May 17, 2026
TL;DR — Major Blocks
• Anthropic is in talks to raise $30B+ at a $900B+ pre-money (closing end-month), after crossing $1T on Forge on a $44B run-rate (up from $9B at year-end 2025). The 300MW SpaceX compute deal on May 6 unlocks the next leg.
• Neuralink's clinical scope is now sight, speech, and hearing — Blindsight first-in-human gated for 2026 with the new 1,680-channel S2 stim chip (+67%), alongside live speech-decoding and a freshly announced auditory-cortex program.
• ByteDance begins testing paid Doubao subscriptions as MAUs reach 345M, and lifts 2026 AI capex past $30B (¥200B) — a 25% hike. Founder-shareholder demand remains north of $600B.
TL;DR — Watchlist.
• Anduril closed a $5B Series H at $61B on May 13 (Thrive + a16z), doubling its mark in twelve months; 2025 revenue doubled to $2.2B.
• Kalshi raised $1B Series F at $22B on May 7 (Coatue-led, Sequoia / a16z / Paradigm in); ARR ~$1.5B, institutional volume +800% in six months.
• Isomorphic Labs announced a $2.1B Series B in May and launched IsoDDE, which roughly doubles AlphaFold 3 accuracy on protein-ligand benchmarks.
• SpaceX's December tender priced shares at $800B (up from $400B in summer 2025); xAI was fully integrated into a new SpaceXAI division in May, ahead of a mid-to-late 2026 IPO targeted at $1–1.5T.
• OpenAI closed a $122B raise at $852B post-money, ~$24B annualized; Q4 2026 IPO targeted at a $1T mark — Anthropic now trades above OpenAI on Forge.
• Stripe's February tender priced at $159B (+74% YoY); 2025 payment volume $1.9T (+34%). Patrick Collison reiterates an IPO would be "a solution in search of a problem."
━━━━━━━━━━━━━━━━
Anthropic — $900B Primary Round, Surpasses $1T Valuation on Secondary with $44B ARR, Compute Locked with SpaceX
What happened
The single most consequential week of Anthropic's history landed mid-month. On May 6, Anthropic and SpaceX announced a multi-year compute partnership that secures 300 megawatts of incremental AI capacity for Claude training and inference, with a separately scoped exploration of multi-gigawatt orbital compute in collaboration with SpaceX. The Musk–Anthropic thaw — Musk had publicly mocked the company's $380B post-money in February as "misanthropic" — is itself the story here: it confirms that the AI compute market is now too tight for any single hyperscaler bias to sustain. Anthropic's Google Cloud / TPU expansion (announced in April) and its AWS-distribution agreement now compose a true three-hyperscaler footprint.
On May 12, Bloomberg reported that Anthropic is in early talks to raise at least $30B in fresh financing at a pre-money valuation above $900B, with the round expected to close before month-end. Sources cite a target funding envelope that may ultimately exceed $60B as Anthropic builds out the SpaceX deal, Project Glasswing, and its life-sciences and frontier-cyber pillars. If priced, the round would put Anthropic's post-money inside the same band as OpenAI's $852B and would surpass the Forge-implied secondary mark from a quarter ago.
Revenue and secondary pricing
Anthropic's run-rate has continued its near-vertical climb. The company's own disclosures and corroborating press indicate annualized revenue crossed $44B in early May, up from $30B in March and roughly $9B at year-end 2025 — a 233% Q1 print, which Dario Amodei has described as "80x annualized" growth. Brad Gerstner's $100B run-rate-by-year-end thesis from April now looks materially more credible than it did six weeks ago.
The secondary tape has caught up. On Forge Global, Anthropic's implied valuation has been hovering at or just above $1T through May, with Forge CEO Kelly Rodriques publicly confirming the print. For the first time, Anthropic now trades through OpenAI on Forge ($1T vs $880B) — a structural inversion driven by Anthropic's growth-rate gap and OpenAI's heavier dilution arithmetic. Goldman Sachs and JPMorgan have been mandated to advise on a potential October 2026 IPO at a target of $400–500B, which would be the largest tech listing in history if priced at the midpoint.
For our desk, three things follow. First, the $900B+ primary mark is the new floor, not the new ceiling — Forge is already through $1T, and the depth at that level is thin but consistent. Second, our SPV pipeline is seeing indicative bids cluster at $900B–$1.2T. Third, given the impending primary, we are sizing this quarter's allocations to clear ahead of the close — post-close, the next institutional reference point is the IPO band, and we expect SPV bid-ask to widen markedly in the interim.
Sources — Anthropic
• Bloomberg — Anthropic, SpaceX Sign Computing Deal (May 6, 2026)
• Bloomberg — Anthropic In Talks to Raise $30B at $900B Valuation (May 12, 2026)
• Yahoo Finance / Quartz — Anthropic crosses $1T on Forge, surpasses OpenAI
• Axios — How Elon Musk grew to love Anthropic (May 7, 2026)
━━━━━━━━━━━━━━━━
Neuralink — Sight, Speech, and Hearing All in the Pipeline
What happened
Neuralink's clinical surface area expanded materially in the past six weeks. The flagship motor/speech N1 program now has three patients implanted publicly disclosed, with Brad Smith (ALS) demonstrating cursor + synthesized speech, and a more recent patient ("Kenneth") receiving an implant calibrated for imagined-speech decoding — an architectural step from cursor control toward continuous-speech BCI. Separately, Blindsight, Neuralink's vision-restoration implant for the fully blind, is now formally gated for first-in-human in 2026, with Joseph O'Doherty (head of next-gen) confirming at the Neural Interfaces Conference that the Blindsight stack uses a new 1,680-channel S2 stimulation chip — a 67% channel-count increase over the prior generation. Reports indicate the FIH trial may be sited in the UAE, alongside a US arm.
