Investors’ 2026 Predictions: AI, Software Valuations and Employment

Harry Stebbings hosts a year-end discussion with investors Jason Lemkin and Rory O’Driscoll. Recorded in December 2025, the episode reviews founders, funds and products before making predictions about stocks, IPOs and employment in 2026. The forecasts are the guests’ opinions at the time, not verified results or investment advice.

Asked for founder of the year, Lemkin first nominates Anthropic’s Claude models as his products of the year. He credits improvements in coding models with helping tools such as Cursor, Replit and Lovable become more useful, then gives Anthropic CEO Dario Amodei credit for delivering them.

O’Driscoll also selects Amodei for operating execution while disagreeing with some public warnings about unemployment. His judgments about product quality, growth and future profitability are personal assessments rather than independently verified comparative results.

Stebbings nominates SpaceX executive Gwynne Shotwell for leading through a difficult year and predicts a large future IPO. He also points to Scale AI’s Alexandr Wang and the returns associated with Meta’s deal, distinguishing a financial outcome from a longer-term assessment of impact.

Lemkin adds Robinhood co-founder Vlad Tenev, praising the company’s expansion from one product to several. The panel’s revenue and share-price figures are recollections from the discussion, not audited operating results or recommendations to buy the stock.

For fund of the year, O’Driscoll chooses Index Ventures, citing early positions in Wiz, Figma and Revolut. He distinguishes announced or paper valuation events from completed exits: the Wiz transaction had not closed at the time of the episode, and Revolut had not gone public.

Lemkin nominates Neo for early exposure to companies such as Cursor and Cognition. He distinguishes the appeal of the firm’s sourcing and approach from realized fund returns, which the panel does not verify.

Lemkin also praises Creandum’s European investments, while Stebbings raises Hummingbird’s early investment in Lovable. The large fund-return figures they discuss are their own claims; the conversation does not supply fund accounts or establish how much has been realized in cash.

Stebbings selects Benchmark for its early-stage portfolio, listing Manus, Sierra, Fireworks, Cerebras and LangChain. O’Driscoll praises a consistent early-stage strategy and argues that small early-stage funds and large multistage funds can both work if executed well.

For growth investing, Stebbings names Thrive and Founders Fund, citing investments in companies including OpenAI, SpaceX, Stripe and Anduril. The discussion reflects the speakers’ selections and reported positions, not an independently ranked comparison of fund performance.

Lemkin suggests Google as an unconventional corporate-investing nominee because of its interests in AI and other technology companies. The ownership percentages mentioned are estimates whose current definitions and dilution are not established in the conversation.

The guests also praise Menlo Ventures’ Anthropic investment. O’Driscoll emphasizes that venture performance ultimately depends on actual exits and returns, even when an investment looks impressive on paper.

Asked for an individual investor, O’Driscoll selects Elad Gil for his ability to raise and deploy capital as a prominent investor. Stebbings notes that Gil has a team, illustrating the difference between a recognizable individual brand and operating a firm alone.

Stebbings names a Kleiner Perkins investor and cites Windsurf and Chronosphere as examples of liquidity. The automatic captions do not reliably establish the person’s identity, and the reported transaction proceeds are not independently verified. O’Driscoll says he keeps a database of completed exits because cash returned is a key measure of venture performance.

O’Driscoll selects OpenEvidence as a breakout company, describing a search tool for medical research. His account of doctor usage and advertiser value is unverified, and the discussion does not establish the accuracy of medical answers or support using the service as a substitute for clinical judgment.

Lemkin nominates Databricks, arguing that the company benefited from the AI wave after building a data business earlier. His revenue and growth comparisons with Snowflake are estimates offered in conversation, not independently checked financial results.

Stebbings selects ElevenLabs, citing rapid revenue growth and competition from larger companies. Lemkin describes using the product to create event announcements quickly and cheaply; his example is a personal anecdote rather than a controlled comparison of price or quality.

O’Driscoll sees a common pattern in tools that give individuals the ability to complete a task quickly. The guests also discuss Lovable and Replit, but Lemkin says their broader economic impact remains to be demonstrated even though he personally uses the tools.

O’Driscoll names competition for AI talent as a major surprise of 2025. He describes unusually large hiring packages and transactions and argues that companies spending heavily on infrastructure may also spend heavily on people; the specific package figures are not independently verified.

Lemkin says very large private-company valuations changed his view of possible venture outcomes. His suggestions of trillion-dollar IPOs are predictions, not completed transactions, and paper valuation increases are not the same as cash returned to investors.

He reflects on a prior model of entering early and relying on later investors to mark up a company. He now sees potential rewards in both large later-stage positions and early ownership, while acknowledging that the nature of venture investing has changed.

Lemkin describes moving from skepticism about early AI agents to greater confidence after using newer tools. His expectation that agent capabilities will expand is a personal forecast, not proof that any particular business model will work.

Stebbings names the shifting Windsurf transaction and proposed large AI infrastructure investments as surprises. The guests use the phrase “circular deals” to discuss strategic funding relationships; the conversation does not verify private transaction terms or establish that announced investment amounts were fully deployed.

