Meta is spending faster than it is producing cash.

What changed: Meta spent $34.91 billion on capital equipment and finance-lease principal in Q3 and narrowed its full-year range to $135–145 billion. Revenue rose 26% to $64.60 billion, but free cash flow was negative $2.12 billion. Meta also approved a new non-voting Class C share that can be used for capital raising, compensation or acquisitions.

Why it matters: The demand signal for chips, networks and power remains very strong. The cash number is the warning. Even Meta’s advertising engine is no longer funding this build without pressure, and the new share class gives it another financing tool.

Negative / caveat: Negative free cash flow, higher lease obligations and potential Class C issuance raise the cost of the build. AMZN, MSFT and GOOGL compete for the same chips, power and customers. Meta did not disclose supplier allocations.

Nebius bought time. It did not solve the funding question.

What changed: Nebius secured a $3.5 billion senior secured facility from Citi. It matures in 364 days and initially costs Term SOFR plus 2.75%, rising as high as 3.50% plus adjustments. Proceeds can fund data centers, GPUs and general corporate needs.

Why it matters: Capacity can keep moving while Nebius arranges longer-term capital. But a one-year facility turns construction and customer-contract timing into a refinancing issue.

Negative / caveat: The $40 billion of customer commitments cited by Nebius are not current revenue or cash. CRWV and IREN face a better-funded competitor. The short maturity, floating rate and secured claim increase leverage and refinancing risk.

Vicor’s royalty model showed up in earnings.

What changed: Q3 product and royalty revenue reached $197.1 million, up 37.5% sequentially and 78.5% year over year. Royalty revenue was $66.6 million. Gross margin rose to 68.3%, backlog reached $406 million and operating cash flow was $69.7 million. Vicor also granted its first VPD license to an AI OEM.

Why it matters: Vicor can earn from high-density power delivery even when another supplier builds the module. That makes the model less dependent on winning every hardware socket and explains the jump in margin and cash.

Negative / caveat: Royalty income was one third of quarterly revenue and can be uneven. Litigation, customer concentration and the plan for nearly five times Fab-1 capacity create execution risk. Competing power-module vendors face licensing or displacement pressure.

Tesla put Cybercabs into service, but the economics are still missing.

What changed: Tesla said it began providing Robotaxi rides with Cybercabs in Austin on September 3 and that cumulative unsupervised miles increased eightfold from Q2. It also said data-center demand is lifting energy-storage demand and that its Texas Megapack factory has started production with 50 GWh of planned capacity.

Why it matters: This moves Cybercab from a manufacturing milestone to operating use. The energy business also gives Tesla a separate way to earn from data-center power demand.

Negative / caveat: Tesla did not disclose Cybercab fleet size, paid rides, revenue or unit economics. Q3 operating income was $718 million on $28.05 billion of revenue, so the new programs still sit inside a capital-heavy business.

Ouster has operating evidence from a hard environment.

What changed: Forterra vehicles using Ouster Rev7 OS1 lidar completed 1,426 missions, 88 casualty evacuations and more than 7,300 miles in Ukraine. Forterra built and delivered 105 vehicles in under six months and plans to use Rev8 in its next generation.

Why it matters: Dust, darkness, weather, vibration and GPS denial are useful tests of sensor reliability. That evidence is more valuable than another pilot announcement and supports Ouster’s defense case.

Negative / caveat: The release gave no Ouster sensor count, contract value, new order or revenue. HSAI and other lidar suppliers still compete on cost and performance.

PHYSICAL AI · OCTOBER 17–23, 2026

Vicor changed the numbers. Ouster changed the evidence.

Every required company was checked. Vicor delivered revenue, margin, backlog and cash. Ouster supplied field-use data but no order value. The other nine did not disclose a new binding order or estimate change.

What matters next: ALGM on October 29; OUST on November 3; VPG and ALNT on November 4; INOD and MP on November 5; and HSAI and NOVT on November 10. AMBA, VICR and SYM had no confirmed next report date at the research cutoff.

The November report has to show that growth is broadening.

What changed: No material change found in the sources checked. Innodata only confirmed that Q3 results will be released November 5. No new customer, order, guidance or filing changed the case during October 17–23.

Next catalyst: November 5 results and call: revenue growth outside the two largest customers, paid work from the motion-capture lab, margin and cash conversion.

Risk: The top two customers represented 71% of Q2 revenue. New labs and datasets do not reduce concentration until they produce repeat revenue.

Valuation relevance: No revenue, margin, cash-flow or share-count change was disclosed this week.

X7 still needs a production order.

What changed: No material change found in the sources checked. Ambarella made no new financial filing or customer announcement in the review window.

Next catalyst: Named X7 production designs, customer commitments and shipment timing. The next earnings date was not confirmed.

Risk: Sampling can take several quarters to become revenue. The Hanwha agreement also needs to convert from potential value into shipments.

Valuation relevance: No change to guidance, margin, cash flow or share count. Fiscal Q3 guidance remains $115–124 million of revenue and 59–60% non-GAAP gross margin.

This is field use, not another lab demonstration.

What changed: On October 21, Ouster said Forterra vehicles using its Rev7 OS1 lidar had completed 1,426 missions, 88 casualty evacuations and more than 7,300 miles in Ukraine. Forterra built and delivered 105 vehicles in under six months and plans to use Rev8 in next-generation vehicles.

