Deep Tech Industry Report: Q3 2026

On 17 September 2026, Crusoe closed the initial tranche of a $3.9 billion Series F at a $30.9 billion post-money valuation, citing more than $140 billion in total contracted value across its vertically integrated AI infrastructure platform (Crusoe). Two weeks earlier, Figure signed for up to 100,000 NVIDIA Vera Rubin GPUs with Nscale at an initial $3.5 billion compute commitment (figure.ai). In July, Commonwealth Fusion Systems raised another $1 billion, taking its total capital to $4 billion (CFS).

That stack is the Q3 2026 Deep Tech story. Capital is concentrating on electrons, tokens, robots, and fusion machines that still have to clear physics and factory gates. Public semiconductor markets, meanwhile, are telling a demand story through ASML's July upgrade: Q2 total net sales of €9.3 billion and a raised 2026 sales guide of €43 billion to €45 billion (ASML / SEC).

This is Inside Deep Tech's Deep Tech Industry Report for Q3 2026 (July through September, with context into late September). Below: five takeaways, five snapshot tables with sources, sector deep-dives across AI infrastructure and semiconductors, physical AI and robotics, quantum, fusion and batteries, and power for AI, plus honest limits, Inside Deep Tech's take, methodology, FAQ, and what to watch in Q4. Companion context: the Physical AI Market Report: September 2026 and the Top 100 Companies in the Deep Tech Industry (2026).

Key Takeaways

• Crusoe's 17 Sep 2026 Series F: $3.9B at $30.9B post-money; company cites $140B+ TCV and 6 GW+ contracted capacity (Crusoe).
• ASML Q2 (15 Jul 2026): €9.3B sales, 86 new lithography systems; FY 2026 sales guide raised to €43–45B; Q3 guide €11–12B (ASML / SEC).
• CFS raised $1B on 30 Jul 2026 (total $4B, ~30% of fusion industry capital to date per CFS) (CFS).
• Physical AI capital: Figure×Nscale $3.5B Vera Rubin compute (3 Sep); XPENG robotics >$900M at >$6.3B post (24–25 Aug); Unitree STAR listing 19 Aug (Figure; XPENG; IDT Physical AI report).
• Quantum + power: Quantinuum Helios C4-Helix QEC results on arXiv (2 Sep); NVIDIA 800 VDC path with LVDC SST Spec v0.3 (Jul 2026) and 80+ ecosystem builders (arXiv; NVIDIA Blog).

Last updated: 30 September 2026 (Q3 close edition).

Q3 2026 snapshot tables

Numbers below are dated and sourced. Where a mark is press-reported rather than company-confirmed, the table says so. Research-house forecasts are labeled as research-house views.

Table 1. Funding and infrastructure commitments

Source line: company newsrooms and named press for Generalist. Here's why that matters. Q3 capital is biased toward AI factories, humanoid compute, fusion equity, and robotics perception silicon, not toward undifferentiated SaaS.

Table 2. Semiconductor and lithography public marks

Source line: ASML 15 July 2026 Q2 press materials via SEC EDGAR and ASML IR presentation. For process context, see Inside Deep Tech's High-NA EUV full guide and EUV lithography / ASML machines full guide. Packaging attach: hybrid bonding full guide.

Table 3. Physical AI and robotics Q3 marks

Source line: company PRs plus TechCrunch for Generalist and Inside Deep Tech's September Physical AI edition for Unitree filing detail. For the monthly Physical AI deep-dive, start here:

Physical AI Market Report: September 2026

Figure's $3.5B Nscale deal, Unitree's STAR IPO, AMR/cobot forecasts, and named movers with sources.

Inside Deep Tech

Table 4. Fusion, batteries, and quantum milestones

Source line: CFS newsroom, Quantonation, arXiv, and Inside Deep Tech solid-state Full Guides. That means… energy deep tech in Q3 was mostly capital and standards work, not new audited MWh or GWh of all-solid-state EV packs in series production.

Table 5. Networking and power standards for AI factories

Source line: Ultra Ethernet Consortium Spec 1.0.3 PDF and NVIDIA's 11 August 2026 800 VDC blog. Primers: Ultra Ethernet full guide, 800 VDC power distribution full guide, co-packaged optics (CPO) full guide.

Sector deep-dives

Let's start with the five buckets that moved budgets in Q3 2026: AI infrastructure capital, semiconductor lithography demand, physical AI hardware and models, quantum error correction proofs, and fusion plus battery factory work.

AI infrastructure: electrons-to-tokens balance sheets

Here's the problem with reading Q3 only through model launches. The scarce inputs for frontier AI remain power, interconnect, and rack-scale systems that ship on a calendar.

Crusoe's 17 September announcement is the clearest AI-infrastructure equity mark of the quarter: an initial close of $3.9 billion Series F at $30.9 billion post-money, co-led by Atreides Management, Mubadala Capital, and Valor Equity Partners, with NVIDIA, GIC, QIA, Founders Fund, Radical Ventures, and TPG among named backers (Crusoe). Company-published operating claims attached to the raise include more than $140 billion TCV, 6 GW+ of gross contracted capacity (with 1 GW delivered and operational), and more than 20× year-over-year growth in Crusoe Cloud bookings year to date. Those are company disclosures, not third-party audits.

