AI to ROI News & Analysis: April 3, 2026

The $1 Trillion Week: OpenAI Tops $852B Valuation, SpaceX Files for the Biggest IPO in History, Anthropic Leaks Its Own Claude Code Secrets, and More

The Biggest AI News This Week

💰 OpenAI Closes $122 Billion Round at $852 Billion Valuation - the Largest Funding Event in Silicon Valley History

🚀 SpaceX Files Confidentially for IPO at Up to $1.75 Trillion Valuation - the Largest IPO in History

🔓 Anthropic Accidentally Leaks 512,000 Lines of Claude Code Source Code which included the Mythos Model and ‘Kairos’ Agent

📈 Anthropic Considers October IPO at Potential $2 Trillion Valuation - Racing OpenAI to Market

🤝 Microsoft Debuts Copilot ‘Critique’ Using Anthropic and OpenAI Models in Sequence

🏗️ Oracle Cuts Up to 30,000 Jobs to Fund $50 Billion AI Data Center Push

🐉 The Chinese AI Market Accelerates as Domestic Chip Companies Gain Traction, Models Improve, and OpenClaw Enables Global Expansion

📊 US AI Adoption Is Rising, but Public Trust Is Falling - Americans are Conflicted on Regulation

⚖️ Perplexity Sued for Allegedly Sharing User Data with Meta and Google

💼 Salesforce Adds 30 AI Features to Slack as Enterprise Platforms Race to Embed Agents

🃏 Bald eagles in love, rock & pop classics as Irish folk songs (beautiful), a lovely comic about the moon, and photos that show the awesomeness of the Artemis II mission

1️⃣ OpenAI Closes $122 Billion Round at $852 Billion Valuation which is the Largest Funding Event in Silicon Valley History

OpenAI closed a $122 billion funding round at an $852 billion pre-money valuation, the largest single funding event in Silicon Valley history. Investors include Amazon ($50B), Nvidia ($30B), and SoftBank ($30B), plus over $3 billion from retail investors via ARK Invest ETFs. The round came in $22 billion over its original $100 billion target.

OpenAI’s pivot towards the enterprise is underway. Enterprise sales now account for 40% of OpenAI’s $2 billion monthly revenue run rate, with that figure expected to reach 50% by year-end.

The Circular Funding Problem

When you look into the details of the funding round, some interesting details pop out:

• Amazon is investing $15 billion up-front. The remaining $35 billion is contingent on OpenAI either going public or achieving artificial general intelligence. Amazon’s investment also includes a cloud agreement for hosting and distributing OpenAI’s models, meaning OpenAI will spend a significant portion of Amazon’s money back with Amazon.

• NVIDIA committed $30 billion. In return, OpenAI will spend $35 billion on NVIDIA products.

Critics have pointed to this web of circular arrangements as a systemic risk if AI revenue growth disappoints.

“AI is driving productivity gains, accelerating scientific discovery, and expanding what people and organizations can build. This funding gives us the resources to continue to lead at the scale this moment demands.”

OpenAI, company statement, March 31, 2026

The Secondary Market for OpenAI Shares

Next Round Capital founder Ken Smythe said roughly $600 million of OpenAI shares recently came to market and found no buyers. In fact, OpenAI secondary bids are coming in at roughly $765 billion, a 10% discount to the primary round valuation. Next Round said it’s actually looking to buy shares in Anthropic.

“It’s just better risk-reward right now. People are betting that Anthropic’s valuation will catch up with OpenAI’s. But if you buy OpenAI shares, it’s less clear what the return will be in the near term.”

Adam Crawley, co-founder, Augment (secondary market platform), April 2026

The Takeaway

• Watch the milestone gating: Amazon’s commitment doesn’t fully vest unless OpenAI executes an IPO or achieves AGI. Enterprise AI buyers and investors should ignore the $122 billion headline. The amount of capital available to OpenAI now is much lower.

• The circular structure is a systemic risk: OpenAI is spending its investors’ money with its investors. If enterprise AI revenue growth slows, the interconnected commitments between OpenAI, Amazon, NVIDIA, and Oracle could compound rather than buffer a downturn.

