Yes, SpaceX is in the corpus. It's tracked as Space Exploration Technologies Corp (CIK 0001181412, Nasdaq: SPCX) with full coverage, and it has no annual report yet.

Annual report: There is no 10-K on record. The only periodic report is a 10-Q for the quarter ended June 30, 2026, filed August 4. The fiscal year ends December 31, so the first 10-K would cover FY2026 and should arrive in early 2027. The corpus was last synced September 30, and the newest filing it holds is dated September 24. The S-1 would have included audited annual financials, but that's a registration statement, not an annual report.

What it has filed (48 filings since March 30, 2026):

• IPO process: A confidential draft registration (DRS) on March 30 and an amendment on May 7. The public S-1 followed on May 20, with two S-1/As on June 1 and June 3. The registration went effective June 11, and the final prospectus (424B4) was filed June 12. Two Form 8-A12B listing registrations were filed June 10.
• Free writing prospectuses: Seven FWPs between June 4 and June 11, including interview transcripts and region-specific materials.

• 8-Ks: Eight in total.

  • Three in mid-June cover charter and governance changes, a material agreement and unregistered equity sales, and an officer or director change.
  • Three more from June 22–26 cover a senior notes launch, pricing and closing.
  • An earnings release (Item 2.02) followed on August 4.
  • An August 14 8-K reports an acquisition completion and unregistered sales (Items 2.01 and 3.02).

• Employee plans: Three S-8s, one on June 12 and two on August 4.

• Ownership filings: Ten Form 3s (mostly June 11), three Form 4s, one Form 144 on September 22, and four Schedule 13Gs filed August 11–14.

The full filing list, with links to each filing on SEC EDGAR, is at the timeline page.

I can dig into any of these, such as the Q2 10-Q financials, the senior notes 8-Ks, or the insider transactions.

SpaceX treats chip supply as a material risk to its AI segment. The fullest statement is in the IPO prospectus (424B4, filed Jun 12, 2026), under a risk factor headed "Our ability to scale our AI products relies on our terrestrial and orbital AI compute infrastructure…". The filing is long, so I'm keeping direct quotes short and paraphrasing the rest. The links open the filing with the passage highlighted.

No long-term chip contracts. The company says GPU manufacturing is limited to a small number of qualified suppliers. It has no long-term or other material contractual arrangements with its direct chip suppliers and is "procuring all of our GPUs on a purchase-order basis." Risk factors, clause 21

Upstream fab concentration. Its chip suppliers depend on a small group of advanced fabrication plants. The filing names fab capacity limits, shortages of inputs like silicon wafers or rare earths, geopolitical tensions, and natural disasters as things that could stop suppliers filling its orders. It says this could have a material adverse effect on the business. Risk factors, clause 22

Orbital AI needs far more chips than it has. The filing says orbital AI at scale requires significantly more AI chips than are currently available to it. Its answer is "Terafab," an internal chip-manufacturing effort, but it warns Terafab may not succeed or hit its timelines. It also says it will keep sourcing a significant share of compute hardware from third parties. The Tesla relationship is only a framework agreement, and neither Tesla nor Intel is obligated to stay in the project. Same link as above.

Suppliers can favor others. The company says it may be unable to get AI processors at acceptable prices, volumes or timing. Its suppliers also serve competitors and could prioritize them, raise prices on short notice, demand onerous prepayments, or delay deliveries. The stated consequences are service outages and delayed data center and network upgrades. Same link as above.

The same language appears in the June 1 S-1/A, and a June 5 free-writing prospectus repeats it.

10-Q update. The Aug 4 10-Q says there are no material changes to the prospectus risk factors except for one new one. That one is about AI infrastructure and data center operations. It cites "supply chain disruptions, equipment shortages" among the things that could impair its ability to deploy capacity. It adds that the AI infrastructure business is concentrated in a few customers, and that cloud agreements can be ended on 90 days' notice. So the chip-specific language comes from the prospectus, not the 10-Q.

These results are AI-ranked retrievals, so check the highlighted passages in the filings before relying on them. I can also search the free-writing prospectuses and the S-1 business section for what SpaceX says about Terafab and its planned chip capacity.

