OpenAI raised up to $110 billion last week. The pre-money valuation was reported at $730 billion. Amazon, Nvidia, and SoftBank were cornerstones. Those numbers are hard to hold in your head without losing their meaning, so try it this way: that single financing round is larger than the GDP of most countries. It rewrites what “private company” means.
And yet the rest of venture kept moving. A three-year-old AI startup quietly ran its first employee tender at a $4.5 billion valuation. An enterprise authentication company raised $100 million on infrastructure that nobody outside of enterprise software has ever heard of. A hardware company that makes optical interconnects for data centers closed $500 million to help chip makers stop frying their own power grids.
The week had a clear shape. Frontier AI is now financed like sovereign infrastructure. Everything else is reorganizing around the question of what happens after you deploy it.
The Frontier Is Now a Capital Game
The thing that made the OpenAI round unusual was not just the size. It was the composition. Amazon and Nvidia were not writing checks the way traditional investors do. They were making infrastructure commitments that blur the line between equity and partnership. That is a different kind of investor relationship, and it changes how secondaries in frontier AI get underwritten. When a company’s valuation is partly a function of compute access and distribution deals, not just revenue multiples, the usual models do not quite fit.
Separately, Reflection AI is reportedly in discussions for a round that would value it above $20 billion. The company is pursuing an open-weights model strategy, meaning it publishes the underlying model parameters publicly, which is the opposite of OpenAI’s closed approach. The fact that capital is chasing both strategies simultaneously tells you the market has not decided which architecture wins. That uncertainty is worth holding onto when you hear anyone claim the winner is obvious.
Liquidity Before the IPO
Stripe ran an employee tender offer last week, reportedly pricing the company at around $159 billion. Decagon, an AI customer support company founded in 2022, ran its first employee tender at $4.5 billion. Same week. Two very different companies. Same mechanism.
Tender offers work like this: the company or an outside investor buys shares directly from employees who want to sell, without going public. The price is negotiated, not set by a public market. Employees get cash. The company stays private. Investors get exposure to a company that might not IPO for years.
What is changing is the cadence. Stripe has run multiple rounds of secondaries over the past few years. Decagon is three years old and already running one. The message from founders is increasingly: we can give employees liquidity and still avoid the quarterly earnings circus. For investors, this creates a repeating opportunity, not a one-time event. The corollary is that IPO timing becomes less of a forcing function and more of an option that companies exercise when conditions are genuinely favorable, not when their employees need cash.
StepStone, a large private markets firm, also disclosed a fund-level redemption offer during the week, allowing a small percentage of its private venture fund to be redeemed at net asset value. Institutionalized liquidity mechanisms are spreading from company-level tenders to fund-level structures. The private market is slowly building plumbing that makes the absence of public-market liquidity feel less like a bug.
The Infrastructure That AI Actually Runs On
WorkOS raised $100 million at a $2 billion valuation. If that name is unfamiliar, that is part of the point. WorkOS sells single sign-on, directory synchronization, and audit logging to companies that build software for enterprises. In plain language: when your company’s IT department wants to control who can log into a product, WorkOS is the layer that makes that possible. It is the kind of company that never appears in a headline until it raises a meaningful round.
Why does this matter now? Because as companies deploy AI agents, meaning software that acts on your behalf rather than just responding to your questions, the question of who authorized that agent to do what becomes a compliance requirement. Every enterprise deploying an AI agent into their workflow is going to need audit trails and access controls. WorkOS is essentially a tax on AI adoption, collected at the infrastructure layer. The $2 billion valuation reflects that.
Ayar Labs closed $500 million at roughly $3.75 billion. They make co-packaged optics, which are optical interconnects that move data between chips at very high speed without the power consumption of conventional electrical connections. The problem they are solving is concrete: data centers running large AI models are hitting physical limits. The chips can process, but the wires between them cannot move data fast enough without generating heat that the power grid cannot support. Ayar is working on a genuine bottleneck, not a speculative one.
JetStream Security raised a $34 million seed round led by Redpoint, with participation from CrowdStrike’s investment fund. They are building governance tooling for AI agents in enterprise environments. Guild.ai raised $44 million for agent orchestration and operations. The pattern is consistent: every company deploying AI agents needs to answer the question “what is it doing and why,” and right now the tooling to answer that question barely exists.
Defense, Governance, and the Cost of Taking the Money
OpenAI published the terms of its Pentagon contract last week, including a set of prohibitions it negotiated: no mass domestic surveillance, no fully autonomous weapon systems, no high-stakes automated decisions without human oversight. Then Caitlin Kalinowski, who led OpenAI’s robotics and hardware efforts, resigned in direct response to the Pentagon deal. Within days of the contract details becoming public.