• The secondary market matters: Institutional investors’ move from OpenAI to Anthropic signals a real shift in perceived enterprise value. AI solutions companies should track which platform their enterprise buyers are moving toward.

2️⃣ SpaceX Files Confidentially for IPO at Up to $1.75 Trillion Valuation - the Largest in History

SpaceX filed confidentially with the SEC for an initial public offering, targeting a valuation of more than $1.75 trillion and raising $75 billion. If SpaceX succeeds, it would be the largest IPO in history, surpassing Saudi Aramco’s $29 billion debut in 2019. The company has lined up five lead banks including: Bank of America, Citigroup, Goldman Sachs, JPMorgan, and Morgan Stanley, plus at least 16 additional banks in regional roles. SpaceX is aiming for a June 2026 listing.

The Nasdaq Rule Change

Nasdaq moved to smooth SpaceX’s post-IPO path before the filing was even confirmed:

• A rule change effective May 1 will allow newly public companies to join the Nasdaq index after just 15 days of trading, down from the previous three-month waiting period.

• The change would allow SpaceX to be included in major index funds almost immediately, generating forced buying from passive investors and supporting the share price in its early trading weeks.

The Valuation Challenge

Objectively speaking, SpaceX’s $1.75 trillion valuation target is in the stratosphere. SpaceX’s combined revenue is expected to approach $20 billion in 2026. xAI, the AI and social networking business SpaceX acquired in February, is likely generating under $1 billion, and its platform is undergoing a major rewrite. xAI’s social network, X (Twitter), isn’t profitable. SpaceX plans to use its IPO proceeds to pay down X’s debt, fund orbital AI data centers, and bankroll xAI’s AI research . These are three expensive, unproven bets layered onto its highly profitable but capacity-constrained rocket business.

“Elon is saying, ‘I have this window to fund all this craziness for a good period of time based on the SpaceX hype.’ By merging the two losers with one winner, he keeps all the balls in the air.”

Ross Gerber, investment manager and SpaceX shareholder, The New York Times, April 1, 2026

The Takeaway

• SpaceX is the first of three potential mega-IPOs in 2026: OpenAI and Anthropic are both targeting offerings this year. Collectively, these three listings could absorb $150+ billion in institutional capital and significantly reshape the AI investment landscape.

• The bundled xAI/X structure raises governance questions: Enterprise customers and investors evaluating SpaceX’s IPO need to understand they are buying exposure to a Musk conglomerate rather than a pure-play rocket company. So far, xAI and X have underperformed and have been lapped by the competition. Investors will be betting on Elon Musk’s ability to execute the comeback of all comebacks in AI.

• Watch the dual-class structure: SpaceX is considering a share structure that gives insiders outsized voting power. Enterprise procurement teams and investors dealing with SpaceX should monitor whether this creates accountability gaps as the company scales.

3️⃣ Anthropic’s No Good, Horrible, Very Bad Week: Leaks Expose 512,000 Lines of Claude Code Source Code, including the Upcoming Mythos Model, and ‘Kairos’ Agent

In the span of one week, Anthropic had two significant security incidents that exposed its internal architecture, unreleased products, and proprietary engineering techniques to the public. For a company that markets itself as the safety-first AI lab, the timing is damaging on multiple fronts.

The First Incident: Mythos Revealed

On March 26, Anthropic accidentally published a draft blog post describing Mythos - also known internally as “Capybara”, the company’s most powerful model to date. Anthropic confirmed the model’s existence after the post circulated, noting it is compute-intensive and being rolled out cautiously, beginning with enterprise security teams. Security researchers at Axios reported that Mythos’s capabilities may materially lower the barrier to sophisticated cyberattacks.

The Second Incident: 512,000 Lines of Source Code

Five days later, developers noticed that Anthropic had exposed more than 512,000 lines of Claude code and approximately 1,900 related files in its NPM registry. The leak disclosed Anthropic’s proprietary “harness”, which is the techniques and tools it uses to coax its AI models into functioning as coding agents. Claude Code creator Boris Cherny acknowledged on X that “our deploy process has a few manual steps, and we didn’t do one of the steps correctly.”

The code spread quickly: over 8,000 GitHub copies were removed via copyright takedown before Anthropic narrowed its request to 96 repositories. The X post disclosing the leak surpassed 30 million views.

