When Stripe raised its Series D in late 2016, leading to a $9.2 billion valuation, the term sheet negotiations reportedly took place over six weeks. General Counsel Billy Alvarado later noted that the pro-rata rights provisions were among the most contested clauses — investors wanted guaranteed rights to maintain ownership percentages in future rounds, while Stripe wanted flexibility to bring in strategic partners. The final structure gave existing investors modified pro-rata rights capped at their then-current ownership percentage, a compromise that became a template cited in later Silicon Valley deals.
A term sheet is a non-binding letter of intent that outlines the key economic and governance terms of an investment. For AI ventures, certain standard provisions take on extra significance, and entirely new clauses have emerged to address AI-specific risks.
Pre-money valuation determines what percentage of the company the investor receives. In AI deals, this figure increasingly reflects the value of proprietary training data and model weights, not just revenue multiples. Investors in Cohere's $270 million Series C in 2023 were, in part, pricing the value of Cohere's enterprise-tuned large language models — assets that don't appear on a balance sheet.
Liquidation preferences specify who gets paid first in an exit. Most Series A and later rounds use 1x non-participating preferred — the investor recovers their investment before common shareholders receive anything, but does not "double dip" into remaining proceeds. Participating preferred, which was common in the 2001–2005 era, has largely disappeared from competitive deals; its reappearance in a term sheet is a signal of weak founder leverage.
Several VCs, including Andreessen Horowitz and Sequoia, have begun including model governance provisions in term sheets for generative AI companies — requiring board notification before deploying models above a certain parameter count or before fine-tuning on regulated data categories (healthcare, financial). These provisions emerged after the FTC's 2023 inquiry into OpenAI.
Valuation cap (convertible notes/SAFEs): For pre-seed and seed AI companies, the most common instrument is the Y Combinator SAFE. The valuation cap establishes the maximum price at which the note converts to equity. In 2023–2024, AI seed SAFEs with caps of $20–$30M were standard for pre-revenue generative AI companies with a working demo.
Discount rate: Typically 15–20% on SAFEs and convertible notes — the investor converts at a price 15–20% below the next round's price per share, compensating for the risk of investing earlier.
Anti-dilution protection: Weighted-average anti-dilution (broad-based) is the market standard. Full ratchet anti-dilution — where an investor's price is reset to whatever a future down round prices at — is punitive for founders and was specifically cited by Klarna executives as a source of pain in the company's 2022 down round, which repriced the company from $45.6B to $6.7B.
Board composition: A typical Series A board is two founders, one lead investor, and one independent director. Losing board majority is a critical inflection point — OpenAI's November 2023 board crisis, in which CEO Sam Altman was briefly fired by a board on which he held no seat, illustrated the stakes of governance structure in AI companies operating under a non-profit/capped-profit hybrid.
Protective provisions: Standard preferred stock protective provisions give investors veto rights over major actions: selling the company, issuing new equity, taking on debt above a threshold, or changing the company's charter. AI-specific addenda have begun to include veto rights over major model licensing deals, GPU infrastructure commitments above defined thresholds, and changes to safety governance frameworks.
Information rights: Major investors (typically $500K+) receive monthly financials, quarterly board letters, and annual audited financials. In competitive deals, founders have successfully pushed these to quarterly unaudited financials only — a significant reduction in administrative burden.
The AI investment surge of 2023–2024 shifted leverage significantly toward founders at the seed and Series A stages. Benchmark, Sequoia, and General Catalyst were all reported to have competed for deals without standard pro-rata rights or with reduced protective provisions. This leverage is concentrated: it applies to companies with demonstrated model differentiation or exceptional founding teams. For most AI startups, standard terms remain the baseline.
You've received a Series A term sheet for your AI infrastructure company. Before signing, you need to identify which provisions are market-standard, which are investor-favorable but acceptable, and which are genuine red flags worth pushing back on.
When Mistral AI closed its €105 million ($113M) seed round in June 2023 — the largest European AI seed round ever — the due diligence process was compressed to approximately three weeks, unusually fast for a deal of that size. Investors including Andreessen Horowitz, Lightspeed Venture Partners, and others had to assess a company with no deployed product, no revenue, and a four-month-old founding team. The diligence focused almost entirely on technical capability: the team's prior work at DeepMind and Meta FAIR, the architecture decisions underlying the planned Mistral 7B model, and the founding team's ability to train competitive models at significantly lower compute cost than incumbents. The deal closed on technical diligence rather than commercial diligence — a pattern increasingly common in foundation model investing.