SpaceX has disclosed two cloud-capacity deals with specific terms, plus a Cursor compute agreement for which only termination-related amounts appear. The Anthropic deal is in the final prospectus (424B4). The Google deal I found only in a free writing prospectus (FWP).

1. Anthropic PBC (May 2026)

• Capacity: about 325,000 NVIDIA GPUs across the COLOSSUS and COLOSSUS II data centers.
• Fee and term: $1.25 billion per month through May 2029. The fee is reduced during a ramp in May and June 2026.
• Cancellation: after an initial three-month period, either party can terminate on 90 days' notice.
• Source: all four terms appear in the 424B4 Business section.
• Discrepancy: the 424B4 financial statement note gives a narrower account. It describes a single agreement signed May 3, 2026, gives no dollar amount, and mentions a reduced fee only in May. It also states the 90-day termination right without the three-month carve-out.

2. Google LLC (signed June 5, 2026)

• Capacity: about 110,000 NVIDIA GPUs, plus CPUs, memory and related components.
• Fee and term: $920 million per month from October 2026 through June 2029, with a reduced fee while capacity ramps up through September. Source: FWP filed June 5.
• Cancellation: if SpaceX hasn't delivered the committed GPUs by September 30, 2026, then after a one-month grace period Google can terminate immediately. Alternatively, Google can accept the GPUs actually delivered, with a pro rata cut in the monthly fee. After December 31, 2026, either party can terminate on 90 days' notice. Source: same FWP.
• 424B4 check: I did not find Google in the final prospectus. I ran two searches scoped to that filing. One used the Anthropic/Google/GPU/termination language, and the other named Google, 110,000 GPUs, $920 million and the dates. Neither returned a Google passage.
• Later filings: a sweep of filings after June 6, including the 10-Q and the August 8-K, also didn't surface it. The 8-K earnings deck credits the segment's growth partly to "new Cloud Service Agreements" without naming counterparties, so I can't tie it to Google.

3. Cursor / Anysphere (compute agreement of April 19, 2026)

• What's disclosed: the 424B4 says Cursor is owed an $8.5 billion deferred services fee under the compute agreement. That applies if SpaceX terminates the option to acquire Cursor, or if Cursor terminates over SpaceX's material breach (after notice and cure). A separate $1.5 billion fee applies under the option agreement.
• Purpose: an FWP deck note describes the agreement as a collaboration on improving Grok and possibly jointly developing models.
• Not found: GPU count, monthly fee, term and ordinary cancellation rights were not found in the corpus by these searches. I searched the prospectus summary, business section and financial notes.
• Later development: the 10-Q summary says SpaceX exercised its option in June and signed a merger agreement, expected to close in Q3 2026. That may make the termination-fee scenario moot, but I haven't confirmed the closing.

Not counted: a $6.587 billion, five-year Valor equipment lease for AI hardware, where SpaceX is the lessee, and unspecified "third-party cloud capacity" purchase commitments in the 10-Q, with no counterparties named.

My Google conclusion rests on targeted searches, not a full read of the 424B4. I can read the whole prospectus and the 10-Q's commitments note to settle it.

The 10-Q (quarter ended June 30, 2026, filed August 4) adds one risk factor, on AI infrastructure customer concentration and the short notice on its cloud contracts. Its only risk factor update says it supplements the prospectus risk factors, with no other material changes. A clause-level comparison of the two filings agrees: one clause added, and the other 65 are just the prospectus's own risk factors, which the 10-Q doesn't repeat.

What the new risk factor says (10-Q, Part II, Item 1A):

• Growing reliance on AI. SpaceX says an increasing share of consolidated revenue comes from the AI segment, driven mainly by its AI infrastructure business. That business depends on continued AI adoption. Efficiency gains, alternative architectures or weaker competitiveness could cut demand for its capacity.
• Customer concentration. AI infrastructure revenue is concentrated in a small number of customers.
• Short exit rights. The cloud services agreements generally carry monthly fees. After an initial ramp period of a number of months, they "may be terminated by either party upon 90 days' notice."
• Consequences. It warns that losing a major customer, non-renewal or termination, an inability to replace the business on comparable terms, or a shift to in-house infrastructure could materially hurt AI segment revenue, results and cash flows.
• Operational risks. The cloud services also depend on reliable data center operations. It lists physical damage, natural disasters, cybersecurity incidents, construction delays, power constraints, supply chain problems, equipment shortages, regulatory restrictions and permit delays.

