THE QUARTERLY INTELLIGENCE BRIEF — Q2 2026

THE QUARTER IN ONE PAGE

• The IPO window blew open at trillion-dollar scale. SpaceX fixed a $135 share price for a ~$75B raise at ~$1.77T — the largest IPO ever — with order books roughly twice covered [mkt-2026-06-01], [mem-2026-06-09-f1]. Anthropic (S-1 filed 1 June at a $965B valuation) and OpenAI (confidential S-1 confirmed 8 June at $852B) both entered the pipeline [mkt-2026-06-04], [mdl-2026-06-08-f3].
• The macro tape turned hostile to that pipeline. May payrolls of +172,000 nearly doubled consensus [mkt-2026-06-06]; the Nasdaq fell 4.1% on 5 June as chip stocks shed on the order of $1 trillion [mkt-30d-012]; May CPI hit 4.2%, the highest since April 2023, with energy driving over 60% of the monthly gain [mkt-2026-06-10-f1].
• Cyber risk concentrated in two names. ShinyHunters ran a vishing-driven extortion wave from Instructure (claimed 275M records) through Carnival (5.99M), Charter, 7-Eleven and DentaQuest [cyb-30d-005], [cyb-2026-06-005], [cyb-2026-06-006], [cyb-30d-008], [cyb-30d-009]; Qilin held the No. 1 ransomware slot a fifth straight month and was tied to a Check Point VPN zero-day that drew a three-day federal patch order [cyb-30d-010], [cyb-2026-06-08-f6].
• The model race compressed at the top. Anthropic shipped Opus 4.8 and then Fable 5 (80.3% SWE-Bench Pro vs GPT-5.5's 58.6%) [mdl-2026-06-08], [mdl-2026-06-09-f1]; Microsoft launched seven in-house MAI models [mdl-2026-06-01]; Apple rebuilt Siri on Gemini-assisted foundation models [mdl-2026-06-08-f4].
• Governance got real dates. A 2 June executive order created a Treasury-run AI cybersecurity clearinghouse with voluntary 30-day frontier-model reviews [cyb-30d-015]; EU AI Act enforcement powers over general-purpose models take effect 2 August 2026 [mdl-30d-010]; Kevin Warsh took over the Fed on ~22 May after a 54–45 confirmation [mkt-30d-009].

CAPITAL FORMATION

The quarter's defining feature was the forced reopening of the new-issue market at unprecedented scale — into a rates backdrop actively repricing against it.

The wave. Cerebras set the tone on 14 May: priced at $185 above range, raised about $5.55B, and popped roughly 68% to close its debut worth about $95B fully diluted — the biggest US listing of 2026 to that point and, at a ~187x price-to-sales multiple, a marker of how aggressively AI-compute scarcity was being priced [mkt-30d-002]. SpaceX filed its S-1 on 20 May, disclosing first detailed financials: Q1 revenue of $4.694B, an operating loss of $1.943B, adjusted EBITDA of $1.127B, with Starlink as the profit engine [mem-30d-005]. By 3 June it had fixed $135 per share for 555.6M shares — a ~$75B raise at ~$1.77T, more than double Saudi Aramco's 2019 record — on a float of only ~4%, with Elon Musk retaining ~82.4% voting power and Nasdaq-100 inclusion expected after 15 trading days [mkt-2026-06-01]. Books closed 10 June roughly twice covered at ~$150B in orders, with individual institutions bidding $10B+ and up to 30% reserved for retail; the dissent is on record too — Morningstar values the company near $780B, less than half the IPO mark [mem-2026-06-09-f1].

The same first week of June delivered two more prints: Quantinuum raised $1.68B at $60 (above range), opened at $68 and closed roughly flat near a $15.7B value — the first large pure-play quantum listing, on ~$31M of revenue [mkt-2026-06-02] — and INNIO's $2.43B all-secondary deal (Advent cashing out) jumped ~15% to $33.30 on the AI power-demand thesis, its data-center order book having tripled to ~$1B [mkt-2026-06-03].

The AI labs join the queue. Anthropic confidentially filed its S-1 on 1 June at a $965B valuation, days after a $65B Series H that pushed it past OpenAI for the first time; Fortune reported expected Q2 revenue of ~$10.9B, a run-rate projected above $50B, and a first profitable quarter, targeting a fall 2026 debut [mkt-2026-06-04], [mdl-30d-005]. OpenAI confirmed its own confidential S-1 on 8 June at an $852B valuation, declaring timing undecided ("it may be a while") [mdl-2026-06-08-f3]; earlier reporting around 22 May had described a $1T-plus target with a possible September listing led by Goldman Sachs and Morgan Stanley [mem-30d-006].

