The AI Term Sheet Is Changing: Why Founders Are Negotiating Different Deals in 2026
Two Deals, Same Week, Opposite Signals
In May 2026, Anthropic raised $65 billion at a $965 billion valuation — nearly triple its mark from three months earlier. In August, OpenAI closed a $7 billion employee tender offer at a valuation held perfectly flat against its March round: $852 billion, unchanged. One company's price nearly tripled in a single quarter. The other's price froze on purpose. Both moves are being read as signs of strength. That contradiction is the real story in AI venture financing right now, and it's the wrong place to start if the question is who actually holds the leverage.
The headline valuations — $965 billion, $852 billion, a combined $1.25 trillion for the merged xAI-SpaceX entity — are the least useful numbers in any of these deals. The more revealing numbers sit in the mechanics underneath: what happens to a 1x liquidation preference if an IPO prices below the last private mark, who gets first call on the next round's allocation, and what a startup actually owes a hyperscaler once it signs a $30 billion compute commitment. This piece decodes those mechanics across seven of 2026's defining rounds, and answers the question the headline numbers can't: are founders actually gaining negotiating power in 2026, or are investors simply paying more while quietly demanding more in return?
The Deals That Set the Market
Seven transactions from the past nine months define the current market, and what they signal matters more than their size:
- OpenAI's $122 billion Series (March 2026), $852 billion post-money. Anchored by Amazon (~$50 billion), Nvidia ($30 billion), and SoftBank ($30 billion), with Microsoft, Sequoia, Thrive, and BlackRock rounding out the syndicate. Up from $500 billion just five months earlier — a valuation that moved faster than the company's product cycle.
- OpenAI's $7 billion employee tender offer (August 2026), flat at $852 billion. The company bought the shares itself, using its own balance sheet rather than bringing in outside investors — the first liquidity event in OpenAI's history that didn't reprice the company upward.
- Anthropic's $30 billion Series G (February 2026), $380 billion post-money. Led by GIC and Coatue, with disclosed run-rate revenue of roughly $14 billion — a valuation multiple investors could still tie to a real, fast-growing revenue line.
- Anthropic's tender offer (April 2026), $350 billion pre-money. Fell short of its $5–6 billion target because employees held their shares rather than sell — the opposite signal from OpenAI's oversubscribed prior tenders, where roughly $4 billion in investor demand went unmet because so many employees wanted out.
- Anthropic's $65 billion Series H (May 2026), $965 billion post-money. Led by Altimeter, Dragoneer, Greenoaks, and Sequoia, with $15 billion of the total already committed capital, including $5 billion from Amazon. Nearly triple the February mark in three months, on disclosed run-rate revenue that had grown to roughly $47 billion.
- The Microsoft-Nvidia-Anthropic strategic agreement (November 2025). Nvidia committed up to $10 billion and Microsoft up to $5 billion, structurally tied to Anthropic committing to purchase $30 billion of Azure compute capacity, with an option to scale to a full gigawatt.
- The xAI-SpaceX merger (February 2026), $1.25 trillion combined valuation. The largest merger in history, executed specifically to position the combined entity for a blockbuster SpaceX IPO rather than a standalone xAI listing — a structural workaround to the AI valuation question entirely.
What these seven deals signal together is more important than any single number: valuation timelines have compressed from years to months, tender offers have become as strategically significant as primary rounds, and the two companies at the center of the AI financing market are sending opposite signals about their own confidence — Anthropic accelerating its price, OpenAI deliberately freezing it.
Decoding the Term Sheet: Where the Real Negotiation Happens
On the term that gets the most attention — liquidation preference — founders are genuinely winning. Cooley LLP's own deal data shows 98% of venture rounds in Q2 2025 carried a standard 1x non-participating liquidation preference, meaning investors get their money back first and nothing more before common shareholders participate in the rest. That's the founder-friendly baseline, and AI mega-rounds haven't deviated from it; with dozens of investors competing for allocation in every Anthropic and OpenAI round, no single backer has leverage to demand the double-dip economics of participating preferred stock.
But that same Cooley data contains the term that actually matters more: protective provisions — investor veto rights over major corporate decisions — remained in place in over 90% of deals, regardless of how founder-friendly the headline economics looked. This is where the real negotiation has moved. Liquidation preferences protect an investor's money on the way out; protective provisions and pro-rata rights protect their position on the way through — and in a market where valuations are tripling in a single quarter, the right to maintain your ownership stake through the next round has become worth more than any exit-economics clause.
Pro-rata rights specifically have become the single most consequential term in AI mega-rounds, precisely because of the compression described above. An investor who backed Anthropic's $380 billion February round without ironclad pro-rata rights had no guaranteed path to participate at the same relative allocation in the $965 billion May round three months later — missing out on continued exposure to a company whose price had nearly tripled. In a market moving this fast, the option to keep buying at your existing ownership percentage, round after round, has become more economically significant than almost any other clause in the document.