Among the details surfaced:

• An always-on agent called “Kairos” that fields tasks proactively

• A Tamagotchi-style pet named “Buddy”

• A “dreaming” process where the model periodically consolidates its task memories

• Instructions for Claude Code to operate “undercover” in certain environments without disclosing that it is an AI

“The leak is a blow for Anthropic because it risks both undermining its reputation for safety and also revealing valuable trade secrets in the pitched battle for enterprise customers.”

- WSJ, April 1, 2026

Melissa Bischoping, Senior Director of Security at Tanium, said the leak provided “a blueprint for what the code does under the hood,” giving adversaries a roadmap to find vulnerabilities or circumvent protections.

The leaked code does not include model weights, which are considered the true crown jewels, but does expose commercially sensitive engineering decisions that competitors can now replicate without reverse engineering.

The Takeaway

• Evaluate your Claude Code deployment security posture now: The leaked harness gives adversaries a detailed map of how Claude Code orchestrates AI models. Enterprise security teams using Claude Code should review prompt-injection risks and agent boundary controls, given the newly public architecture.

• Two incidents in six days are a pattern, not a coincidence: Both leaks stemmed from manual process failures, not external attacks. Enterprises considering Anthropic as a strategic AI partner should ask pointed questions about internal security governance and deployment automation before signing long-term contracts.

• Mythos is coming, and it requires a risk assessment: Anthropic’s own security researchers flagged Mythos’s potential to lower the barrier to sophisticated cyberattacks. Enterprise security leaders should request briefings on Mythos’s deployment controls before it reaches general availability.

4️⃣ Anthropic Considers October IPO at Potential $2 Trillion Valuation - Racing OpenAI to Market

Anthropic executives have held active discussions with bankers about executing an IPO as soon as Q4 2026. Anthropic’s funding target could exceed $60 billion, making it the second-largest offering in history, behind only SpaceX. The company’s annualized sales more than doubled to $19 billion in the first two months of 2026, driven primarily by Claude Code, which hit a $2.5 billion run-rate revenue in February, one year after its release.

Demand for Anthropic’s potential IPO is expected to be strong:

• Investment firm Coatue has projected Anthropic could reach a $1.995 trillion valuation by 2030.

• The secondary market is validating that trajectory: secondary platforms Augment and Hiive are registering demand valuing Anthropic at roughly $600 billion, more than 50% above its last primary-round valuation.

• Bankers and lawyers working on the IPO expect Anthropic to list before OpenAI, citing investor preference for Anthropic’s enterprise focus and its shorter projected path to profitability.

• Anthropic has projected it will burn approximately $22 billion before reaching positive free cash flow, compared to OpenAI’s prior projection of more than $200 billion.

• The company is working with law firm Wilson Sonsini in preparation for the IPO.

The Takeaway

• Anthropic is positioned as the enterprise leader: Its enterprise-first revenue mix, a tighter path to profitability, and a secondary-market premium signal that institutional investors view it as a better near-term investment than OpenAI.

• Three mega-IPOs in one year will reshape the capital markets: SpaceX, OpenAI, and Anthropic may collectively raise $200+ billion in 2026. Enterprise IT procurement leaders should monitor how this reshapes vendor pricing power and platform stability among their AI providers.

5️⃣ Microsoft Debuts Copilot ‘Critique’ Where Anthropic and OpenAI Models Work Together

Microsoft’s latest Microsoft 365 Copilot update makes the clearest statement yet about where enterprise AI is heading:

Model orchestration, not model selection, is the new battleground.

The headline feature is “Critique,” which routes research tasks through OpenAI’s GPT models to compile findings, then directs Anthropic’s Claude to fact-check and improve the output. Microsoft says the dual-model pipeline produces results that are materially more accurate than either model alone. Separately, Copilot Cowork, which is designed for long-running, multi-step autonomous work within M365, became available via Microsoft’s Frontier early-access program this week.

Microsoft has implemented a model-agnostic strategy while locking enterprise customers into the M365 and Azure ecosystems. By integrating both OpenAI (a $13B+ investment partner) and Anthropic (its second major AI relationship), Microsoft insulates itself from single-model risk and creates a switching cost:

Enterprise buyers who adopt Critique are buying into a workflow, not a model.

