AI to ROI News & Analysis: May 22, 2026

The SpaceX S-1 is a doozy; the IPO is scheduled for June 12th. Google shines at I/O, NVIDIA delivers the numbers again. OpenAI plans its IPO, Anthropic hits profitability with hypergrowth, and more...

The Biggest AI News This Week

🚀 SpaceX's S-1 Arrives with More Than a Few Reasons for Investor Caution

🤖 Google’s New Products Demonstrate a Compelling Agentic Path

💎 NVIDIA Knocks the Earnings Ball Out of the Park - AGAIN

📋 OpenAI Targets a $1 Trillion Valuation and a September IPO. The S-1 Will Tell the Real Story

📈 Anthropic Expects to Post $10.9B in Q2-26 Revenues AND Its First Quarterly Operating Profit

🖥️ Google & Blackstone Launch a Venture to Build & Operate TPU-based Data Centers

🏢 Anthropic Launches New Agentic Products for Small Businesses and Financial Professionals

🏛️ AI Executive Order Gets Nixed as AI Big Wigs and Trump Whisperers Intervene

📜 Report of the Week: Benedict Evans Is Back With His Twice-Yearly “AI Eats the World” Presentation

📊 Data You Can Use: How AI Has Invaded Our Lives

🃏 Definitely Not AI: China and “The Sound of Music.” Eiffel Tower staircase. Fox gets busted. Want to tempt fate? The metric system and the ocean.

1️⃣SpaceX's S-1 Arrives with More Than a Few Reasons for Investor Caution

SpaceX filed its long-awaited “public” S-1, targeting the largest IPO in history, a $75 billion raise at a $1.75–$2 trillion valuation, with a June 12 Nasdaq debut under the ticker SPCX. The filing unveiled the company’s complete financial profile for the first time:

• $18.7 billion in 2025 revenue (up 33%), offset by a $4.9 billion net loss.

• In Q1 2026, SpaceX lost $4.3 billion while burning $9 billion in cash, leaving $16 billion on hand against $29 billion in debt, including a $20 billion bridge loan from the xAI acquisition.

Deeper Dive

The crown jewel is Starlink, which generated $4.4 billion in operating income in 2025 (doubling year-over-year) and reached 10.3 million subscribers at $2.1 billion in Q1 revenue, up 44%. xAI (via Grok) contributed $475 million in Q1. The most significant new revenue line:

A $1.25 billion per month compute deal with Anthropic that runs through May 2029 (~$40 billion total) with a 90-day cancellation clause for either party.

The S-1 included a lot of red flags:

• Capital expenditures on AI tripled to $7.7 billion in Q1 alone.

• Starship, the S-1’s centerpiece technology, which has cost $15 billion to develop, is not yet commercially operational and represents the company’s top risk factor.

• The Terafab chip venture with Tesla and Intel is described as a “general framework” with neither party obligated to proceed.

• Orbital data centers are framed with the frank disclaimer: “No one has previously operated orbital AI compute.”

• Elon Musk’s 85%+ voting control via super-voting shares means that there are no checks on his power. Note that Musk earns a separate 1-billion-share bonus if he achieves a $7.5 trillion market cap and Mars colonization.

Analyst reaction was spirited:

Piper Sandler’s Lauren Webster called the filing “aspirational” and “visionary,” adding the familiar Wall Street caveat: “I don’t know if I have fully truly believed any TAM put in a prospectus.”

Bloomberg Intelligence’s Mandeep Singh was more direct: “When I compare it to Alphabet, I find only the Starlink part to be sustainable in terms of revenue and margins. All in all, I think the free-cash-flow potential is far lower than any of the other hyperscalers.”

Takeaways for AI Solution Vendors and Enterprise Leaders

• SpaceX’s xAI segment grew 23% in 2025, compared to Anthropic’s 1,000%+ and OpenAI’s ~300%. Even with the new Anthropic compute contract, the xAI cloud business remains an afterthought and subscale.

• Orbital AI compute remains entirely unproven. Enterprise leaders evaluating frontier infrastructure vendors should distinguish between operational technology and aspirational roadmaps.

2️⃣Google’s New Products Demonstrate a Compelling Agentic Path

Google used its annual I/O conference to declare what CEO Sundar Pichai called “the agentic Gemini era”. Google’s collection of new and upgraded products threads agentic AI through Search, Gmail, YouTube, shopping, and coding in ways that actually reinforce one another.