How new it is. The 10-Q labels this a supplement, but the prospectus wasn't silent on adjacent points, so "new" depends on how strictly you read it:

• The prospectus already said SpaceX has a limited number of customers for its AI products.
• It also said some compute customers may not be cash flow positive and may rely on outside capital to pay SpaceX.
• It disclosed the 90-day termination terms of the Anthropic agreement in its business section.
• What I didn't find in the prospectus's risk factors, using targeted searches, is a dedicated risk factor tying revenue concentration to the 90-day termination rights and their effect on revenue and cash flow. That framing is what the 10-Q adds.

The comparison tool is AI-paired and the searches are AI-ranked, so check the exact wording against the linked filings.

Revenue rose 91.943012% and R&D rose 81.205312% from Q2 2025 to Q2 2026. These come from the XBRL-tagged figures in the 10-Q for the quarter ended June 30, 2026, not from the text. The percentages were computed exactly, not estimated.

Three months of the compute fees, as a share of Q2 2026 revenue ($7,814M)

I used the contractual monthly fees from the earlier answer ($1.25B for Anthropic, $920M for Google), each multiplied by three.

This is a full-rate comparison, not what SpaceX recognized in Q2.

• Anthropic: its fee was reduced during the May and June ramp, so Q2 recognized revenue was lower than $3,750M.
• Google: its $920M monthly fee doesn't start until October 2026, with reduced fees before then. It contributed little or nothing to Q2.

The Q2 earnings deck reports AI solutions and infrastructure revenue of $2,194M (slide). That is well below the $3,750M that three months of the full Anthropic rate would imply. The percentages are best read as the size of the contracts against the quarter's revenue base.

The "R&D" comparison uses the single consolidated tag, ResearchAndDevelopmentExpense. The two year-ago figures are the comparatives presented in the 10-Q, not separately filed prior-year reports, since SpaceX had no earlier periodic filings.

Revenue concentration among NVIDIA's direct customers has risen over the last three fiscal years. Its fiscal year ends in late January, so FY2026 ended Jan 25, 2026. NVIDIA only names direct customers at or above 10% of revenue, so the figures below are a floor on concentration, not a full top-N breakdown.

• Overall trend: the share of revenue from customers at or above 10% went from 13% to 34% to 36%. The share from the largest single direct customer went 13%, then 12%, then 22%.
• Sharp jump in FY2026: the largest customer's share nearly doubled from FY2025 to FY2026.
• Indirect concentration: NVIDIA also estimates indirect customers, which buy through its direct customers. One indirect customer was estimated at about 19% of FY2024 revenue, and an indirect customer was estimated at 10% or more in FY2025.

Latest interim periods (fiscal 2027)

• Q1 (ended Apr 26, 2026): three direct customers at 21%, 17% and 16%, or 54% combined. A year earlier, two were at 16% and 14%. Source
• Q2 (ended Jul 26, 2026): only one direct customer was at 10% or more, at 16%. A year earlier, two were at 23% and 16%. For the first half, three customers were at 16%, 15% and 13% (44% combined), against two at 20% and 15% (35%) a year earlier. Source

So concentration peaked in Q1 and eased in Q2, but the first half is still above the prior-year first half. The Q2 easing may reflect a customer slipping just under the 10% threshold, which makes it disappear from the disclosure. The 10-Qs also say an AI research and deployment company contributed a meaningful amount of revenue by buying cloud services from NVIDIA's customers.

Caveats

• Customer labels: NVIDIA's "Customer A/B/C" labels are reassigned each filing, so I can't say the same customers are involved from year to year.
• Definition change: the FY2026 10-K describes direct customers as including cloud service providers and AI model makers. The FY2025 10-K did not list those. Some of the FY2026 rise may reflect that change in how NVIDIA categorizes direct customers, not only new buying patterns.
• Method: these are search-retrieved passages from the filings' MD&A and notes, not XBRL data. The corpus has no tagged customer-concentration facts I could pull.