On the robotics-compute side, Figure's 3 September Nscale deal commits an initial $3.5 billion of NVIDIA Vera Rubin compute, with intent past $6 billion and up to 100,000 GPUs targeted for initial deployment starting H2 2027 in Barstow, Texas (figure.ai). That is a compute partnership with a strategic investment from Nscale, not a replacement for Figure's September 2025 Series C mark at $39 billion post-money.

Of course, interconnect and packaging still gate how those racks scale. Ultra Ethernet Spec 1.0.3 landed on 16 July 2026 with clarifications and a note adding support for 200 Gb/s per lane signaling (UEC). Optical attach remains a parallel path; see Inside Deep Tech's CPO full guide for when copper scale-up runs out of reach.

Semiconductors: ASML's July upgrade and High-NA readiness

At first glance, Q3 semiconductor news is quieter than model demos. The public tape that matters is ASML's mid-July upgrade, which frames customer CapEx into late 2026.

On 15 July 2026, ASML reported Q2 total net sales of €9.3 billion, gross margin 54.0%, and net income €2.9 billion. New lithography systems sold rose to 86 from 67 in Q1. Management guided Q3 total net sales to €11.0–€12.0 billion and raised full-year 2026 total net sales to €43–€45 billion with gross margin 54–56% (ASML / SEC).

Capacity plans attached to that demand signal: about 65 low-NA EUV tools in 2026 with a planned 30% add for 2027 (and investigation of another 30% for 2028), plus about 130 DUV immersion tools in 2026 on the same +30% / investigate-another-30% cadence (ASML / SEC). ASML's Q2 IR materials separately highlight a High-NA EUV readiness milestone with first high-volume Logic product language and an Intel Foundry quote (ASML IR presentation). That is a process-integration milestone, not a claim that High-NA has replaced low-NA economics overnight. Context: High-NA EUV full guide.

Physical AI and robotics: capital ahead of audited industrial mix

On the flip side, physical AI spent Q3 raising and listing faster than it published new audited industrial revenue mixes.

XPENG's robotics business announced share purchase agreements raising over US$900 million at a post-money valuation over US$6.3 billion (XPENG, 24–25 August 2026), with IDG Capital leading and Tencent and Alibaba as strategic participants. Company guidance: IRON mass production by end of 2026, with broader China and overseas deliveries targeted for 2027. Specs cited include 76 DoF and three Turing AI chips at up to 2,250 TOPS. Those are company figures.

Unitree's 19 August STAR Market listing remains the only mainland public breakout that forces industrial honesty on humanoid revenue mix. Inside Deep Tech's September Physical AI edition walks the filing: industrial application at 9.01% of humanoid revenue in 9M 2025, with guided tours a large slice of that industrial bucket (Physical AI Market Report).

Perception silicon also took a named round: Lyte's $165 million Series C on 2 September at $1.6 billion post-money, led by Maverick Silicon, with production shipping claimed into inspection, logistics, and manufacturing customers (lyte.ai). Separately, TechCrunch reported on 25 August that Generalist reached about a $3 billion valuation after a nearly $200 million extension; Generalist and 8VC did not comment (TechCrunch). Treat that as reported, not company PR.

Quantum: architecture bets and early fault-tolerant demos

But it's not just that. Quantum in Q3 mixed early-stage architecture capital with peer-reviewed error-correction demos.

On 28 July 2026, Quantonation led a $25.5 million seed in ETH Zürich spin-out ZuriQ for a 2D trapped-ion architecture using Penning micro-traps (Quantonation). That is a physics-architecture bet after a $4.2 million pre-seed in 2025, not a claim of thousands of logical qubits in production.

On 2 September 2026, arXiv posted experimental validation of a compact fault-tolerant architecture (C4-Helix) on Quantinuum Helios, a 98-qubit trapped-ion processor. Reported highlights include repeated QEC with error about 4.6×10⁻⁵ per logical qubit per cycle and two-qubit logical Clifford error about 2.8×10⁻⁴ under active error correction, with encoded implementations outperforming unencoded baselines without postselection (arXiv:2609.03194). That is a laboratory architecture result. It is not a commercial SLA.

Fusion and batteries: capital and factories still ahead of MWh

That means… fusion's Q3 headline is equity depth, while solid-state batteries' Q3 headline is licensing and factory prep.

Commonwealth Fusion Systems announced $1 billion of additional equity on 30 July 2026, bringing total capital to $4 billion. CFS states that figure is about 30% of total capital raised by the fusion industry to date, and the largest single fusion round since its own $1.8 billion Series B in 2021 (CFS). Capital funds SPARC demonstration progress and ARC commercial design. It does not equal grid MWh delivered.