The centerpiece model, Gemini 3.5 Flash, went straight to general availability, skipping the preview stage, and is now deployed across Google’s consumer products. It costs 5x more than Gemini 3 Flash, but Pichai positioned it as cheaper than rivals:

“If the top companies in Google Cloud shifted 80% of their workloads from other frontier models to 3.5 Flash, they would save over $1 billion annually.”

Gemini 3.5 Pro will be released in June.

Deeper Dive

Four announcements deserve particular attention from enterprise leaders:

• Gemini Spark: a 24/7 personal agentic assistant that integrates with Gmail and calendar, running tasks in the background even when a laptop is closed - it is Google’s answer to OpenAI’s operator-style agents.

• Antigravity 2.0: an updated coding agent with a new desktop app that absorbed Gemini CLI, positioned as a cost-effective alternative to Claude Code.

• Universal Cart: a cross-retailer AI shopping agent integrated across Search, Gmail, YouTube, and Gemini.

• AI Search: Search VP Liz Reid called it “the biggest upgrade to the search box in 25 years”, with continuous-monitoring agents for apartments, stock prices, and product drops replacing the static results page. The line between Gemini and Google Search is effectively gone.

The Information reported that an Uber ML engineer found that Gemini still lags behind Claude in context and reasoning; however, for developers primarily concerned with cost and native Google integration, Antigravity 2.0 is now a credible alternative.

Google also introduced AI Ultra at $100/month (5x the usage of AI Pro) and smart glasses in partnership with Warby Parker and Gentle Monster, launching fall 2026 to compete with Meta’s Ray-Bans.

The Takeway

• Google is building a big and connected AI surface area. The convergence of Search, Gemini, and Gmail into a unified agentic layer creates a distribution advantage no AI startup can match.

• Pichai’s savings claim directly targets Anthropic’s Claude, which is both expensive and capacity-constrained. Expect Google’s enterprise team to target Claude’s perceived high price.

• Gemini Spark is the first serious challenge to standalone productivity AI tools. If it works as demonstrated, it can simultaneously compete with Copilot, Claude Cowork, and ChatGPT’s operator mode.

3️⃣NVIDIA Knocks the Earnings Ball Out of the Park - AGAIN

NVIDIA reported Q1 fiscal 2027 results on May 20 that, for the 14th consecutive quarter, beat Wall Street’s estimates. Revenue hit $81.6 billion (up 85% year-over-year), GAAP net income reached $58.3 billion (up 211%), and free cash flow reached $48.6 billion. That’s roughly half of the company’s entire fiscal 2026 free cash flow generated in a single quarter. Management projected ~$91 billion in revenue for fiscal 2Q27, representing ~95% year-over-year growth. It would be the fourth consecutive year of growth acceleration.

“Demand has gone parabolic. Agentic AI has arrived. The build-out of AI factories is the largest infrastructure expansion in human history and is accelerating at extraordinary speed. We built ahead of this moment so that when agentic AI arrived, Nvidia would be ready. It has arrived.”

Huang called out Anthropic specifically as “an important, relatively new customer at large scale.”

- CEO Jensen Huang

Deeper Dive

The drivers behind the numbers are compelling.

• Hyperscaler capex is projected at $725 billion combined in 2026, scaling to $3–$4 trillion by 2030.

• Networking revenue of $14.8 billion crushed the $12.7 billion estimate as NVLink demand for GB200 systems surged.

• NVIDIA also disclosed it expects ~$20 billion in standalone CPU revenue this year, positioning it as the world’s largest CPU supplier alongside its GPU dominance.

• The company raised its dividend from $0.01 to $0.25 per share and authorized $80 billion in new buybacks.

Despite the huge earnings beat, NVIDIA stock declined by 2%, reflecting investor fatigue more than fundamental concern:

UBS analyst Tim Arcuri described a “marked apathy” among investors.

Bloomberg’s headline: “Nvidia Gives Disappointing Forecast as Chip Competition Mounts“ referenced that Q2 guidance of $91 billion missed the most bullish analyst forecasts of $96 billion, not the consensus.

NVIDIA still has no sales in China, even though the US government has approved sales of previous-generation H200 chips to 10 Chinese firms.