The redline is open in the panel above, and this link opens the full Item 1A comparison of the FY2025 and FY2026 10-Ks in your browser: full redline. It is an AI-paired comparison, so some ADDED/REMOVED labels may be imperfect. The filings themselves are FY2025 and FY2026.

The redline flags 83 changed clauses, but 63 are the same language re-wrapped, leaving about 20 substantive changes. Several that look new are not. The purchase-obligations-and-prepaids warning, the customer-insolvency list and the spread of export-control harm to Europe, Latin America and Southeast Asia all appear in the FY2025 10-K, so I excluded them. I can only reproduce short quotes from filings, so most of what follows is paraphrase and the links open the exact passages.

1. NVIDIA may finance customers' and partners' data center buildouts.

• Quote: "We have not entered into any financing arrangements."
• What's new: NVIDIA says it may provide such financing and warns of counterparty risk. That includes customers or partners being unable to meet financial commitments, project delays, and counterparty distress or insolvency. Any financing could mean lower upfront cash through extended or multi-year payment terms and higher credit risk.
• Check against FY2025: that filing had extended-payment-terms language but nothing on financing data center buildouts, as far as my searches found.
• Link: FY2026 10-K, Item 1A
• In the redline: clause P1.I1A.23.

2. Customers' power, data center and capital constraints. This is a partly new, expanded risk.

• What's new: a dedicated paragraph says the availability of data centers, energy and capital for customers' and partners' AI buildouts is "crucial". It says expanding energy capacity is a complex multi-year process, and that less-capitalized companies may struggle to finance large projects. Any of this could delay deployments and reduce the scale of AI adoption. A new bullet also lists customers' and partners' ability to secure capital and energy and build data center infrastructure on time as a demand factor.
• Check against FY2025: the prior filing only had two shorter bullets, on data center capacity or energy for customers and on customers' ability to invest in AI infrastructure. The full paragraph and the capital emphasis are new.
• Link: FY2026 10-K, Item 1A
• In the redline: the ADDED clause P1.I1A.6. The first sentence there is carried-over text, so read the second half.

3. China's public questioning of H20 chip security.

• What's new: after US legislative proposals for mandatory chip features, China's government publicly questioned whether the H20 has built-in vulnerabilities, which discouraged customers from buying it. NVIDIA says it publicly responded that its GPUs, H20 included, have no such vulnerabilities. The same passage also warns about restrictions on products supporting open-source models from China such as DeepSeek and Qwen. I didn't separately verify whether that part is new.
• Check against FY2025: that filing covered Chinese restrictions on Micron and an Action Plan on data center standards, but not this episode. The surrounding export-control discussion is largely carried over.
• Link: FY2026 10-K, Item 1A
• In the redline: clause P1.I1A.33.

Other changes in the redline I did not rank or verify:

• The customer-concentration risk has updated figures (a 14% customer appears in the FY2026 wording).
• Rewritten language covers new software, services and cloud business models.
• NVIDIA's own cloud-services arrangements now carry a warning that timing and availability have shifted.
• Geopolitical wording on Israel has been added.

My "new versus carried over" calls come from targeted searches of the FY2025 text, not a full read of both filings. The redline will show anything those searches missed.

Yes. NVIDIA's 13F filing for the quarter ended June 30, 2026 (filed Aug 14) reports 122,764,805 Class A shares of SpaceX. That is the latest quarter available, since the September 30 13F isn't due until mid-November.

The position

• Size and value: the reported value is $20,975,594,582, or about $21.0 billion. That implies about $170.86 per share, against the $135 IPO price.
• Share of NVIDIA's portfolio: it is about 33% of NVIDIA's $63.4 billion of reported 13F holdings, spread across eight positions.
• Control: NVIDIA has sole investment and voting discretion. This comes from the 13F data (accession 0001045810-26-000065).
• Ownership percentage: it is a minority stake, roughly 1.6% of the Class A. I estimated that using Alphabet's Schedule 13G as the share-count yardstick (551.2 million shares reported as 7.2%). It is a rough figure. NVIDIA is not among the four holders above 5% that filed Schedule 13G (Musk, Alphabet, Gracias, Thiel).

