cursor. /OpenAI
OpenAI invited its first (small) batch from the waitlist to try the Codex app on Windows. /OpenAI
Sam Altman thanked a developer for choosing Codex 5.3 over Opus 4.6. /@sama
In Atlas, OpenAI were working on landing several major features, including multi-account login, referencing multiple tabs when talking to ChatGPT, Agent mode upgrades, and Windows. /OpenAI
OpenAI launched ‘Library for ChatGPT' to save the files you upload to ChatGPT so you can find and reuse them later. /OpenAI
OpenAI added an ‘Add sources’ feature to projects in ChatGPT. /OpenAI
OpenAI integrated the ChatGPT Voice “orb” into the chat view in its latest iOS release. /OpenAI
OpenAI topped the Audio MultiChallenge speech-to-speech (S2S) leaderboard with the latest gpt-realtime-1.5 release. /OpenAI
OpenAI announced that users can now start new threads in the Codex app. /OpenAI
OpenAI announced that the ‘long-awaited theme picker’ was available in
Codex v0.105.0. /OpenAI
OpenAI’s Ed Bayes went live with Figma to talk through round-tripping between Code and Canvas. /Figma
OpenAI shared the article ‘Building frontend UIs with Codex and Figma’ with its developer community. /OpenAI
OpenAI announced that the first episode of Builders Unscripted went live with
Peter Steinberger and Romain Huet [VID]. /OpenAI
OpenAI was working on an Ad Manager platform. /@btibor91.blaho.me
OpenAI was noted as a customer of Nvidia who plan to unveil a new AI inference chip at its GTC conference in March. /WSJ
STRATEGY/IMPACT ↓
Public Policy, CSR, ESG practices, ethics & AI safety, risk management, competitive strategy, thought leadership, industry impact, market positioning & competitive strategy
OpenAI President and Co-founder Greg Brockman spoke to Rick Ruben about some intense moments at OpenAI. /Tetragrammation with Rick Rubin + @sama
OpenAI said ChatGPT refused to assist a user associated with Chinese law enforcement in planning an online campaign to discredit Japanese PM Sanae Takaichi. /Bloomberg
OpenAI fired an employee for prediction market insider trading. /WIRED
MISCELLANEOUS ↓
Legal, scandals, industrial action, revenue, profiles, public policy, CSR, ESG, corporate culture, ethics, risk management, Sam Altman investments, +
OpenAI said it will overhaul safety protocols and establish direct contact with Canadian police, after not alerting authorities about the Tumbler Ridge suspect. /Politico
OpenAI welcomed the Court's decision in the Musk vs OpenAI lawsuit to throw out the specific trade secret and employee-poaching claims. /OpenAI + @jasonkwon
OpenAI were bashed by Musk in a deposition, saying ‘nobody committed suicide because of Grok.’ /Tech Crunch
RESOURCES ↓
POVs, data visualisations & dashboards, key insights & metrics, industry reports & white papers, competitive analysis, recommended reading & book lists, analysis pieces & case studies +
Sam Altman would like to remind you that humans use a lot of energy, too. /Tech Crunch
GPT-5.2 can’t stop recommending nuclear strikes in war game simulations. /New Scientist
Thrive Capital bought shares in OpenAI at a fraction of the current valuation talks. /WSJ
Inside OpenAI’s scramble to get computing power after Stargate stalled. /The Information
Workday CEO said OpenAI uses his software tools. /Bloomberg
OpenAI reached an agreement with the Pentagon to use AI models. /axios
Not subscribed yet?
Join the thousands who are getting updated every week.
Try it for a month. Cancel anytime.
