Summary of 20VC: NVIDIA Crushes Quarter and Buys Hugging Face | OpenAI Cuts Off Cursor | Instinct Hits $2.5BN Valuation and The Race for AI Assistants | Cognition Raises at $46BN, Linear $2.5BN and Clay $7BN

Overview of 20VC with Harry Stebbings

This episode is a high-energy AI market roundtable focused on what the latest wave of model, infra, and agent news means for founders, investors, and enterprise software. The big themes were NVIDIA’s blowout quarter, the emerging race for AI assistants and agentic workflows, the OpenAI/Cursor fallout, and why the hosts believe we’ve entered an era of “compound startups” that must ship far more software, far faster, to stay relevant.

NVIDIA: Record Quarter, Supply Constraints, and the Compute Boom

What happened

• NVIDIA reported a record $96.2B quarter and guided to ~70% revenue growth for the next fiscal year, well above Street expectations.
• The hosts framed this as a signal that compute demand remains extremely strong and likely stays that way for at least another 12 months.
• The key takeaway: NVIDIA appears supply constrained, which reduces near-term miss risk but also suggests the entire AI infrastructure stack is still in expansion mode.

Why it matters

• If NVIDIA is still selling everything it can make, then hyperscalers, model providers, and application companies are likely still in a “green light” environment.
• The discussion emphasized that the main risk is no longer circular financing chatter alone, but whether end-user demand for AI keeps growing fast enough to justify the spend.
• One memorable takeaway: the hosts argued that the pressure on free cash flow at Google and Oracle is the other side of NVIDIA’s strength.

NVIDIA’s Hugging Face Deal: Open Weights and Market Power

The thesis behind the move

• The reported $12.9B move toward Hugging Face was interpreted as NVIDIA trying to own more of the open-weights stack and ensure that as AI usage grows, more of that value still flows through NVIDIA’s compute.
• The logic: if end users spend a trillion dollars on tokens and inference, NVIDIA prefers that flow through lower-margin, open tooling where more compute is consumed.

Strategic implication

• The hosts viewed this as NVIDIA trying to win every part of the market, not just sell chips.
• It was framed as rational, even obvious, behavior for a company that wants to defend its dominance as AI becomes more software-driven and more competitive.

OpenAI Cuts Off Cursor: Competition, Grudges, and IP Protection

What happened

• OpenAI reportedly cut off Cursor, which prompted commentary about how much of the decision was real vs. theatrical.
• Cursor’s response emphasized that the loss represented only a small portion of traffic, and the hosts noted users can still bring their own keys in some cases.

The deeper read

• The hosts argued that this was not just pettiness: it was a rational response to competition and model distillation risk.
• OpenAI has a strong reason to avoid subsidizing a direct competitor that could learn from its models while also competing for customers.
• The episode repeatedly returned to the idea that when two companies are competitors, partnership becomes unstable very quickly, especially when there is existing personal conflict.

AI Cybersecurity: The Real Wake-Up Call

Why the Hugging Face/OpenAI incident mattered

• A major segment focused on the OpenAI research/blog about agents swarming to find weaknesses.
• The hosts strongly pushed back on the anthropomorphic language around “agent collaboration” and “civilizations”:
  • These are not human-like societies.
  • They are persistent optimization systems running with enough compute and enough time to find flaws.

Core takeaway

• This is a cybersecurity warning, not sci-fi theater.
• Persistent agents can:
  • probe systems constantly,
  • string together multiple weaknesses,
  • remain undetected for long periods,
  • and scale attack attempts to near-zero marginal cost.
• The message to CISOs and enterprise teams: you have months, not years, to harden systems against AI-assisted attacks.

Instinct and the Race for AI Assistants

What Instinct represents

• Instinct was positioned as a leading example of the new consumer AI assistant category:
  • access to calendar,
  • email,
  • credit cards,
  • and other personal data,
  • so the assistant can “manage your life.”
• The hosts noted its $2.5B valuation and said the category is attracting enormous venture excitement.

The key debate: usefulness vs. reward hacking

• The conversation focused on the challenge of reward hacking and whether consumer assistants can safely make decisions on your behalf.
• Main concern:
  • if an assistant is told to optimize for your happiness or convenience,
  • it may take actions you did not explicitly want,
  • especially when multiple objectives conflict.

Bottom line

• The hosts believe this category is real and useful, but still fraught with safety and control issues.
• Likely near-term use cases:
  • scheduling,
  • booking,
  • inbox triage,
  • task management,
  • and low-risk personal admin.
• But they remain skeptical about handing over high-trust actions like payments without strict caps and controls.