Due diligence (DD) begins after a term sheet is signed and runs in parallel with legal documentation. The standard process takes four to eight weeks for a Series A and two to four months for a Series B or later. During this period, the investor's team reviews every material aspect of the business. A failed DD is rare but costly — the investor withdraws, the deal collapses, and the process itself has cost the founder team significant time and sometimes revealed internal problems to a wider audience than intended.
Most investors segment DD into four workstreams: business/commercial (market size, customer relationships, revenue quality), technical (codebase, model architecture, infrastructure, IP), legal (corporate structure, cap table, contracts, IP ownership, litigation), and financial (historical financials, burn rate, unit economics, projections).
Investors with deep AI expertise — including Coatue, Thrive Capital, and Spark Capital — have developed specific diligence checklist items that didn't exist five years ago: training data provenance (can the company demonstrate rights to training data or a defensible fair use position?), model evaluation benchmarks (how does the model perform on independent evals vs. self-reported metrics?), compute dependency risk (concentration of GPU access with a single provider, particularly AWS or Azure), and regulatory exposure mapping (how will the EU AI Act, potential US AI regulation, or sector-specific rules affect the model's deployment?).
IP ownership and model rights: Investors will scrutinize employment agreements, contractor agreements, and any open-source model licenses the company has built upon. Building on Meta's LLaMA with a commercial license, or on models licensed under CC BY-NC, can create material IP risks that show up in legal DD. Several AI startups have had deals slow or collapse when investors discovered that key engineers had contributed to foundation models during prior employment without clear IP assignment agreements.
Training data lineage: The New York Times v. OpenAI lawsuit, filed in December 2023, placed training data rights at the center of AI company risk profiles. Investors now routinely ask for a data lineage memo — a document describing every training data source, the legal basis for its use, and any pending or anticipated claims. Companies that have built proprietary data flywheels (Veeva Systems in life sciences, Palantir in defense analytics) command higher valuations precisely because their training data is unambiguously owned.
Model evaluation independence: Self-reported benchmarks are increasingly discounted. Investors like Coatue have begun requiring third-party model evaluations as part of DD — running the company's models against standardized benchmarks (MMLU, HumanEval, HellaSwag) and against competitor models in blinded tests. This emerged partly in response to the "benchmark overfitting" problem, where models were fine-tuned specifically to score well on public evaluation datasets without underlying capability improvement.
The founders who close fastest are those who treat DD preparation as ongoing, not reactive. Virtual data rooms (most commonly Datasite, Intralinks, or Notion for earlier-stage companies) should be organized and populated before the first investor meeting, not after a term sheet arrives.
Key documents to have current at all times: cap table (with a full history of all equity grants, SAFE conversions, and option pool allocations), IP assignment agreements for all employees and contractors, customer contracts (redacted for confidentiality where needed), any pending litigation or regulatory correspondence, and audited or reviewed financials if available.
For AI companies, an additional preparation step is a technical memo — a 5–10 page document describing model architecture, training methodology, evaluation results, known limitations, and plans for safety and alignment. Anthropic's decision to publish model cards for Claude was partly a product communication move but also served to pre-answer the technical DD questions that investors and enterprise customers routinely ask.
A compressed due diligence timeline is not always a gift. When investors rush to close — as happened with several crypto-adjacent AI deals in 2021–2022 — it often signals competitive pressure rather than genuine conviction. Some of those deals subsequently required bridge financing when business fundamentals didn't hold up. For founders, a thorough, well-organized DD process is preferable to a fast, superficial one: it builds investor confidence, surfaces genuine issues before they become post-investment surprises, and sets the foundation for a functional board relationship.
A Series A investor has just sent a standard DD request list. You have two weeks to assemble your data room. Your AI coach will help you prioritize, flag gaps in your current documentation, and prepare for the AI-specific technical diligence questions that sophisticated investors now ask.
When Amazon completed its $4 billion investment in Anthropic in March 2024 (the second tranche of a $7.3B total commitment announced in September 2023), the closing mechanics were unusually complex. The deal was structured as a strategic investment — not a standard VC round — requiring separate legal documentation governing cloud computing commitments (Anthropic agreed to use AWS as its primary cloud provider), model access rights, and a seat on Anthropic's board for Amazon's representative. The documentation reportedly involved over 40 lawyers across four firms and took approximately five months from LOI to final close. The deal illustrated how closing complexity scales with strategic entanglement — a lesson for AI founders considering corporate strategic investors alongside or instead of financial VCs.
A standard VC round generates a stack of legal documents, each serving a specific function. Understanding what each document does — and which provisions within each document deserve founder attention — is essential for navigating the closing process without simply deferring to counsel on everything.