The Takeaway

• NVIDIA still has plenty of room to grow. Its 95% growth projections for 2Q27 show a company clicking on all cylinders. The Category 2 customer cohort – AI clouds, industrial, and enterprise grew 74% YoY. The enterprise AI wave is just getting started.

• The Networking Segment at $14.8B (+263% YoY) signals that AI infrastructure is no longer just about compute. Interconnect and fabric spending are scaling in parallel, which is relevant for enterprise buyers designing production AI stacks.

4️⃣OpenAI Targets a $1 Trillion Valuation and a September IPO. The S-1 Will Tell the Real Story

OpenAI is preparing to file confidentially for an IPO as soon as the week of May 25th, targeting a September 2026 public debut at a valuation that could exceed $1 trillion. That’s up from its current private valuation of $852 billion. Goldman Sachs and Morgan Stanley are leading the process; Latham & Watkins is advising the banks; Cooley and Wachtell Lipton are advising OpenAI. CEO Sam Altman has been pushing to go public ahead of rival Anthropic, which is targeting a fall debut of its own.

Deeper Dive

The path is clear for OpenAI. On May 19, a jury sided with OpenAI in its legal case against Elon Musk, dismissing his claims that the company abandoned its nonprofit mission on technical grounds. Musk has said he will appeal. The verdict cleared the most significant legal obstacle to a public offering.

The IPO comes against a challenging backdrop:

OpenAI missed multiple internal revenue and user growth targets.

CFO Sarah Friar has told company leaders the offering may need more time to get its internal systems in order.

An $18 billion financing snag has stalled the partnership with Broadcom to build next-generation custom chips.

OpenAI has $600 billion in infrastructure spending commitments. The current growth rate can’t cover the cost.

The competitive stakes are enormous:

• Anthropic overtook OpenAI in enterprise adoption for the first time in April (34.4% vs. 32.3% of businesses per Ramp data).

• ChatGPT remains ahead in weekly consumer users (~900 million), but the gap is narrowing.

• Increasingly capable and cheaper Chinese AI models could eat into OpenAI’s growth story.

The Takeaway

• OpenAI has been on a product-release heater over the past 60-90 days, with ChatGPT 5.5, Image 2.0, and Codex getting great reviews and increasing adoption; however, the business side of the company feels a bit shaky. Stories about slowing growth, hard-to-believe forecasts, and executive turmoil are in the news every day.

• OpenAI would be better off listening to its CFO and getting its business and financial act in order. The AI race won’t be won this fall.

5️⃣Anthropic Expects to Post $10.9B in Q2-26 Revenues AND Its First Quarterly Operating Profit

Anthropic is on track to post $10.9 billion in revenue for the June quarter, up 130% from approximately $4.7 billion in the first quarter. More importantly, it is projecting its first-ever operating profit of $559 million. That would make Anthropic profitable on an operating basis roughly two years ahead of its internal forecasts.

The company’s annualized revenue run rate has surpassed $30 billion, up from $9 billion at the end of 2025, and it doubled the number of customers spending $1 million or more annually from 500 to over 1,000 in just two months.

Deeper Drive

The profitability may be temporary, because Anthropic is about to spend huge amounts of money on compute:

• Anthropic has committed to spending approximately $1.25 billion per month on SpaceX compute through May 2029, and roughly $40 billion over the contract term)

• It has also signed a five-year, $200 billion deal with Google for cloud infrastructure beginning in 2027.

Those commitments will likely drive operating losses in subsequent quarters as these capital expenditures ramp. The company is also in talks and is eyeing an October 2026 IPO, while watching SpaceX’s June debut as a potential market bellwether.

On the plus side:

• Anthropic’s revenue grew over 1,000% in 2025, while OpenAI grew roughly 300%.

• Anthropic now leads OpenAI in enterprise adoption, per Ramp’s AI Index (34.4% vs. 32.3%).

• Financial institutions account for approximately 40% of Anthropic’s top 50 customers.

Anthropic CEO Dario Amodei has said the company planned to grow 10x this year but is now on track for 80x growth.

The Takeaway

• Anthropic has “hit” products and an almost vertical growth path. The company is now projected to generate $50 billion in revenues this year. It is addressing its shortage of compute. They are the company to beat in AI at the foundation model layer.

• The biggest risks are: 1) increased traction for foundation model competitors OpenAI, Google, Microsoft, xAI; 2) competition from Chinese model companies; and 3) high prices that cause enterprise customers to look for less expensive options that accomplish the same things.