Next catalyst: November 3 results: defense sensor volume, Rev8 conversion, acquisition integration and whether operational proof produces orders.

Risk: The release disclosed no Ouster unit count, contract value, new binding order or revenue. Battlefield performance does not establish commercial scale.

Valuation relevance: The product case strengthened, but estimates did not. Q2 remains the baseline: $55 million of revenue, 49% GAAP gross margin and a $4 million adjusted EBITDA loss.

November 10 is the next financial test.

What changed: No material change found in the sources checked. On October 20 Hesai only set its Q3 report date. It did not add an order, shipment figure or estimate change after last week’s Kosmo and actuator launch.

Next catalyst: November 10 results: Kosmo revenue, ETX ramp, actuator orders and whether the full-year SGI revenue guide changes.

Risk: China competition, export controls and U.S. market access remain material. Partnerships and delivery targets are not binding backlog.

Valuation relevance: No change to revenue, margin, cash flow or share count this week. The SGI revenue guide remains RMB200–300 million.

October 29 should show whether data-center mix is holding.

What changed: No material change found in the sources checked. Allegro issued no new order, filing or estimate change during October 17–23.

Next catalyst: October 29 results: data-center sales, robotics design wins, automotive demand and gross margin.

Risk: Automotive still dominates revenue. A robotics premium needs disclosed production revenue, not broad market labels.

Valuation relevance: No new input. The prior fiscal Q2 revenue guide remains $265–275 million.

The next report is about integration and cash.

What changed: No material change found in the sources checked. No filing or customer announcement changed the automation and robotics case this week.

Next catalyst: November 9 results and November 10 call: Riverpoint integration, organic automation growth, leverage and cash conversion.

Risk: Acquisition integration and return on invested capital. The NVIDIA Halos relationship is a certification path, not an order.

Valuation relevance: No change to guidance, margin, cash flow or share count. Full-year revenue guidance remains $1.130–1.140 billion.

The humanoid nomination still needs booked orders.

What changed: No material change found in the sources checked. VPG made no new filing or customer disclosure during the review window.

Next catalyst: November 4 results: humanoid orders, production shipments from the August nomination and gross-margin recovery.

Risk: A vendor nomination is not a binding order. Mix, foreign exchange, ERP execution and small initial robotics volumes can delay earnings.

Valuation relevance: No change to the $84–89 million Q3 revenue guide, margins, backlog, cash generation or share count.

A report date is not a business change.

What changed: No material change found in the sources checked. On October 21 Allient set its Q3 reporting schedule. The October 22 Vyxalon controller launch disclosed no order, customer, price or revenue.

Next catalyst: November 4 results and November 5 call: conversion of the $298 million backlog, defense and robotics orders, gross margin and cash.

Risk: Product launches and certifications remove technical hurdles but do not prove demand. Backlog timing and industrial execution remain the key risks.

Valuation relevance: No weekly guidance, margin, cash-flow or share-count change. Q2 orders of $201.3 million and 1.31 book-to-bill remain the baseline.

The royalty model is now visible in the numbers.

What changed: On October 20, Q3 product and royalty revenue reached $197.1 million, up 37.5% sequentially and 78.5% year over year. Royalty revenue was $66.6 million. Gross margin rose to 68.3%, backlog reached $406 million and operating cash flow was $69.7 million. Vicor also disclosed its first VPD license with an AI OEM and plans capacity approaching five times Fab-1 across three fabs.

Next catalyst: The Q3 10-Q, additional VPD licenses, product-versus-royalty mix and evidence that the $406 million backlog converts without heavy working-capital needs. No date is confirmed.

Risk: One third of Q3 revenue was royalty income, and licensing can be uneven. Customer concentration, litigation dependence, fab execution and spending ahead of demand remain material.

Valuation relevance: The business value improved: revenue, gross margin, backlog, earnings and cash all rose. Net income was $63.3 million, or $1.33 per diluted share; capex was $16.1 million and cash ended at $513.1 million.

Production and qualification evidence still comes next.

What changed: No material change found in the sources checked. MP disclosed no new qualification, binding magnet order or production update during October 17–23.

Next catalyst: November 5 results: finished-magnet qualifications, Independence and 10X yields, capital needs and the split between precursor and finished-magnet revenue.

Risk: Capital intensity, commodity exposure, policy dependence, geopolitics and production yield. Price-protection income is not product revenue.

Valuation relevance: No change to guidance, margin, cash flow, capital needs or share count. Q2 Magnetics revenue remains the $16.5 million baseline.

Cash conversion remains the issue.

What changed: No material change found in the sources checked. No deployment, contract or filing changed the fiscal Q3 baseline during October 17–23.

Next catalyst: Fiscal Q4 results: performance against the $760–780 million revenue and $100–105 million adjusted EBITDA guides, system acceptance and cash conversion. The report date was not confirmed.

Risk: Customer concentration, installation timing and working capital can separate reported growth from cash. The public-share count understates the full ownership structure.

Valuation relevance: No weekly guidance, backlog-quality, cash-generation or share-count change. Q3 free cash flow was negative $164.6 million.