Inferential

Anthropic filed a confidential draft S-1 with the SEC on Monday June 1, the first formal step toward an initial public offering reportedly targeted for October 2026 at a valuation that begins where the $65 billion Series H closed four days earlier — a post-money mark of $965 billion. The filing itself stays confidential until fifteen days before any roadshow, but the company is now publicly attaching numbers to the trajectory: annualised revenue run rate crossed $47 billion earlier in May, total revenue is approximately $10.9 billion and more than doubling year over year, and management expects to post its first operating profit during the June quarter with a break-even target of 2028 — two years ahead of OpenAI’s 2030 self-disclosed timeline. The S-1 itself is procedural. The $47 billion is the data point. ...

Anthropic shipped Claude Opus 4.8 on Thursday May 28 — forty-one days after Opus 4.7 landed on April 16. That is the fastest Opus cycle the company has ever run. Opus 4.6 to 4.7 took roughly seven months. The previous accelerations in the line were measured in months, not weeks. The pricing did not change — $5 per million input tokens and $25 per million output tokens, the same as 4.7, with a Fast Mode now running at 2.5× speed for 3× lower cost than the prior generation. Anthropic also raised $65 billion at a $965 billion valuation on the same day. The combination — fastest-ever cycle, same pricing, larger fast-mode discount, fresh capital at unprecedented scale — is the closest thing to a 2025-style cadence the frontier has produced this calendar year, and it is exactly the test H2 was set up to detect. ...

At midnight UTC this morning, GitHub Copilot’s flat-fee era ended. The product that grew under “$10 a month for everything” — and that Microsoft disclosed last quarter is now contributing to a Copilot business clearing roughly $2 billion in annual recurring revenue — switched all paid plans onto usage-based billing. Base subscription prices did not change: Copilot Pro remains $10 a month, Pro+ $39, Business $19 per user per month, Enterprise $39. What changed is what those fees buy. Each plan now includes a monthly allotment of GitHub AI Credits — $10 buys 1,000 credits at $0.01 per credit, $39 buys 3,900, and so on — and once the credits run out, users either stop or pay token-metered overages. The fallback to a cheaper model that used to keep work flowing after the old premium-request budget exhausted no longer exists. The meter is running, and for the first time in the product’s history a heavy user will know exactly what their agentic workflow costs. ...

Seven Q1 FY27 prints landed in the eight calendar days from Workday on May 21 to MongoDB on May 28. SaaS, data, identity, hardware, and silicon — five layers of the enterprise stack, seven income statements, one week. The result is the cleanest cross-layer evidence the substitution thesis has produced so far, and it does not point where the original hypothesis framing said it would. The vendors building the substitution layer are mostly capturing it; the market is pricing them on whether their AI revenue line is large enough and credible enough to look like genuine new value rather than a re-labelled extension of the seat business; and the dispersion of stock reactions inside the cohort is doing more analytical work this week than the headline numbers. ...

Salesforce closed Wednesday’s print with Agentforce ARR at $1.2 billion, up 205% year over year and roughly 50% sequentially from the $800 million figure it reported at the end of January. Snowflake reported the same day what CEO Sridhar Ramaswamy called “a clear inflection point” — product revenue of $1.33 billion at 34% year-over-year growth, the strongest sequential dollar growth in the company’s history. Workday’s Q1 FY27 print six days earlier ran the same playbook from a different starting point: agentic AI annualised revenue near $500 million, customer agent count doubled quarter-over-quarter, GAAP operating margin expanded 1,154 basis points in one year. Three Q1 FY27 prints in a six-day window, three different cost structures, three different customer bases, one pattern. ...

Aneel Bhusri co-founded Workday in 2005, ran it for sixteen years, and stepped back from the CEO seat in 2022. He came back on the first day of fiscal 2027, signed the Q1 10-Q on May 22, and used the earnings call to tell investors what he believes the company actually has to do now: “The 150th feature in HR or finance is not going to move the needle for our business. The next agentic application will.” The stock rallied roughly 10% on the day. The print and the founder’s statement together set up the cleanest test of the central H1 mechanism we have seen this quarter — the canonical seat-based vendor, in an SEC filing, in two adjacent sentences, admits seat erosion at renewal and books the substitution revenue offsetting it. ...

Intuit told 3,000 employees on Wednesday that their jobs were gone — 17% of an 18,200-person global workforce — in an internal memo sent hours before the Q3 FY26 earnings release. The same release raised full-year revenue guidance to $21.341–21.374 billion, posted Q3 revenue of $8.6 billion (+10%), Online Ecosystem revenue of $2.5 billion (+19%, +22% ex-Mailchimp), and non-GAAP EPS of $12.80 (+10%). The stock entered the print down 41% year-to-date. The arithmetic is now familiar — beat the quarter, raise the guide, cut the workforce, watch the multiple compress — but the mechanism Intuit is funding with the cut is different from the ones SAP, Atlassian, and Cisco have used in the past six weeks. ...

Nine mathematicians signed a 19-page companion paper on Wednesday confirming that an internal OpenAI reasoning model — general-purpose, not math-specialised — produced a counterexample to Erdős’s 1946 planar unit-distance conjecture. The list of names is the story: Noga Alon (Princeton), W.T. Gowers (Fields medalist, Cambridge/Collège de France), Will Sawin (Princeton), Daniel Litt, Arul Shankar, Jacob Tsimerman, Victor Wang, Melanie Matchett Wood, and Thomas Bloom — the same Bloom who maintains erdosproblems.com and called OpenAI’s previous math announcement “a dramatic misrepresentation” because the model had retrieved existing solutions from the literature. Sawin’s separate paper the same day gives an explicit improvement exponent δ = 0.014, derived through class field towers and Golod-Shafarevich theory — techniques from algebraic number theory that had not previously been brought to bear on what was thought to be a purely geometric problem. Gowers’ written verdict: “a milestone in AI mathematics… I would have recommended acceptance to the Annals of Mathematics without any hesitation.” ...

Cisco raised its FY26 AI infrastructure order target from $5 billion to $9 billion on Wednesday — an 80% guide-raise nine months into the fiscal year — and lost 260 basis points of non-GAAP gross margin doing it. Hyperscaler AI orders printed $1.9 billion in Q3 alone against $600 million in the year-ago quarter. Networking revenue grew 25% to $8.82 billion. Total revenue hit a record $15.8 billion. The stock jumped roughly 15% after-hours. And the same press release announced 4,000 layoffs with up to $1 billion in restructuring charges, $450 million of which lands in Q4. ...

Anthropic committed $200 billion to Google Cloud last week — five gigawatts of TPU capacity over five years, beginning in 2027 — first reported by The Information on May 5. The contract represents roughly 40% of Alphabet’s $462 billion cloud backlog disclosed at Q1 earnings. Alphabet shares are up about 160% over the past twelve months and briefly passed Nvidia in after-hours market cap the day the report ran. Wall Street read the news as confirmation that “owning the stack” — chips, models, distribution — is the durable AI position. ...