6️⃣Google & Blackstone Launch a Venture to Build & Operate TPU-based Data Centers

Google and Blackstone are creating an AI neo-cloud company that will rent Google’s Tensor Processing Units to AI developers in a direct challenge to CoreWeave’s GPU-rental dominance. Blackstone is committing $5 billion to fund the company, and Google is providing TPU chip infrastructure. The vehicle positions Google to expand TPU sales beyond its own cloud to third parties. Blackstone is providing the financing and participating in the profits.

Deeper Dive

Google has spent years building TPU infrastructure but has struggled to penetrate the broader AI developer market dominated by Nvidia GPUs, NVIDIA-centric hyperscalers, and the CloudWeave-style GPU rental model. The Blackstone partnership creates a commercial vehicle to change that by using Blackstone’s financial resources and distribution relationships.

The Takeaway

• This is a low-risk, high-reward partnership for Google. If Blackstone’s relationships and capital bring customers, it’s a validation of Google’s TPU architecture and a big new revenue stream. If it doesn’t work out, Google can put 100% of those GPUs to use in its Google Cloud infrastructure, as part of its relationship with Anthropic, or internally, because its engineers are clamoring for more capacity to build products.

7️⃣Anthropic Launches New Agentic Products for Small Businesses and Financial Professionals

Anthropic has launched two significant vertical expansions in the past two weeks. One targets small businesses; the other targets Wall Street. Both are based on the same premise:

That Claude’s most durable advantage lies in the depth of workflow integration, not model benchmarks alone.

Claude for Small Business Launched May 13th

Anthropic launched Claude for Small Business on May 13th. It’s available as a toggle inside Claude Cowork. The package connects Claude to QuickBooks, PayPal, HubSpot, Canva, DocuSign, Google Workspace, and Microsoft 365, with 15 ready-to-run workflows covering payroll planning, invoice chasing, month-end reconciliation, sales campaigns, contract routing, and cash-flow forecasting. Users approve before anything is sent, posted, or paid.

“Small businesses make up nearly half the American economy, but they’ve never had the resources of bigger companies. AI is the first technology that can finally close that gap.”

- Daniela Amodei, President, Anthropic

Claude for Financial Services Launched May 5th

Earlier, on May 5, Anthropic launched 10 ready-to-run agent templates for financial services:

Pitchbook creation, KYC screening, earnings analysis, credit memos, underwriting, month-end close, statement audits, and insurance claims.

Each ships as a plugin in Claude Cowork and Claude Code, and as a cookbook for Claude Managed Agents. Claude now works natively in Microsoft Excel, PowerPoint, Word, and Outlook (coming soon) via Microsoft 365 add-ins.

A Moody’s MCP integration brings proprietary credit ratings and data on 600 million+ companies directly into Claude workflows. Claude Opus 4.7 now leads the Vals AI Finance Agent benchmark at 64.4%. Production clients include JPMorganChase, Goldman Sachs, Citi, AIG, and Visa. Financial institutions represent approximately 40% of Anthropic’s top 50 customers.

The Takeaway

• Anthropic’s vertical expansion strategy, which now spans coding, legal, financial services, and SMB, is reducing its dependence on any single enterprise segment while widening its moat against OpenAI’s broader enterprise push.

• The Microsoft 365 integration is particularly significant: it embeds Claude into the existing workflows of hundreds of millions of enterprise users without requiring behavior change.

• For vendors building on Anthropic’s platform, the finance agent templates and MCP connectors represent a template for how Anthropic wants to expand. Expect healthcare and legal verticals to get similar treatment in the coming months.

8️⃣AI Executive Order Gets Nixed as AI Big Wigs and Trump Whisperers Intervene

Thursday, May 21, 2026, was supposed to be the day when President Trump signed the big AI executive order. That is, until Big Tech leaders and departed Trump tech advisers intervened. It seems that the President really didn’t want to regulate AI in the first place.

How It Went Down

According to Axios, ahead of the proposed signing, Trump, AI adviser David Sacks, and some in the industry discussed the executive order, sources familiar with the matter said:

• The main reason the executive order signing was delayed was that “he just hates regulation,” one source familiar with the matter said of Trump, adding that Sacks also “hated it.”

• “The whole thing was unnecessary” and “just something doomers wanted,” the source added.