“Compound Startups”: The New Operating Model

The central idea

• One of the biggest themes was that AI has changed how companies build.
• The hosts argued that startups must now become compound startups:
  • shipping more products,
  • more features,
  • more integrations,
  • and often adjacent products too.

Why this matters

• AI makes code cheaper to produce, but not all software spend falls:
  • winners can add more product surface area,
  • competitors can move faster,
  • and the best companies compound into adjacent workflows.
• This creates a “build more, faster” environment where:
  • point solutions get squeezed,
  • multi-product platforms pull ahead,
  • and slower teams risk becoming irrelevant quickly.

Venture implication

• VC capital is increasingly used to accelerate compounding, not just fund a single product.
• The hosts explicitly connected this to a “stuff capital into winners and stay out of the way” mindset.

Cloud and Enterprise Software: Salesforce, Anthropic, and Outcome-Based Pricing

Salesforce’s evolution

• The hosts discussed Salesforce’s partnership with Anthropic and the broader move toward multi-surface AI access:
  • users may interact through Slack, Claude, or other surfaces,
  • while Salesforce remains the underlying system of record.
• They also highlighted the shift toward outcome-based pricing, which they see as a major strategic change.

Why it matters

• The enterprise market is moving away from purely seat-based or tool-based pricing toward measurable outcomes.
• That favors companies that can:
  • deliver results,
  • remain open to different user surfaces,
  • and adapt to agent-led workflows.

Investor read

• Salesforce was characterized as a compounder, not a rocket ship:
  • strong cash flow,
  • improving positioning,
  • but not necessarily the kind of explosive upside seen in newer AI-native names.

Other Notable Company and Market Updates

Cognition

• Reportedly raising around $46B valuation, with huge ARR trajectory.
• The hosts used it as evidence that coding and software automation remain an enormous market.
• Their conclusion: the market for AI coding tools is still much larger than many initially thought.

Linear

• Linear’s tender round at $2.5B was seen as potentially undervalued.
• The argument:
  • project management is being redefined for an agent-heavy world,
  • and Linear is well positioned as a system of record for software tasks created by humans and agents.

Clay

• Clay’s $7B valuation was described as increasingly compelling.
• The hosts argued that Clay is becoming a key platform for agentic GTM:
  • AI agents running prospecting, enrichment, and outbound workflows continuously.
• They believe agent-driven go-to-market could expand usage dramatically.

Andreessen Horowitz Growth Fund

• The increased growth fund size was interpreted as a sign that investors believe the opportunity set is still expanding.
• More capital is being deployed because the biggest winners are still getting bigger.

Stripe / PayPal deal

• The rumored deal fell apart, and the hosts framed it as a price-driven negotiation that never bridged.
• It was used as a reminder that late-stage deals often require multiple rounds of back-and-forth before they die or close.

Flock Safety

• The hosts noted growing backlash around surveillance and misuse of police data.
• Their takeaway:
  • the product may be strong,
  • but misuse and public perception can become a real business risk.

Main Takeaways

• NVIDIA’s strength is a signal for the entire AI stack: demand remains intense, and the boom is not over.
• OpenAI and other AI labs will cut off competitors when necessary; partnerships are fragile in a competitive model market.
• Cybersecurity is the urgent near-term AI risk: persistent agents can find and exploit weaknesses at scale.
• AI assistants are real, but trust is the bottleneck: useful for low-risk tasks, dangerous without hard constraints.
• The best startups are becoming compound startups: more features, more surfaces, more speed.
• Agent-friendly software is becoming the new wedge: products that work well with third-party agents may win disproportionately.
• Capital is concentrating into winners: venture is increasingly about backing the companies already showing momentum.

Practical Implications for Founders and Operators

If you’re building in AI

• Ship faster than before.
• Design for agent compatibility.
• Expect adjacent products to be built quickly by competitors.
• Treat security and guardrails as core product features, not afterthoughts.

If you’re investing

• Look for products that are:
  • agent-friendly,
  • outcome-oriented,
  • and capable of expanding into multiple workflows.
• Pay close attention to whether a company is becoming a platform rather than a single feature.

If you’re running an enterprise

• Assume AI-assisted attacks are already here.
• Revisit access controls, permissions, and cybersecurity posture.
• Be prepared for software buying to shift toward outcomes rather than seats or modules.