What it tells you

• NVIDIA has a large financial interest in SpaceX, so it gains both as a chip seller and as a shareholder.
• It is a long-term holding by a company with sole discretion over it. This is a snapshot, so it shows nothing about changes over time.

What it doesn't tell you

• Chip purchases: a 13F reports a securities holding, not a supply relationship. It says nothing about how many GPUs SpaceX buys, from whom, or on what terms.
• Where the dependency evidence is: that comes from SpaceX's own filings. The prospectus says the capacity sold to Anthropic is about 325,000 NVIDIA GPUs. The Google agreement covers about 110,000 NVIDIA GPUs. Its risk factors add that it has no long-term contracts with direct chip suppliers and buys all GPUs by purchase order. SpaceX's planned answer is its in-house Terafab chip effort, which it says may not succeed.
• NVIDIA as a customer concentration: NVIDIA doesn't name its large customers, so I can't tell whether SpaceX is one of the direct customers at 10% or more of its revenue.
• Acquisition: I found no filing explaining how NVIDIA acquired the shares. SpaceX only listed in June, so this is the first 13F where the stake could appear. I didn't verify whether it came from an earlier private investment.
• Scope: 13Fs show only long US-listed positions as of the quarter end. The share price has moved, and NVIDIA's holding may have changed since June 30.

The stake shows NVIDIA is financially tied to SpaceX, but the evidence that SpaceX relies on NVIDIA GPUs comes from SpaceX's contracts and risk disclosures, not from the shareholding.

The memo came back as a three-page Word document with 49 links. The whole memo, as written, opens from the button below, and the Word file itself can be downloaded. In brief, its opening section and its two closing sections are below. The middle two sections, the table of the three compute arrangements and the NVIDIA context, carry the same facts as Steps 3 and 6 to 8 above.

Read the whole memo ↗ Download the Word file

1. The dependency

SpaceX buys all of its GPUs on purchase orders from a small number of qualified suppliers, with no long-term supply contract, while it sells GPU capacity under large monthly contracts: about 325,000 NVIDIA GPUs to Anthropic at $1.25 billion a month, and about 110,000 to Google at $920 million a month, the Google deal carrying a 30 September 2026 delivery condition. It also says orbital AI needs far more chips than it has, and that its in-house chip project, Terafab, may not succeed. The memo's conclusion: SpaceX needs outside chip supply both to build and to honor its revenue contracts, and has not secured that supply under long-term contract.

4. What this analysis does not establish

The memo names six things it could not establish. First among them is who supplies SpaceX's GPUs, and in what volume: the filings name NVIDIA GPUs only as the hardware in the capacity SpaceX sold, and the memo's searches found no supply agreement with NVIDIA, which it says is not proof that none exists. The other five are whether SpaceX is one of NVIDIA's 10%-plus customers, whether NVIDIA's stake affects SpaceX's allocation, pricing or priority, whether Google's 30 September delivery condition was met, the full terms of the Google and Cursor deals, and NVIDIA's position after 30 June.

What I inferred rather than read in a filing

It then marks five things as its own inference rather than something read in a filing: NVIDIA's ownership share, roughly 1.6% of the Class A, which it calls rough because two 13G filers imply different share counts; that NVIDIA holds under 5%; its own arithmetic on the 13F figures, such as the stake being about a third of NVIDIA's reported portfolio; its “two-sided dependency” characterization; and how NVIDIA came to hold the shares, which it did not find stated and calls any link to SpaceX's merger with xAI speculation.

That list is part of the output, not an apology for it. A research tool that cannot tell you where its answer stops is not one you can file work product behind.

How it was produced

Nobody chose those lanes by hand. The assistant routed between them based on the kind of question being asked, sometimes from the plan the first call returned and sometimes on its own, which is what the tool is designed to make possible.