This is what the “defense AI” opportunity actually looks like in practice. The revenue is real. The procurement dollars are large. But the stakeholder surface area, meaning employees, enterprise customers, regulators, and journalists, is enormous. A company that takes a defense contract cannot un-take it. The reputational and talent costs can move faster than the revenue.
For founders and early-stage investors, the practical question is not whether to engage with government. It is whether you have thought carefully about what your usage policy actually says, and whether you can defend it to the people who work for you. OpenAI had published red lines and still faced leadership exits. The implication is that process alone is not enough. Your people have to agree with the decision, not just be informed of it.
Related: the US government is reportedly drafting procurement guidelines that would require AI contractors to allow “any lawful” use of their models. That phrase is doing a lot of work. It would effectively prohibit the kind of conditional usage restrictions that Anthropic and others have built into their deployment terms. Startups selling into federal channels should watch this closely. The choice between accepting government procurement terms and maintaining your commercial customer commitments may arrive sooner than expected.
Fintech: Staying Private, Going Deeper
Revolut publicly confirmed that it filed for a US bank charter and named a new US chief executive. This is a meaningful strategic move. Revolut currently operates in the US through partnerships with licensed banks, paying fees to access payment rails and deposit infrastructure it does not own. A bank charter would let it control those rails directly.
The timing is deliberate. The current regulatory environment in the US is more permissive toward fintech applications than it has been in years. Revolut is betting that the window is open. If they are right, they gain structural cost advantages and the ability to offer credit products that require a banking license. If the window closes before approval, they will have spent years in the charter process for nothing. That is the kind of strategic bet that either looks prescient or reckless depending on what happens next.
Clear Street, which had previously filed to go public, asked the SEC to deregister its shares and withdrew its IPO registration. The company cited market conditions. The practical meaning is that even companies close to the public markets are choosing to stay private when the conditions are not favorable. This adds secondary market pressure: investors and employees in companies like Clear Street are not getting IPO liquidity, which means they need other options.
Xflow raised a $16.6 million Series A led by General Catalyst, with PayPal Ventures participating, to expand cross-border business payments. The company had just received regulatory authorization for cross-border payment flows. This is the classic milestone-driven fintech story: you raise money after you have the license, not before, because the license is the hard part.
The Macro Backdrop
The US Labor Department reported that nonfarm payrolls fell by 92,000 in February, with unemployment rising to 4.4 percent. Separately, oil crossed $90 per barrel for the first time in nearly two years, driven by supply disruptions from the conflict in Iran.
For late-stage private markets, this combination matters in a specific way. When rate cut expectations become uncertain, meaning investors are not sure when or whether the Federal Reserve will reduce borrowing costs, the willingness to pay up for illiquid, long-dated assets shrinks. Secondaries in private companies already (sometimes) price at discounts to the last primary round. In a volatile macro environment, those discounts widen. Buyers ask for more protection: preferred return structures, price ratchets tied to future events, or earn-outs that keep value at risk until liquidity materializes. The deals still happen. They just get more complicated.
M&A: What the Strategics Are Buying
Thoma Bravo announced it would acquire WWEX Group, a freight logistics network, and merge it with Auctane, which owns ShipStation, Stamps.com, and several other shipping software platforms. The transaction was financed with roughly $5 billion in private credit from Blackstone and Ares. Existing WWEX investors are rolling equity into the combined company.
The thesis is straightforward: combine the software layer with the network layer and you control both the interface and the execution. The rollover equity is a signal that existing investors believe the combination is worth more than the exit. That is worth noting in an environment where buyout exits often involve sellers taking everything off the table.
Accenture agreed to acquire Ookla, which operates Speedtest and Downdetector. Ookla generates data about network performance and outages at global scale. Accenture is buying that data to support enterprise clients managing distributed digital infrastructure. The pattern here is that telemetry and monitoring data, the exhaust of running large systems, is increasingly valuable as an input to AI-based operations.
Thomson Reuters acquired Noetica, a company that builds market intelligence tools for legal and financial professionals. AI trained on deal data, embedded into the workflow where lawyers and bankers are already spending their day. That is the right way to deploy AI in a professional services context: not a new product your users have to learn, but a better version of something they already use.
What This Week Means for Early-Stage Investors
The most durable pattern from the week is not the OpenAI or Anthropic headlines. It is the consistency of what got funded below the headline: governance tooling, access control, orchestration, audit infrastructure. Every company on that list is solving a problem that gets more expensive to ignore as AI deployment scales. WorkOS, JetStream, Guild.ai. These are picks-and-shovels bets on the idea that enterprises will deploy AI agents at scale and will need the compliance infrastructure to do it safely.