This mirrors Microsoft’s historical playbook with Office: own the productivity layer and let the underlying technology evolve beneath it.

The Takeaway

• Multi-model orchestration is now a Microsoft enterprise product, not an experiment: Enterprise IT teams evaluating Copilot should test Critique’s accuracy improvements against their specific use cases. If the dual-model pipeline outperforms single-model solutions in your critical workflows, the switching-cost case for staying in M365 becomes very strong.

• AI solutions providers need a model-agnostic story: Microsoft’s willingness to combine OpenAI and Anthropic in a single product sets a new expectation. Customers will increasingly ask AI solutions companies why they are locked into a single model provider and what their orchestration strategy is.

6️⃣ Oracle Cuts Up to 30,000 Jobs to Help Fund $80 Billion AI Data Center Push

Oracle began terminating employees across its US and India operations on March 31, notifying workers via early-morning emails from “Oracle Leadership” with no advance notice from managers or HR. The cuts span legacy database support, traditional ERP services, and back-office functions. Analysts at TD Cowen estimate the total reduction could reach 30,000 positions, which equals roughly 18% of Oracle’s 162,000-person global workforce. Oracle’s stock rose 6% on the day of the announcement.

The Data Center Strategic Shift + A Huge Capex Commitments Drive the Cuts

The layoffs are the direct consequence of one commitment:

Oracle’s participation in Stargate, the $500 billion AI infrastructure joint venture with OpenAI and SoftBank, was announced at the White House in January 2025.

Oracle’s Stargate contract includes a $156 billion commitment to build much of the data center capacity required to run OpenAI’s models, including the purchase of approximately 400,000 NVIDIA GPUs.

Oracle’s CapEx spending will grow rapidly:

• Oracle’s FY2025 capital expenditure reached $6.9 billion. Apparently, that’s just a down payment.

• Its commitments under the Stargate contract imply capex of $30 billion or more annually for multiple years.

• The flagship Stargate site in Abilene, Texas, is partially operational, but Oracle won’t see meaningful revenue from the overall project for several years.

• The company is partially funding this commitment by eliminating headcount across its applications, database businesses, and back-office functions.

• Oracle’s OCI cloud revenue is projected to grow from $18 billion in FY2026 to $144 billion by 2030, according to CEO Safra Catz’s guidance. Those numbers assume AI infrastructure demand will continue to accelerate. Most of the projected revenue growth is tied to OpenAI's success.

Oracle’s stock has fallen nearly 50% over the past six months amid concerns about the feasibility of its financing plan. Oracle recently allocated an additional $500 million to cover restructuring costs in the current fiscal year.

The Takeaway

• Oracle’s bet is the AI infrastructure boom’s biggest stress test: If AI demand scales as projected, Oracle’s capex-heavy strategy could generate enormous returns. If enterprise AI adoption plateaus or Stargate’s revenue timeline slips (likely), Oracle has limited operational flexibility to course-correct.

• Oracle’s enterprise software customers face service risk: The cuts span database support and ERP services. Enterprise buyers with Oracle in their critical infrastructure stack should assess whether support capacity for their tier-one systems has been impacted and plan accordingly.

• Watch Oracle’s OCI revenue growth rate in Q1 results: Guidance implies 77% OCI growth in FY2026.

7️⃣ The Chinese AI Market Accelerates as Domestic Chip Companies Gain Traction, Models Improve, and OpenClaw Enables Global Expansion

Three developments this week have materially strengthened China’s position in the global AI race:

• Huawei’s AI chips made a big leap forward.

• Alibaba shifted its AI model strategy to focus on delivering closed models.

• Overnight, OpenClaw became a powerful distribution engine for Chinese AI model makers.

Alibaba Closes the Open-Source Door

Alibaba released its new Qwen3.5-Omni model as its first closed model, marking a major strategic inflection point. Chinese AI companies spent two years building global developer mindshare with open-weight models; now they are pivoting to monetization. A follow-on Qwen3.6-Plus model touting “significant advancement in agentic coding” was released this week. It’s also a close model.