Anti-dilution protection, historically a fought-over term, has become genuinely complicated by the structure of strategic investment itself. Broad-based weighted-average anti-dilution remains the market standard on paper, but when a startup's own supplier — Nvidia selling compute, Microsoft selling cloud capacity — is also its investor, the "price" of the next round is no longer a clean, arm's-length signal the way anti-dilution math assumes. A hyperscaler that both prices a round and sells the compute that round is meant to fund has influence over the company's economics that no anti-dilution formula was designed to capture.
Strategic Investors: Control Without a Board Seat
The Microsoft-Nvidia-Anthropic agreement is the clearest template for how strategic capital now works in AI: Nvidia committed up to $10 billion and Microsoft up to $5 billion, structurally bundled with Anthropic's commitment to purchase $30 billion of Azure compute capacity, with an option to scale to a full gigawatt. This is not a passive check. It is capital, compute, and distribution sold as a single package — and it gives the strategic investor something more durable than a board seat: a multi-year, dollar-denominated commercial obligation the startup cannot easily unwind, regardless of how its board is composed.
Google's move in April 2026 followed the same logic at an even larger scale — a commitment of up to $40 billion to Anthropic, with $10 billion funded immediately, dwarfing its prior roughly $2–3 billion in cumulative investment and tied to Anthropic's adoption of Google's custom TPUs for training its next model generation. Amazon, meanwhile, remains Anthropic's primary training partner through Project Rainier, an $11 billion AI data center built specifically to run Anthropic workloads on Amazon's custom Trainium2 chips, with $5 billion of Anthropic's $65 billion Series H arriving as previously committed capital. Three hyperscalers, three separate multi-billion-dollar compute relationships, all layered onto a single company's balance sheet.
Nvidia's own posture is the most telling data point in this section. After finalizing its two largest AI lab investments ever — $30 billion into OpenAI and $10 billion into Anthropic — the company reportedly signaled it was done making mega-investments of that size directly into frontier labs, even as its broader dealmaking (cloud infrastructure bets like CoreWeave and Nebius, plus an 80-company portfolio run by its separate NVentures unit) continued unabated. In August, Nvidia went further, signing memoranda of understanding to mobilize over $500 billion in third-party capital for AI compute infrastructure — shifting from writing direct equity checks into frontier labs toward structuring the financing that lets others buy the GPUs, such as the Apollo-backed $3.5 billion facility financing Valor Equity Partners' $5.4 billion compute lease to an xAI subsidiary. Nvidia is not stepping back from AI financing; it is stepping back from the specific structure — direct frontier-lab equity — that ties its own balance sheet most tightly to a handful of companies' fortunes.
Does this create advantages or constraints for startups? Both, and the two are inseparable. The advantage is real: guaranteed compute access at a moment when compute is the binding constraint on every AI company's growth, credibility that helps close enterprise deals, and distribution through Azure, AWS, and Google Cloud's existing sales channels. The constraint is that a $30 billion compute commitment functions economically like a fixed liability — it doesn't shrink if growth decelerates, the way a discretionary marketing budget might. Anthropic's decision to spread commercial dependency across Amazon, Google, Microsoft, and Nvidia simultaneously, rather than concentrating it with one partner, is itself a signal that even the industry's best-capitalized company is actively managing this constraint rather than treating the capital as free money.
Founder and Employee Economics: Why Liquidity Now Beats an IPO Later
Tender offers have become the dominant liquidity mechanism in AI, and the reasoning is straightforward: employee compensation at these companies is overwhelmingly denominated in illiquid private stock, and the AI talent war makes retention worth paying for directly. OpenAI has now run three liquidity events in under two years — a $1.5 billion tender in 2024, a $6.6 billion secondary in October 2025 at a $500 billion valuation, and the $7 billion buyback in August 2026 at a flat $852 billion. A staff engineer who joined OpenAI in 2019 on a market-rate package can now be sitting on $50 million to $200 million in equity value, with an actual mechanism to convert some of it to cash without waiting for a public listing that may still be a year or more away.
The more interesting signal is in how the two leading labs' tenders diverged. OpenAI's prior tenders were oversubscribed on the sell side — so many employees wanted to cash out that roughly $4 billion in investor demand went unfilled. Anthropic's April 2026 tender ran the opposite direction: it fell short of its $5–6 billion target because employees chose to hold, betting the company's value would keep climbing into a prospective IPO, while outside investors wanted to buy more shares than employees were willing to part with. Read literally, that means Anthropic's own workforce had more conviction in near-term appreciation than OpenAI's did — a real-time employee sentiment indicator that the headline valuations alone can't provide.