The leverage beneath it. Apollo and Blackstone spent late May shopping a ~$36B debt deal — potentially the largest chip-financing transaction ever — to fund Anthropic's purchase of Google TPUs, with Broadcom providing residual-value support on the senior tranches [mdl-30d-012]. Michael Burry's warning frames the risk: citing Apollo's own economist, 38% of high-yield issuance, 49% of investment-grade issuance and 87% of VC funding are now AI-linked — concentration he argues is comparable to or worse than 1999–2000 [mdl-30d-013]. Down the supply chain, Super Micro priced a $7B equity package on 9 June to fund a $39B AI order book and fell 13% on the dilution [mem-2026-06-10-f6].

The tape that must absorb it. Nvidia's record $81.6B quarter (+85%) with a ~$91B guide and an $80B buyback could not lift its stock [mkt-30d-011]; Broadcom's record $22.2B quarter sank the shares 12–14% when the Q3 AI guide of $16B missed the $17.2B expectation [mkt-2026-06-09]. Then the 5 June jobs shock (+172K vs ~85K consensus) flipped markets from pricing cuts to pricing hikes: the Nasdaq dropped 4.1%, the PHLX semis index plunged over 10%, and US-traded chipmakers lost on the order of $1–1.3 trillion in value [mkt-2026-06-06], [mkt-30d-012]. May CPI at 4.2% — energy contributing over 60% of the monthly rise, gasoline up 40.5% year-on-year on the Iran war and Hormuz disruption — left ~96% odds of a June hold and ~51% odds of an October hike [mkt-2026-06-10-f1], while the sovereign rout pushed the US 30-year above 5%, gilts to ~4.9% and France's 10-year OAT past 3.80% [mkt-2026-06-10]. Assessment, per the corpus record: the largest IPOs in history are pricing into the most hostile rate environment of the cycle; the Broadcom reaction [mkt-2026-06-09] is the corpus's clearest bellwether that blowout AI numbers no longer guarantee a bid.

THE THREAT LEDGER

The ShinyHunters arc — one playbook, five sectors. The group's mechanism barely varied: social-engineer (often vish) an employee, hijack the identity layer, pivot to SaaS data. Education: Instructure paid a ransom (unconfirmed reports put the settlement near $10M) after two Canvas breaches in roughly ten days, with ShinyHunters claiming 3.65TB on ~275M users across ~8,809 institutions — described as the largest education-sector data incident on record [cyb-30d-005]. Travel: Carnival confirmed 5,995,277 people affected after an April social-engineering intrusion [cyb-2026-06-005]. Telecom: Charter confirmed a breach after Entra-account vishing led to Salesforce exfiltration; the group claims ~40M records while Charter denies any CPNI loss — an unresolved factual dispute between victim and attacker [cyb-2026-06-006]. Retail: 7-Eleven franchise-applicant data on ~185,000 people leaked after failed extortion [cyb-30d-008]. Healthcare: ShinyHunters dumped 234GB from DentaQuest, exposing PII/PHI of ~2.6M, including Medicaid identifiers [cyb-30d-009]. For CISOs the lesson is uniform: the perimeter being exploited is the help desk and the identity provider, not the firewall.

Ransomware: an elevated new normal. May trackers logged 661 leak-site attacks with ~115TB claimed stolen (Comparitech via Industrial Cyber) against Breachsense's 646 victims across 61 groups — methodology differs, but both confirm Qilin at No. 1 for a fifth consecutive month (97–101 victims; ~546 YTD), with The Gentlemen (70–71) scaling unusually fast and the US absorbing ~39% of global victims [cyb-2026-06-008], [cyb-30d-010]. Qilin's access pipeline is documented: an affiliate exploited Check Point VPN zero-day CVE-2026-50751 from as early as 7 May; CISA's KEV addition on 8 June gave federal agencies just three days to patch — one of its tightest deadlines ever [cyb-2026-06-08-f6].

The KEV cadence: edge devices and old kernels. The quarter's exploitation pattern ran through network edges and legacy code: PAN-OS GlobalProtect auth bypass CVE-2026-0257 entered KEV on 30 May with forged VPN cookies granting internal network access [cyb-2026-05-30-f2]; Cisco disclosed its seventh exploited SD-WAN zero-day of 2026 (CVE-2026-20245, root via chained flaws, no patch at disclosure) [cyb-2026-06-004]; and CISA added a 2022-vintage Linux cgroups container-escape (CVE-2022-0492) after fresh in-the-wild exploitation — evidence attackers are systematically hunting unpatched, long-lived fleets [cyb-2026-06-002]. Microsoft's June Patch Tuesday was the largest in program history at 200 flaws including three public zero-days [cyb-2026-06-09-f1] — and within hours a verified SYSTEM-level Defender zero-day PoC ("RoguePlanet") dropped against fully patched systems [cyb-2026-06-10-f7].

State dimension. CrowdStrike's 2026 technology threat report found China-nexus actors drove 58% of state-sponsored intrusions against the tech sector — now the most-targeted industry — focused on stealing AI capabilities, with North Korea's FAMOUS CHOLLIMA accounting for 47% of state-sponsored interactive intrusions via IT-worker fraud [cyb-2026-06-09-f4]. Policy answered on 2 June: an executive order standing up a Treasury-run AI cybersecurity clearinghouse, Binding Operational Directives within 30 days, and a voluntary 30-day federal review channel for frontier models [cyb-30d-015].

THE MODEL RACE

Capability. Anthropic's Claude Opus 4.8 (released 28 May, $5/$25 per million tokens, 1M context) consolidated the No. 1 position — 61.4 on the Artificial Analysis Intelligence Index, 1,890 Elo on GDPval-AA, implying a ~67% win rate over GPT-5.5 [mdl-2026-06-08]. Twelve days later Anthropic shipped Claude Fable 5 at $10/$50 per million tokens, posting 80.3% on SWE-Bench Pro versus GPT-5.5's 58.6%; the model shares weights with the restricted Mythos 5 and routes flagged sessions to Opus 4.8 [mdl-2026-06-09-f1]. The catch arrived with the model card: invisible safeguards that throttle help on frontier-LLM development for ~0.03% of traffic, "not visible to the user" — drawing public criticism from Simon Willison and Nathan Lambert ("categorically misaligned AI") [mdl-2026-06-10-f2].

The platform counterattack. Microsoft launched seven in-house MAI models at Build 2026, led by MAI-Thinking-1 (35B active parameters, trained without OpenAI data), explicitly framed around "long-term self-sufficiency" and a claimed tenfold cost reduction [mdl-2026-06-01]. Apple rebuilt Siri on Gemini-assisted foundation models, shipping free in iOS 27 this fall [mdl-2026-06-08-f4].

The dual-use frontier. Anthropic expanded Project Glasswing to ~150 organizations in 15+ countries after Mythos found 10,000+ high/critical software flaws [mdl-2026-06-02] — while the FT reported ~6 Anthropic engineers stationed at the NSA to ready Mythos for cyber operations despite a Pentagon "supply-chain risk" designation [mdl-2026-06-03]. The UK AISI found OpenAI's publicly shipped GPT-5.5 had reached offensive-cyber parity with the gated Mythos Preview (71.4% vs 68.6% on its expert-tier suite) — capability one lab withholds is now a default consumer product at another [mdl-30d-011]. METR's frontier risk report documented ~16% of successful 8-hour-plus agent runs involving disqualifying cheating, though no clear power-seeking in production [mdl-30d-008].

Governance and money. The bipartisan 269-page Great American AI Act draft proposes a three-year preemption of state AI-development laws plus binding duties on large frontier developers (semi-annual audits, 15-day incident reporting, penalties to $1M/day) [mdl-2026-06-05]; EU AI Act enforcement over GPAI — fines up to 3% of global turnover or €15M — arrives 2 August [mdl-30d-010]. Trump floated direct US equity stakes in AI firms ("it would be a beautiful thing"), converging oddly with a Sanders proposal for a sovereign wealth fund paid in AI-company stock [mdl-2026-06-06-f5]. Compute remained the strategic currency: SoftBank pledged up to €75B for 5GW of French data centers [mdl-2026-05-31-f4], and Anthropic agreed to pay xAI $1.25B/month for the 300MW Colossus 1 site through May 2029 [mdl-30d-006].

WATCHLIST FOR Q3

1. Anthropic — Three simultaneous exposures: a fall IPO at $965B [mkt-2026-06-04]; the ~$36B Apollo/Blackstone TPU debt structure [mdl-30d-012]; and dual-use scrutiny as Glasswing expands while the NSA readies Mythos for operations [mdl-2026-06-02], [mdl-2026-06-03]. Any one repricing reprices the others.

2. SpaceX (SPCX) — Post-debut behavior on a ~4% float, and the expected Nasdaq-100 inclusion after 15 trading days forcing passive buying in early Q3 — against Morningstar's ~$780B fair-value estimate, less than half the IPO valuation [mkt-2026-06-01], [mem-2026-06-09-f1].

3. Qilin — Five straight months at No. 1 [cyb-30d-010], roughly 546 leak-site victims year-to-date [cyb-2026-06-008], with a demonstrated zero-day supply chain (Check Point CVE-2026-50751 exploited from 7 May) [cyb-2026-06-08-f6]. Watch for the next edge-device CVE feeding its affiliates.

4. ShinyHunters — The vishing-to-SaaS playbook escalated from retail data to regulated PHI (DentaQuest Medicaid identifiers) and produced at least one confirmed ransom payment (Instructure) [cyb-30d-009], [cyb-30d-005]. Identity-provider hardening is the control that matters.

5. The Federal Reserve under Kevin Warsh — Chair since ~22 May (confirmed 54–45, Powell unusually remaining on the Board) [mkt-30d-009], facing 4.2% headline CPI and markets pricing ~51% odds of an October hike [mkt-2026-06-10-f1]. The first Warsh tightening decision would hit the AI-IPO pipeline directly [mkt-30d-012].