OpenClaw as a Distribution Engine

OpenClaw’s role as a global distribution channel for Chinese AI tokens:

• OpenClaw agents require an LLM provider at the back end; while Anthropic’s Claude is widely preferred for quality, its cost is high for users running intensive agentic tasks.

• Affordable Chinese alternatives from Zhipu, Minimax, Moonshot’s Kimi, and ByteDance cost roughly one-third as much as Claude and are capturing a significant share of OpenClaw token usage globally.

• ByteDance’s ClawHub partnership now gives Chinese AI companies a dedicated marketplace to reach global developers.

OpenClaw creator Peter Steinberger, who has since joined OpenAI, observed:

“In China, there are many companies where you get fired when you do not use OpenClaw.”

China’s AI Chip Market Reaches a Tipping Point

Chinese AI chip performance is improving rapidly:

• Huawei’s new 950PR AI chip is winning orders from ByteDance and Alibaba after testing showed significant improvements over its predecessor, including better compatibility with NVIDIA’s CUDA software.

• Huawei plans to ship approximately 750,000 of the 950PR chips this year, with mass production starting next month.

In fact, Chinese AI chipmakers are gaining market share. According to IDC data reviewed by Reuters, Chinese GPU and AI chip makers captured nearly 41% of China’s AI accelerator server market in 2025, shipping approximately 1.65 million cards. Huawei alone shipped around 812,000 chips, which is roughly half of all Chinese vendor shipments. The demand driver is clear:

China’s shift from model training to real-world AI inference deployment is turbocharged by the adoption of OpenClaw, which demands fast, cost-efficient inference at scale.

The Takeaway

• Chinese AI providers are winning on price via OpenClaw: Enterprise procurement teams evaluating agentic AI infrastructure need to explicitly evaluate whether their workflows will default to Chinese model providers on cost grounds and whether that creates data security or regulatory exposure.

• Huawei’s CUDA compatibility is a significant technical milestone: The 950PR’s improved compatibility with NVIDIA’s software ecosystem removes the main adoption barrier for Chinese chips.

• Alibaba’s open-source pivot signals a shift in market maturity: When the world’s largest open-source AI contributors start charging for their best models, it shows that the era of free Chinese AI models is ending.

8️⃣ US AI Adoption Is Rising, but Public Trust Is Falling, and Americans Are Conflicted on Regulation

In the US, consumer AI adoption and AI anxiety are both accelerating at the same time:

• A Quinnipiac University poll released March 30 found that 55% of Americans say AI will do more harm than good in their daily lives. That’s an 11-percentage-point increase since April 2025.

• 70% believe AI will reduce job opportunities, up 14 points year-over-year.

• Nearly two-thirds think AI will worsen education.

The polling was conducted in mid-March among 1,397 US adults, with a margin of error of ±3.3 points.

A separate Axios survey found that nearly two-thirds of Americans use AI regularly and want stronger government oversight but resist the trade-offs that tighter regulation requires, including slower innovation and higher costs. Adding a workplace dimension, Lean In data shows that managers are not giving women equal credit for AI-driven productivity gains, creating an equity gap in how AI’s benefits are recognized across workforces.

The Takeaway

• The trust gap is a deployment problem, not just a PR problem: Rising public skepticism will translate into employee resistance, union demands, and customer backlash if AI deployment is not paired with credible communication about impact on jobs, data, and decision-making. Build your internal change management strategy before the backlash arrives.

• Track the gender credit gap in your AI rollout: If women on your teams are doing AI-assisted work but not receiving credit, your productivity metrics and retention risk are both affected. Build AI attribution into performance review processes now.

9️⃣ Perplexity Sued Again - This Time for Allegedly Sharing User Data With Meta and Google

A federal class-action lawsuit filed April 1 in the Northern District of California accuses Perplexity AI of secretly sharing user conversation data with Meta and Google via tracking software embedded in its home page, including when users are logged into Perplexity’s “Incognito” mode. The lawsuit alleges that trackers are downloaded onto user devices upon login, giving Meta and Google “full access” to conversations and enabling them to exploit that data for ad targeting and to resell it to additional third parties.