Meta CEO Mark Zuckerberg, xAI CEO Elon Musk, and Sacks all spoke with Trump between Wednesday night and Thursday morning, arguing against the order. President Trump commented later on Thursday:

“I didn’t like certain aspects of it. I postponed it. I think it gets in the way of - you know, we’re leading China, we’re leading everybody, and I didn’t want to do anything to get in the way of that lead.”

President Trump has been trying to thread the needle between allowing American AI companies to flourish without strict rules while weighing growing public anti-AI sentiment, including within his own party. For now, the no guardrails team has won out. One government official told Axios:

“It could be CEOs, or egos in general. Everyone hates each other in the political tech space.”

Interestingly, the largest AI model makers didn’t oppose the executive order. While lingering questions remain about which AI models would participate in the voluntary testing program, OpenAI, Anthropic, Microsoft, and Google have been broadly supportive of AI model testing and guardrails. Plus, leading frontier or cutting-edge models already undergo voluntary testing through the Commerce Department’s Center for AI Standards and Innovation.

The Takeaway

• This EO would have been a good move by the Trump administration. As AI gains power, it should be partly the government’s responsibility to ensure that powerful models aren’t broadly deployed without their risks being known.

• Unfortunately, AI regulation will remain essentially a free-for-all for the time being, with states like California and New York enacting well-defined rules while other states pass no rules at all.

9️⃣ Report of the Week: Benedict Evans Is Back With His Twice-Yearly “AI Eats the World” Presentation

Benedict Evans is a prominent independent technology analyst who spent years in equity research, venture capital, and strategic advisory in London and Silicon Valley. Known for his macro-level insights, he produces highly regarded semi-annual presentations that contextualize sweeping technological platform shifts.

Key Takeaways

• The Nature of Platform Shifts: Technology advances in 10- to 15-year cycles. Generative AI represents the latest shift, fundamentally altering innovation, capital deployment, and corporate structures both within and beyond the tech sector.

• The Capex Explosion: Tech giants operate on the philosophy that under-investing poses a far greater risk than over-investing. This has triggered an unprecedented infrastructure spending boom, with the “Big Four” planning $700 billion in capital expenditures in 2026. Consequently, US data center construction spending is actively overtaking traditional office construction.

• Commoditization of the Frontier: Despite massive capital deployment, early data indicate that large language models are behaving like commodities. Models across different labs exhibit similar benchmark scores and lack clear network effects. Long-term pricing power for raw models remains unlikely, suggesting value capture will inevitably move further up the stack into applications and specialized workflows.

• The “Capacity Gap” in Adoption: While OpenAI reports over 900 million weekly active users, only 5% are paying. Consumer and enterprise adoption remains “a mile wide and an inch deep,” characterized by experimentation rather than entrenched daily habits. Coding has emerged as the clear frontrunner for enterprise spending and immediate product-market fit.

• Task vs. Job Automation: Historically, automation either turns a specific job into a baseline utility (like the elevator button) or changes the job’s entire nature to unlock massive price elasticity (like the spreadsheet-based expansion of accounting). AI acts as an “infinite intern,” automating boring, logical tasks. This shifts the human role away from routine execution and entirely toward taste, judgment, and original ideas.

Benedict Evans Report

🔟Data You Can Use: How AI Has Invaded Our Lives

More of us think we’re good enough lawyers on our own. Last year, self-represented litigants accounted for about 17% of federal non-prisoner filings, up from the historical average of 1%, according to a study by Anand Shah of the Massachusetts Institute of Technology and Joshua Levy of the University of Southern California. They ascribe the increase to AI.

The total number of cases also increased modestly, and so did the average number of documents in each case, which the researchers said means more work for judges. The duration of cases has remained steady, suggesting judges have managed to handle the influx so far.

It’s way too easy to make AI-generated music. “Make a jazz song about watering my plants,” suggests the website of Suno, one of several companies that lets users generate music, complete with lyrics, simply by typing in some text. The results are not Beethoven quality.

There are 3 more very interesting charts from “These 5 charts show how ChatGPT is flooding our lives” in The Washington Post (Gift Link)

Definitely Not AI

Why China LOVES “The Sound of Music (Gift Link).” Own a piece of the original Eiffel Tower staircase. Fox gets busted for stealing hot dogs.

Want to tempt fate? Guess her age. Metric versions of nautical measurements.

The AI to ROI Podcast

If you want to listen to the latest in AI in more detail, Ray and Peter 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.