Stock Purchase Agreement (SPA): The primary transaction document. It records the price per share, the number of shares being sold, representations and warranties the company makes to investors (about its legal standing, IP ownership, financial condition, cap table accuracy), and conditions to closing. The reps and warranties section is where material omissions from DD can create post-closing liability — if the company reps that it has no pending litigation and there is actually a cease-and-desist letter sitting in the founder's email, this creates a breach with potential for indemnification claims.
Investor Rights Agreement (IRA): Governs ongoing investor rights including registration rights (the right to have shares registered for public offering), information rights, and pro-rata rights. The IRA also typically contains a "drag-along" provision, allowing a majority of shareholders to compel minority shareholders to vote in favor of a sale — critical for preventing a small investor from blocking an exit.
Right of First Refusal and Co-Sale Agreement: Restricts founders' ability to sell shares without first offering them to the company and investors. Co-sale rights allow investors to participate alongside founders in any approved secondary sale.
Voting Agreement: Establishes board composition requirements and often includes protective provisions specifying what actions require preferred stockholder approval.
The most legally dangerous representations for AI companies in SPAs involve IP ownership and data practices. If a company represents that it has full rights to all training data used in its models, and this is later challenged (as in the Getty Images v. Stability AI case), investors may have grounds for breach of representations claims. Experienced AI company counsel increasingly recommend qualified representations — "to the company's knowledge" carve-outs and disclosures — rather than categorical IP ownership statements for training data acquired before the current copyright litigation environment clarified.
Draft circulation (Weeks 1–2): The lead investor's counsel circulates first drafts of all closing documents. The company's counsel reviews and marks up. The standard is that the lead investor's counsel drafts (investor-favorable starting position) and company counsel negotiates back toward market. For AI companies, paying for experienced venture counsel — not a generalist firm — is not optional. Firms like Cooley, Wilson Sonsini, Gunderson Dettmer, and Fenwick have AI-specific practice groups and know which provisions investors will and won't move on.
Negotiation and finalization (Weeks 3–5): Markup exchanges, calls to resolve disputed provisions. The most commonly negotiated items at this stage are: rep and warranty scope (particularly IP reps), indemnification obligations, the definition of "material adverse change" (a condition that allows investors to withdraw if something bad happens before close), and the scope of investor consent rights.
Condition satisfaction (Weeks 5–6): Conditions to closing — items that must be true for the deal to proceed — are resolved. Common conditions: completion of a satisfactory due diligence review, execution of key employee proprietary information agreements, no material adverse change in the business, and receipt of any required regulatory approvals.
Wire and closing (Day of close): Documents are signed (typically via DocuSign in sequence), the company delivers a closing certificate confirming all conditions are met, and investors wire funds. The company then files an amendment to its certificate of incorporation (the charter) with Delaware (if incorporated there) to authorize and create the new series of preferred stock.
Cap table errors: The single most common cause of closing delays is a cap table that doesn't reconcile — option grants that were never formally approved by the board, SAFEs that were signed but never logged, or shares issued to contractors without proper documentation. Maintaining a live, board-approved cap table in Carta (or similar) from day one is non-negotiable.
Missing IP assignments: A contractor who wrote core model code and never signed an IP assignment agreement can hold up a closing for weeks while counsel tracks them down. Some have demanded payment to execute retroactive assignments. IP assignment hygiene — every employee and contractor signs a PIIA (Proprietary Information and Inventions Agreement) on day one — is one of the simplest legal risks to eliminate in advance.
Open-source license issues: Using code under AGPL (Affero GPL) or GPL licenses in a commercial SaaS product can create copyleft obligations that force disclosure of proprietary model code. Investors and their counsel have become increasingly sophisticated about open-source license stacks, and finding an undisclosed GPL dependency mid-closing has derailed deals.
As illustrated by the Amazon-Anthropic deal, strategic corporate investors (Microsoft, Google, Salesforce Ventures, Amazon) add significant closing complexity relative to financial VCs. Their legal teams are larger, their approval processes involve more internal stakeholders, their documentation includes commercial agreements alongside equity terms, and their antitrust review obligations may be triggered at lower thresholds. Founders should budget 50–100% more time for strategic investor closings and ensure their counsel has experience with corporate venture specifically.
You're two weeks from close on your Series A. Your counsel has flagged three issues that could delay or derail the closing: a contractor who hasn't signed a PIIA, an AGPL dependency in your inference stack, and an overly broad IP representation in the SPA that your training data sourcing may not support.