A Pattern of Legal Exposure

This is not Perplexity’s first legal confrontation, and it is its most consequential:

• The company has faced copyright claims from News Corp (Dow Jones and NY Post), The New York Times, the Chicago Tribune, Nikkei, Asahi Shimbun, and Reddit, all centered on allegations of scraping copyrighted content without authorization and generating outputs that substitute for the original sources.

• A 2024 Wired investigation found Perplexity was using undisclosed IP addresses to access content from sites that had opted out of being scraped.

• The new user-privacy lawsuit is structurally different: it targets Perplexity’s relationship with its own users, not with publishers.

The Threat to Perplexity’s Agentic Pivot

The lawsuit arrives at a critical moment for Perplexity, which is attempting to pivot from AI search toward agent-based products, including Computer (a $200/month orchestration tool), Personal Computer, and the Comet browser — all of which require users to grant deep access to personal data, files, and browsing behavior. A finding that Perplexity shared even basic conversational data with third parties without disclosure would severely undermine the trust premise that these agentic products demand. Enterprises evaluating Perplexity for AI search or agent deployment should conduct careful data governance due diligence before proceeding.

“Publishers have been suing new tech companies for a hundred years, starting with radio, TV, the internet, social media, and now AI. Fortunately, it’s never worked.”

Jesse Dwyer, Head of Communications, Perplexity AI, TechCrunch, December 2025

The Takeaway

• Conduct data governance due diligence before deploying Perplexity in enterprise environments: If the lawsuit’s allegations about tracker behavior, including tracking in Incognito mode, are accurate, Perplexity could pose a direct threat to your enterprise data security requirements. Verify Perplexity’s data practices with your legal and security teams before expanding usage.

• The agentic product trust gap is the bigger strategic risk: Perplexity’s agentic pivot requires users to hand over far more sensitive data than a search query. It’s hard to trust a company that has repeatedly been accused of unauthorized data handling as it tries to enter the high-trust agentic market.

🔟 Salesforce Adds 30 AI Features to Slack as Enterprise Platforms Race to Embed Agents

Salesforce unveiled more than 30 new AI-powered features for Slack this week, covering meeting transcription, CRM updates, desktop assistance, and enterprise agent workflows. It’s the most ambitious Slack update since Salesforce acquired Slack for $27.7 billion in 2021. The update turns Slack from a messaging layer into an AI-powered work orchestration surface that can push updates to Salesforce CRM, automatically summarize meetings, and execute multi-step tasks without leaving the interface.

The release fits a broader pattern that every major enterprise platform is now racing to execute:

Embed AI agents into the workflow layer before a native AI alternative occupies the position.

Microsoft is doing it with Copilot in M365. ServiceNow is doing it with its AI platform. SAP is doing it with Joule. The window for incumbent platforms to establish dominance in AI workflows is finite, and Salesforce is moving rapidly.

The new Slack features reinforce Salesforce’s Agentforce strategy, positioning Slack as the front end for enterprise AI agents rather than a standalone communication tool. Customers who standardize on Slack for agentic workflows face a significantly higher switching cost than those who use it only for messaging.

The Takeaway

• Evaluate Slack’s new agent features against your existing workflow stack: If your enterprise already uses Slack and Salesforce CRM, the new Slackbot features offer a low-friction path to agentic automation. Test the CRM update and meeting transcription features against your current tooling before committing to alternative agentic platforms.

• The agentic workflow layer creates platform lock-in: The decisions you make today regarding your agentic workflows will be difficult and expensive to reverse in 2 to 3 years.

🃏 Definitely Not AI

The US National Arboretum’s prize pair of Bald Eagles, Mr. President and Lotus, had a very public spat. Apparently, they made up, though, because they are expecting. What if “With or Without You” were an Irish folk song? Hint: It’s beautiful. And so are “The Sound of Silence” and “Don’t Stop Believing.”

Artist Grant Snider creates a beautiful comic about the moon.

The NASA SLS rocket for the Artemis II mission is awe-inspiring.

The AI to ROI Podcast

If you want to listen to the latest in AI in more detail, Peter and Ray break down our big AI story of the week on the AI to ROI podcast. We also interview leaders at AI companies and mid- to large-sized enterprises about their challenges and successes in AI.

Click here to listen or subscribe using your favorite podcasting app.