The most consequential addition came in early April but compounded in May: Musk publicly described a third sensory program targeting hearing restoration by directly stimulating the auditory cortex — bypassing the cochlea and auditory nerve entirely. While this program does not yet have a disclosed FIH date, it directly extends the "sensory restoration" platform thesis from a one-indication (motor) to a three-indication (motor/speech, vision, hearing) franchise on a shared neural-interface stack.
Operationally, Neuralink remains on its 2026 plan to move to near-fully automated surgery (dura-piercing threads) and high-volume implant production, with a publicly stated target of 1,000+ implants in 2026 — an order-of-magnitude step-up from current single-digit patient counts.
Secondary pricing
Neuralink's last priced primary remains the $650M Series E in June 2025 at a $9B pre-money — now nearly a year old and conspicuously stale relative to where comparable BCI / neuro-platform names are trading. The three-indication platform thesis (motor/speech + vision + hearing) materially expands the underwriteable TAM, and the surgical automation roadmap directly compresses the cost-of-care argument that has historically suppressed BCI multiples.
Our pipeline this month is seeing indicative bids that have stepped up meaningfully — clusters now sit in the $50B–$60B range versus $44B–$48B in mid-April. Inventory remains thin (Series A/B Founders Fund and Khosla blocks are the deepest known pools), and spreads remain wider than the AI-software comps, but the clinical derisking compounding across three programs has materially tightened the bid side. We expect the next primary mark to print well above the indicative range as Blindsight reaches FIH and the speech-decoding patient delivers its first month-out readout.
Sources — Neuralink
• MobiHealthNews — Elon Musk provides update on Blindsight implant trials
• Dataconomy — Neuralink wants to restore hearing by bypassing the ear (Apr 6, 2026)
• Neurapod — Neuralink Blindsight Trial: What to Expect (S2 1,680-channel chip)
• Teslarati — Neuralink Blindsight human trials expected to start in the UAE
━━━━━━━━━━━━━━━━
ByteDance — Doubao Monetization Begins, $30B+ AI CapEx, 345M MAUs
What happened
Two events define the May tape for ByteDance. First, on May 5–6, ByteDance began testing paid subscription tiers for Doubao — a three-tier structure at ¥68 / ¥200 / ¥500 per month (roughly $10 / $28 / $69) targeting heavier-compute use cases like slide generation, advanced analytics, and video production. The basic free tier persists; the introduction of paid tiers is the cleanest signal yet that ByteDance has crossed the line where unit economics, not user acquisition, define its AI roadmap. The pricing is competitive with ChatGPT Plus / Pro but materially below domestic peers — the take-rate question for analysts is whether Doubao's 345M MAU base converts at OpenAI-class rates (8–10%) or at lower China-specific rates (2–5%).
Second, on May 11, multiple outlets (Yahoo Finance / FT, BigGo, WinBuzzer) reported that ByteDance has lifted its 2026 AI capex envelope to roughly ¥200B (~$30B+) — a 25% upward revision versus the initial plan and a meaningful step-up versus the ~¥150B 2025 outturn. The capex revision reflects (a) higher memory-chip costs in the current cycle, (b) the Doubao 2.0 training and serving load, and (c) the Volcano Engine enterprise pipeline. For LPs underwriting ByteDance secondary, the implication is unambiguous: ByteDance now competes on compute commitments at the scale of the top three US hyperscalers and frontier-lab pairs.
Underneath these two events, the operating cadence remains exceptional. QuestMobile data put Doubao at 345M MAUs as of March 2026 — roughly 2x Qwen (166M) and ~2.7x DeepSeek (127M). Daily token consumption has reached 120 trillion (a roughly 1,000x lift in two years). On the enterprise side, Volcano Engine held a 49.2% share of China's public-cloud LLM token-inference market in H1 2025 (IDC), and full-year 2025 Volcano Engine revenue reached ~¥20B (~$2.9B), doubling YoY. The number of trillion-token-cumulative enterprise customers grew from 100 at year-end 2025 to 140 by May.
What it means for secondary pricing
The pricing tape continues to absorb. The February 2026 General Atlantic secondary at $550B remains the cleanest reference print of the year, and the more recent founder-shareholder equity sale process (Reuters / SCMP, April) cleared demand at $600B+ with multiple bidders competing to upsize the $900M block. The May capex and Doubao-monetization news add to that thesis without yet re-marking the screen.
Our desk this month is seeing indicative bids in the $580B–$650B range, a roughly 3–5% step-up from the April band ($560B–$620B). Bid-ask remains wider than Anthropic but block sizes remain materially better — we can clear $25–50M SPVs cleanly without the crowding that characterizes Anthropic supply. For our Asia-based LPs, ByteDance continues to carry the cleanest currency-and-familiarity profile in our book. We are biased toward filling allocations into the next ByteDance vintage if the founder-equity process formally re-anchors above $600B.
Sources — ByteDance
• Caixin Global — ByteDance Plans Subscriptions for Doubao (May 5, 2026)
• WinBuzzer — ByteDance Plans Over $30B for AI Expansion (May 11, 2026)
• South China Morning Post — ByteDance valuation surges to record $600B+ on proposed equity sale
• The Tech Portal / Reuters — ByteDance touches $550B in GA secondary share sale
━━━━━━━━━━━━━━━━
As always, please reach out with questions or to discuss potential allocations.
— The Interval Capital Team
Disclaimer
This memo is prepared by Interval Capital for the information of existing and prospective secondary target companies. It is not an offer to buy or sell any security and should not be construed as investment advice. Valuations, bid-ask indications, and pricing references reflect our reading of the secondary market as of the publication date and may change without notice. Private company information is sourced from public reporting and our proprietary pipeline; figures should be independently verified before any allocation decision.