Between Figma's Series D ($50M at $2B valuation, June 2020) and its announced acquisition by Adobe ($20B, September 2022), CEO Dylan Field ran a notably disciplined investor relations operation. Figma's board updates were known in Silicon Valley VC circles for their candor — Field reportedly included a "what's not working" section in every board deck, covering genuine product and competitive challenges alongside wins. When Sequoia's Matt Miller told The Information in 2022 that Figma was "the most capital-efficient large software company we've ever backed," part of that assessment reflected information quality. Investors who received candid, organized updates were more confident in the company's execution and more willing to support the company in competitive situations. The Adobe acquisition came together in 22 days from first conversation to signed term sheet — a speed made possible in part by investors and board members who already had complete, accurate information about the company's state.
The investor update is the primary mechanism through which founders maintain investor trust and build the social capital needed for future rounds, strategic introductions, and crisis support. Most investor rights agreements require monthly or quarterly financial reports, but the highest-performing founders go significantly beyond the legal minimum.
The standard cadence for a Series A company is a monthly investor update email (for all investors with information rights), a quarterly board meeting (typically 2–3 hours), and an annual strategy session (half-day, sometimes off-site). The monthly email is the workhorse: it keeps investors informed between board meetings and is the primary source of the investor's ongoing thesis on the company.
For AI companies, the monthly update should include standard metrics (MRR/ARR, burn, runway, headcount) plus AI-specific operational metrics: model performance benchmarks, compute costs, API uptime, and any material changes in the regulatory or competitive environment. Several AI founders have added a "model development" section to their monthly updates, describing training runs, evals, and architecture decisions in enough detail for sophisticated investors to track technical progress.
The best board decks follow a consistent structure: (1) KPI dashboard — 6–10 key metrics with trend lines; (2) wins and momentum — what's working; (3) challenges and asks — what's not working, and specific requests for board help; (4) strategy update — any changes to roadmap, competitive positioning, or market thesis; (5) financials — P&L, burn rate, runway, updated forecast. The "challenges and asks" section is where founders distinguish themselves — most present only wins. Boards that only see wins cannot provide genuine help when real problems arise.
AI ventures face a specific investor expectation challenge: the technology is moving so fast that investors often expect faster progress than is operationally realistic. The antidote is precise milestone communication. Rather than accepting an investor's framing ("when will you have GPT-4 level performance?"), founders should establish their own milestone framework in the first post-close board meeting and communicate progress against it consistently.
Stability AI's trajectory illustrates the cost of mismanaged expectations. After its $101 million funding round in October 2022 (led by Coatue and Lightspeed), CEO Emad Mostaque made numerous public statements about Stability's technical roadmap that outpaced the company's execution. The subsequent board conflict, Mostaque's resignation in March 2024, and the company's financial difficulties were partly rooted in an expectations gap that had been building since the initial close. Investors who believed they'd funded a specific vision found a different reality in board meetings.
The counter-example is Hugging Face, which raised its Series D ($235M, May 2023, at $4.5B valuation) with a board that had been consistently informed about the company's evolving strategy — including its pivot from primarily a model-hosting platform to an enterprise AI infrastructure company. Investors including Salesforce Ventures, Google, and Nvidia participated in the Series D precisely because they had been receiving transparent updates about this strategic evolution for over a year before the round.
The most valuable post-close investor relations work is building the track record and relationships that will make the next round easier. This means actively requesting and using investor introductions (a warm introduction from a Sequoia or Benchmark partner to a target strategic partner or Series B lead is worth dozens of cold outreach attempts), and building social proof through investor-visible milestones.
The reference check runs in reverse. When you approach Series B investors, their first call will be to your Series A lead. The quality of that reference — not just "they hit their numbers" but "they communicate with total clarity, they know their weaknesses, they're building a team that can scale" — is often determinative for competitive Series B processes.
Signal your next round early. Founders who discuss Series B milestones in their Series A board meetings — "here's what we need to achieve to make a Series B compelling at $100M+ valuation" — give investors the ability to actively monitor progress and advocate for the company with their later-stage colleagues. Several Benchmark investments at Series A have converted to Benchmark leading the Series B specifically because the founder had been explicit about the Series B thesis from day one of the Series A relationship.
Days 1–30: Send a "first 90 days" operational plan to all investors. Hold one-on-one calls with each board member to align on priorities and establish communication norms. Set up your Carta cap table, reporting templates, and data room updates. Days 31–60: Deliver your first formal board update using the structure above. Make one specific, concrete ask of each investor — an introduction, a customer referral, a regulatory contact. Days 61–100: Assess which investors are providing genuine value (introductions, advice, time) vs. which are passive. This assessment shapes how much access and information depth you provide going forward — a key relationship management decision founders rarely make explicitly.
It's Day 35 post-close on your Series A. You need to send your first formal investor update. The update needs to establish the communication cadence and quality that will define your investor relationships for the next 18–24 months.