The secondary liquidity story is separate but related. Stripe and Decagon are not similar companies, but they both ran tenders this week. The mechanism is normalizing. For a syndicate that sources secondary positions, the implication is to build sourcing infrastructure for tender processes, not to wait for IPOs. That means knowing which companies have recurring tender programs, which law firms run them, and which boards are open to giving early-stage investors access.
On governance: the OpenAI Pentagon situation should be read as a warning, not an anomaly. The founders who will navigate this decade well are the ones who can articulate their usage policy before they have government customers, not after. That includes being honest about which contracts they would and would not take. It is easy to say “we have red lines” when no one is testing them. The test comes with the revenue.
The macro environment does not change the investment thesis. It changes the terms. Secondaries get more structured when bid-ask spreads widen. That is a reason to be more disciplined about what you underwrite, not to stop underwriting. The companies that will trade well through volatile macro are the ones with clear paths to liquidity, strong governance, and valuations set by something more than hope.
Sources
OpenAI funding announcement: openai.com/index/scaling-ai-for-everyone
OpenAI $110B round: techcrunch.com/2026/02/27/openai-raises-110b-in-one-of-the-largest-private-funding-rounds-in-history
OpenAI backers: apnews.com/article/openai-amazon-nvidia-softbank-altman-microsoft-a0a915c32b85337d799fe2f9525a932a
Stripe tender: ft.com/content/42639c42-2e95-42e7-aacf-3a39743657f5
OpenAI Pentagon details: techcrunch.com/2026/03/01/openai-shares-more-details-about-its-agreement-with-the-pentagon
Decagon tender: techcrunch.com/2026/03/04/decagon-completes-first-tender-offer-at-4-5b-valuation
WorkOS Series C: workos.com/blog/series-c
Ayar Labs Series E: ayarlabs.com/news/ayar-labs-closes-500m-series-e-accelerates-volume-production-of-co-packaged-optics
Revolut US bank charter: revolut.com/en-US/news/revolut_files_u_s_bank_charter_application_names_new_u_s_ceo
BLS February payrolls: bls.gov/news.release/archives/empsit_03062026.htm
FirmPilot Series A-1: prnewswire.com/news-releases/firmpilot-closes-oversubscribed-22m-series-a-1
JetStream seed: newswire.com/view/content/cybersecurity-heavyweights-launch-jetstream-with-34m-seed-round-to-22734722
Alibaba Qwen lead: techcrunch.com/2026/03/03/alibabas-qwen-tech-lead-steps-down-after-major-ai-push
Xflow Series A: newsroom.paypal-corp.com/2026-02-24-Xflow-Raises-16M-Series-A
Clear Street deregistration: clearstreet.io/news/press-releases/clear-street-announces-voluntary-sec-deregistration
Thoma Bravo / WWEX / Auctane: thomabravo.com/press-releases/thoma-bravo-to-acquire-wwex-group-and-combine-with-auctane-to-form-global-logistics-leader
Blackstone/Ares WWEX financing: news.bloomberglaw.com/private-equity/blackstone-ares-provide-5-billion-for-thoma-bravo-wwex-buyout
Accenture / Ookla: newsroom.accenture.com/news/2026/accenture-to-acquire-ookla
Thomson Reuters / Noetica: legal.thomsonreuters.com/blog/thomson-reuters-acquires-noetica
Reflection AI: ft.com/content/07073c8f-7176-471c-ac69-ef1458845fb2
Guild.ai: axios.com/pro/enterprise-software-deals/2026/03/03/guildai-khosla-agentic-enterprise-tech-ai
Flux: flux.ai/p/blog/we-raised-37m-to-take-the-hard-out-of-hardware
NationGraph: axios.com/pro/enterprise-software-deals/2026/02/25/govtech-nationgraph-18-million-menlo
StepStone redemption offer: go.stepstonegroup.com/rs/664-QRQ-541/images/SPRING_US_Redemption_Offer_to_Purchase.pdf
US AI procurement guidelines: economictimes.indiatimes.com/tech/artificial-intelligence/us-draws-up-strict-new-ai-guidelines-amid-anthropic-clash-ft
SEC tender offer CDIs: sec.gov/rules-regulations/staff-guidance/compliance-disclosure-interpretations/tender-offer-rules-schedules
Iran oil shock: investing.com/news/commodities-news/us-pump-prices-surge-as-iran-war-upends-global-energy-supply-4548254
OpenAI Kalinowski exit: techcrunch.com/2026/03/07/openai-robotics-lead-caitlin-kalinowski-quits-in-response-to-pentagon-deal
Loveholidays IPO delay: ft.com/content/1325d4ff-34ae-4d34-8b34-b3a7ef7d5acc
Moneyview DRHP: m.economictimes.com/tech/technology/fintech-lender-moneyview-files-drhp-with-sebi-for-ipo