On batteries, Inside Deep Tech published a late-September Full Guide series covering QuantumScape's QSE-5 licensing path toward 2029 (QuantumScape guide), Factorial's FEST / Solstice quasi-solid path (Factorial guide), ProLogium Gen4 ceramic cells (ProLogium guide), and Toyota / Idemitsu sulfide factory work with a 2027–28 ASSB BEV target and Chiba electrolyte pilots (Toyota Idemitsu guide). The shared honest limit: OEM slides and pilot tons are not high-volume all-solid-state BEV series production.

Power for AI: 800 VDC leaves the white paper

Of course, none of the AI factory balance sheets work if facility power stays stuck on legacy 54 V rack distribution.

NVIDIA's 11 August 2026 blog restates the 800 VDC AI-factory path developed with Google and Microsoft through OCP, citing a joint white paper from March 2026 and the LVDC Solid-State Transformer Specification v0.3 from July 2026, with more than 80 equipment manufacturers and infrastructure companies building to the specification (NVIDIA Blog). The same post describes an MGX-compatible 800 VDC power rack arriving in the second half of 2026 as a hybrid path into existing AC halls. For the operator-level guide, see 800 VDC power distribution for AI data centers. SMR and other firm power remain part of long-dated campus planning, but Q3's concrete published standard work is the 800 VDC / SST stack, not a new fleet of grid-connected SMRs.

Honest limits: what this data does not show

• Crusoe's $140B+ TCV, 6 GW+ contracted capacity, and 20× Cloud booking growth are company disclosures attached to a raise, not a third-party audit of all contracts.
• Figure's Nscale deal is compute commitment and partnership language, not a new primary equity valuation replacing the Sep 2025 Series C.
• XPENG robotics' $6.3B+ post-money and end-2026 mass-production target are company statements, not audited IRON shipment tables.
• Unitree's industrial percentage remains one issuer. It does not describe Figure, Apptronik, XPENG IRON, or Tesla Optimus.
• ASML High-NA readiness language in IR materials is a process milestone with a named customer quote, not proof that High-NA is the economic default for every advanced Logic layer.
• CFS $4B total capital does not equal SPARC scientific breakeven completed or ARC MWh on the grid.
• Quantinuum Helios C4-Helix results are peer-reviewed-style experimental claims on arXiv. They are not a commercial fault-tolerant cloud product metric.
• Solid-state battery OEM timelines (Toyota 2027–28, QS licensing toward 2029, Factorial/ProLogium pilots) are not evidence of already-scaled all-solid-state BEV volume.
• Generalist's ~$3B mark is TechCrunch sourcing. Company confirmation was not provided in that report.

This report is written for operators, investors, and technical leaders who need citable numbers. It is not a buy list, a valuation model, or a substitute for primary diligence.

Inside Deep Tech's take

The team at Inside Deep Tech will keep industry reports boring on purpose: tables first, opinions labeled, and every metric cut if it cannot be sourced.

Methodology and sources

Primary preference for this edition: company newsrooms (Crusoe, Figure, XPENG, CFS, Lyte, NVIDIA Blog), ASML Q2 2026 materials via SEC EDGAR and ASML IR PDF, Ultra Ethernet Consortium Spec 1.0.3, Quantonation / ZuriQ seed note, and arXiv:2609.03194. July through late September 2026 sources are prioritized. Older figures appear only when they remain the best public authoritative number and are date-stamped inline (for example Unitree 9M 2025 mix via the September Physical AI report).

Every quantitative claim in the tables and body above maps to a named source and date in Inside Deep Tech's research log for this edition. Unsourced metrics were cut. Reported-only private marks (Generalist) are labeled as reported.

Internal context used for cross-links (HTTP 200 as of 30 September 2026): 800 VDC, CPO, High-NA EUV, hybrid bonding, Ultra Ethernet, QuantumScape, Factorial, ProLogium, Toyota/Idemitsu, Physical AI Market Report September 2026, EUV / ASML machines guide, and Top 100 Deep Tech companies.

FAQ

What to watch in Q4 2026

Q4 is for evidence that survives outside raise announcements.

• ASML's Q3 print versus the €11.0–€12.0 billion guide and any High-NA shipment/attach commentary that moves beyond IR milestone language.
• Crusoe follow-through on contracted GW energization and Cloud bookings after the Series F close.
• Figure Index throughput plus Nscale / Vera Rubin site milestones into 2027.
• XPENG IRON end-2026 mass-production claims versus first customer-visible deployments.
• Any second public humanoid issuer forced to break out industrial revenue the way Unitree did.
• CFS SPARC construction and science milestones that change the 2027 breakeven narrative.
• 800 VDC power-rack deployments into existing AC facilities, not only partner announcements.
• Solid-state: PowerCo / QS licensing milestones and Toyota/Idemitsu Chiba pilot progress toward 2027.

For the wider company map, see the Top 100 Companies in the Deep Tech Industry (2026). The next industry report in this series will cover Q4 2026 after year-end primary disclosures land.

Related Inside Deep Tech guides: 800 VDC · CPO · High-NA EUV · Hybrid bonding · Ultra Ethernet · QuantumScape · Factorial · ProLogium · Toyota/Idemitsu.