OpenAI's decision to hold its August tender flat at $852 billion, rather than stepping up the price the way every prior liquidity event had, is itself a deliberate signal. Using its own cash rather than outside investor capital meant OpenAI didn't need a new external price-setting event at all — it could give employees liquidity while avoiding the two-sided risk of a fresh valuation mark: overpaying if sentiment has softened, or under-pricing if it hasn't. Freezing the number is a controlled, pre-IPO move, and it explains why founders and employees increasingly prefer a tender over rushing a public listing — it delivers real cash without forcing the company to defend a number to public markets before it's ready.
Founder-Friendly or Investor-Friendly? The Honest Answer Is Both, at Different Layers
On headline economics, 2026 is the most founder-friendly financing environment in venture history. Valuations are compressing years of appreciation into months — Anthropic's near-tripling in a single quarter has no real precedent even at the 2021 peak. Liquidation preferences have genuinely normalized to the founder-friendly 1x non-participating standard. And the sheer number of bidders per round — sovereign wealth funds, crossover hedge funds, hyperscalers, and traditional VCs all competing for allocation in the same deal — gives founders real pricing power that founders in a normal, two-term-sheet negotiation simply don't have.
But investors have not given up leverage — they've relocated it. Protective provisions persist in over 90% of deals regardless of price. Pro-rata rights have become the mechanism by which early investors compound their stake through every subsequent mega-round without renegotiating terms. And strategic investors have built something more durable than a board seat: multi-billion-dollar commercial dependency that constrains a startup's behavior as effectively as any veto right, without ever showing up as a governance line item. This is a meaningfully different playbook than traditional venture financing, where control ran primarily through board composition and negotiated protective provisions attached to a clean primary round. In 2026's AI deals, control increasingly runs through the compute contract sitting alongside the cap table.
The downside scenario is where these two layers reconnect. The bull case for current valuations rests on real, fast-compounding revenue — Anthropic's run rate went from roughly $1 billion to $47 billion in about fifteen months, a trajectory with no analog in the dot-com era's revenue-free unicorns. But the aggregate math is still uncomfortable: the five largest hyperscalers are budgeting somewhere between $700 billion and $755 billion in AI-related capital expenditure for 2026 against total AI revenue estimated at well under $100 billion industry-wide, and AI-exposed mega-caps now represent roughly 40–45% of S&P 500 market capitalization — a concentration level that exceeds the dot-com peak. Nvidia's own decision to stop writing further mega-checks directly into frontier labs, even while dramatically expanding third-party financing facilities, reads as the most exposed strategic investor in the ecosystem quietly hedging its own concentration risk.
The CODEW Take
Who has the leverage: it's genuinely split by layer, not by side. Founders have real, unprecedented leverage over price — valuation, round size, and the survival of founder-friendly 1x non-participating economics all reflect a market desperate for allocation in a handful of companies. Investors have quietly regained leverage over protection and control, just through different mechanisms than the last cycle: pro-rata rights that let them compound through every re-up, protective provisions that persist regardless of headline terms, and — for strategic investors specifically — commercial lock-in through multi-billion-dollar compute commitments that function as de facto control without ever appearing as a board seat.
Which provisions matter most right now: pro-rata rights and compute-purchase commitments, not liquidation preference multiples. The exit-economics fight that defined prior cycles has been won by founders; the fight over who gets to keep compounding their position, and who controls a startup's supply chain, is where 2026's real negotiation is happening.
Is it sustainable: more sustainable than the dot-com comparison suggests, because the revenue behind the top two labs is real and compounding fast — but far less sustainable than the headline valuations alone imply, given that industry-wide AI infrastructure spending outpaces industry-wide AI revenue by a factor of roughly seven to one, and the private financing system's stability rests on a small number of companies continuing to grow at a rate almost nothing in corporate history has sustained for long.
What to watch in the next 12 months: whether OpenAI and Anthropic's IPOs price at or above their last private marks — a below-mark listing would instantly reprice every tender offer and secondary transaction the entire ecosystem has used as a benchmark. Whether tender-offer demand flips at either company, since oversubscription versus undersubscription has become a real-time proxy for internal confidence. Whether other strategic investors follow Nvidia's lead in stepping back from direct frontier-lab mega-checks while expanding third-party financing instead. And most concretely: whether any AI lab's growth deceleration forces a renegotiation of its hyperscaler compute commitments — the first crack in the strategic-investor lock-in structure would show up there before it shows up in any headline valuation.
The Term Sheet is The CODEW's flagship deal intelligence series. The Term Sheet summarizes the most important technology business news, including mergers and acquisitions, startup funding, executive moves, partnerships, earnings, product launches, regulatory actions, and market developments. For corrections, tips, or partnership inquiries, contact the editorial team.
Reviewed by Erwin Castro
on
Saturday, August 15, 2026
Rating: