AI
A filing at the model layer, not the silicon layer
Eight applications published this week carry an NVIDIA assignee. The one worth reading is about how many tokens reach the language model — a cost question dressed as an architecture claim.
Latest Articles
24 articles indexedAI
Five jurisdictions cleared, one vote left
A Bulgarian foreign-investment approval on 20 July removed the last regulatory condition to the SoundHound-LivePerson mergers. What remains is a shareholder vote, and both sides are now filing to win it.
Filing Signal
AI
Wintergreen's S-4/A Values AdTech Target KIKA at $80 Million, With No Fairness Opinion
The amended de-SPAC prospectus fixes KIKA Technology's valuation at $80,000,000 and discloses that Wintergreen's board obtained a valuation report but not a fairness opinion. Nine-month revenue rose 1,293.8% while the customer count fell from 32 to 25.
AI
Sony Application Binds Generative AI Output Quality to a Licence Key
A Sony Group Corporation application published July 16, 2026 describes generative AI output that embeds its own authentication key and comes out better when that key is valid. The mechanism treats licensing as a property of the content, not a gate in front of the model.
Artificial Intelligence
Microsoft's July 16 Filings Point at Inference Cost — and the Fix Is Software, Not Silicon
Twelve Microsoft applications published July 16, 2026. Three of them attack the cost of running a model rather than the cost of training one, and two share an inventor team.
Artificial Intelligence
SoundHound Restates Its LivePerson Merger Agreement: A New Two-Merger-Sub Structure and Revised Terms
A July 2 Form 8-K and Rule 425 filing disclose an Amended and Restated Merger Agreement that replaces SoundHound AI's April 21 deal for LivePerson with a two-step, Merger Sub I/II structure, a $42,784,532.64 base consideration amount, a $7-to-$12 stock-price collar, and an October 21, 2026 outside date.
Artificial Intelligence
A Week of Filings Points at the Control Plane for AI Agents
In the July 2, 2026 publication drop, Microsoft published a cluster of applications on agent permissions, delegation, state, and authoring — the governance scaffolding around autonomous AI, not the models themselves. Read together, they suggest where the enterprise-AI effort is pointed.
Artificial Intelligence
NVIDIA's Latest Grant Batch Is Mostly Software: What Two Dozen Patents Issued in One Day Signal
On June 30, 2026, more than twenty patents assigned to NVIDIA issued at once — and the cluster leans heavily toward the software around the silicon: CUDA APIs, serverless GPU orchestration, model compression, and an LLM that troubleshoots systems. Read as a group, the grants signal where NVIDIA is directing its patented R&D.
AI Business
Applied Digital Upsizes Its Revolver to $430M and Lifts a Preferred Commitment to $2B in One 8-K
A single June 26 filing enlarges Applied Digital's secured revolving credit facility to $430.0 million, expandable to $550.0 million through 2029, and raises a Series G preferred-equity commitment from $1.59 billion to $2.0 billion. The capital stack is being widened around data-center project financing.
Artificial Intelligence
This Week's AI Filings Have No Owner: Read the Pub Drop and AI IP Looks Horizontal, Not Concentrated
Roughly 280 machine-learning and neural-network applications published in a single week, and no assignee accounts for more than a low-double-digit share. The spread, across telecom, chips, and enterprise, is itself the signal.
AI Business
Spin AI Files S-1 to Register a $100,000 Best-Efforts Offering at $0.01 a Share
An early-stage Wyoming SaaS company that scores research-commercialization signals registered up to 10,000,000 shares with the SEC on June 23, 2026. The filing discloses three active subscribers, a going-concern paragraph, and net proceeds that could fall to roughly $68,102.
AI Business
Verses AI 8-K: President and COO Resigns Citing Non-Payment of Earned Wages as Company Says It Will Discontinue AI R&D
A Form 8-K filed June 22 reports President and COO James Hendrickson resigned effective immediately over disagreements that included non-payment of earned wages to W-2 employees, and discloses the company's decision to discontinue its AI research and development and pursue alternative business opportunities.
AI Business
Getty Images Files 8-K Disclosing a Multi-Year Display Agreement to Put Its Licensed Content Inside ChatGPT
A Form 8-K filed June 22 attaches a press release announcing that Getty Images' licensed content will appear across OpenAI search and discovery experiences in ChatGPT. The filing says the deal was referenced on an unnamed basis in the Q3 2025 earnings call.
Artificial Intelligence
What 'MD&A' Is in a 10-K — and What Item 303 Requires AI Companies to Explain
MD&A is the section where management explains the numbers in its own words. Here is what Regulation S-K Item 303 requires, and why it is the place AI companies must address known uncertainties in their results.
Artificial Intelligence
What 'Remaining Performance Obligations' (RPO) Disclose About an AI Company's Backlog
RPO is the contracted revenue a company has not yet recognized. For AI and cloud businesses it functions as a backlog signal. Here is what the figure means and where the revenue-recognition rules require it.
Artificial Intelligence
What 'Purchase Obligations' Are in a 10-K — and Why AI Compute Commitments Live There
Beyond the capex line, AI companies commit billions to future supply and cloud capacity. Those commitments are disclosed as purchase obligations. Here is what the term means and where to find it.
Artificial Intelligence
What a 'Non-GAAP' Financial Measure Is — and the Reconciliation the SEC Requires
AI companies routinely report adjusted earnings and margins alongside their official numbers. Here is what makes a measure non-GAAP, and the disclosure rules — Regulation G and Item 10(e) — that govern how it must be presented.
Artificial Intelligence
What 10-K 'Risk Factors' Are — and How AI Companies Disclose AI Risk Under Item 105
The Risk Factors section of a 10-K is governed by a specific SEC rule. Here is what Item 105 requires, and how AI companies use it to disclose the legal, regulatory, and reputational risks of the technology they sell.
Artificial Intelligence
Form 8-K Item 2.02: Why an AI Company's Earnings Release Is 'Furnished,' Not 'Filed'
When a company reports quarterly results, the press release reaches the SEC through Form 8-K Item 2.02. Here is what that item requires, and why the distinction between 'furnished' and 'filed' matters to investors.
Artificial Intelligence
What a 'Reportable Segment' Is — and Why NVIDIA's Data Center Revenue Isn't a Separate One
Segment reporting tells investors how a company divides its business for financial reporting. Here is what the term means, where the numbers live, and why NVIDIA reports two segments rather than a standalone 'Data Center' line.
Artificial Intelligence
What 'Capital Expenditures' Mean in an AI Company's 10-K — and Where to Find Them
Capital expenditures are the cash a company spends to acquire or build long-lived physical assets like data centers and servers. Here is where the figure appears in SEC filings and what the rules require management to say about it.
Artificial Intelligence
Read Microsoft's New Filings and You See a Company Optimizing the Cost of Running AI, Not Just Building It
A cluster of newly published Microsoft applications is dominated not by bigger models but by cheaper inference, custom packaging, and generative-AI productivity. That points the R&D budget at the unit economics of serving AI at scale.
Artificial Intelligence
Intel's New Applications Point at Cheaper, Provable AI Inference
A week of published applications clusters on compressing transformer caches, proving an inference ran correctly, and routing generative-AI jobs across enterprise hardware.
Artificial Intelligence
Qualcomm's Week of Grants Lands on the Mechanics of Running AI on the Device
Among 95 issued patents, a thread covers attention-cache eviction, tunable activation functions, and graph neural networks for perception — the techniques that make models run on power-limited hardware.