O’Driscoll argues that both the potential size of technology companies and the portion of their growth occurring in private markets have increased. He warns that private marks still need to be tested against liquid market prices and that attractive financial structures can be pushed too far.

Stebbings asks for the best-performing technology stock in 2026. O’Driscoll resists the premise, saying the answer depends on the universe and market-cap cutoff; he cites smaller prior-year winners as a reminder that a top-down prediction is difficult.

Lemkin instead proposes a narrower group of public software and infrastructure companies and predicts some current leaders will remain leaders. He acknowledges that he is not a public-market specialist and may have the historical arithmetic wrong.

He discusses companies including Palantir, Cloudflare, Shopify, CrowdStrike and Snowflake as exposed to different AI-related themes. His forecast that some will continue to benefit is a speculative view at the end of 2025, not an assurance of performance.

The group discusses whether recent reacceleration can persist. O’Driscoll frames continued gains in high-multiple companies as a momentum-dependent bet: even a strong business can be repriced if expectations change.

Lemkin raises Salesforce as a different possibility, arguing that its large customer base may want useful AI agents. The scenario depends on products delivering measurable value and customers paying for them; the panel does not establish adoption or future revenue.

O’Driscoll contrasts the Salesforce idea with betting on expensive momentum stocks. He says a lower starting revenue multiple may offer a different risk profile, while acknowledging that share prices can fall. He discloses owning Salesforce shares and prior involvement with HubSpot.

Lemkin lists several software companies that had performed poorly and asks whether their AI offerings will generate incremental revenue. He says that a company can have a capable feature without proving that customers will pay materially more for it.

O’Driscoll discusses Bill, where he says he served on the board, and potential AI applications in financial workflows. Lemkin identifies himself as a customer. Their views are informed by those relationships, and any proposed product or pricing benefit remains a hypothesis.

Stebbings describes a risk in bundled AI pricing: attributing existing contract value to an AI feature is different from generating incremental revenue. If the customer later drops the add-on, discounting the core product could make renewal economics worse.

Lemkin criticizes broad “AI-influenced” revenue measures, citing Adobe, and says he would prefer evidence of net-new bookings. His interpretation of Notion’s growth and pricing is explicitly an inference; the panel does not have the company’s detailed revenue breakdown.

O’Driscoll agrees that vendors need to deliver and charge for real added value. He notes that including costly AI capabilities without increasing revenue can pressure margins, although the magnitude depends on the company’s actual costs and pricing.

Lemkin predicts demand for Salesforce sales agents if they can automate valuable work reliably and be deployed easily. His example of a $100,000 annual product and assumed customer uptake illustrates a potential use case, not an established product outcome.

The guests say expensive copilots that provide too little added value can disappoint customers. O’Driscoll suggests large organizations need executives who examine actual product value rather than rely only on reassuring internal reports.

Asked to buy one large technology stock and short another for 2026, O’Driscoll chooses Google and NVIDIA respectively in the hypothetical exercise. He explains the choice in terms of AI capital-spending expectations and makes clear that market timing is difficult; it is not a recommendation for viewers.

O’Driscoll discloses that he owns Apple shares and sold some while retaining Google. He sees valuation risk in Apple but treats the stock’s future and the effects of AI as uncertain; the comparison is his view at the time of recording.

Lemkin disagrees with a 2026 short on NVIDIA, arguing that competing chip capacity takes time to build and deploy. He thinks Amazon could underperform in the near term, while emphasizing that his time horizon is a prediction and that he is not a public-market expert.

O’Driscoll distinguishes two possible risks to NVIDIA: competitors taking share and demand for compute slowing. He notes that being directionally right about a short can still lose money if the timing is wrong.

The guests discuss reported AI infrastructure financing and NVIDIA’s potential investments. Claims that compute spending maps one-for-one to revenue are presented as an asserted relationship, not a demonstrated causal rule; announced funding or investment amounts should not be assumed to be cash already delivered.

The conversation moves to potential 2026 IPOs. O’Driscoll identifies Anthropic and SpaceX as possibilities while saying Stripe seems less likely. These are December 2025 predictions, not announcements or confirmations of any listing.

He describes the challenge of listing a company at an extremely high valuation: a relatively small initial float can be followed by a much larger supply of shares when restrictions expire. Generating sufficient demand at a high price is, in his view, a difficult underwriting problem.

O’Driscoll notes that companies can have different combinations of growth, cash burn, profitability and market position. He would be surprised if every proposed mega-IPO occurred in the same year, and he does not claim to know which will actually happen.

Stebbings and O’Driscoll speculate about SpaceX timing and value, while recognizing that its founder has previously argued for remaining private. Their dates and valuation estimates are guesses, not issuer guidance.

Lemkin predicts four listings—SpaceX, Canva, Databricks and Anthropic—in a particular order, and suggests OpenAI could wait longer. He acknowledges a tendency to be too optimistic on timing; none of these forecasts establishes an IPO plan.

Lemkin asks whether AI-driven unemployment will appear in official statistics by the end of 2026. The question is a proposed future test; the panel does not provide a measurement showing that AI caused a given number of job losses.

O’Driscoll separates actual causation from political perception. He predicts that a general rise in unemployment could generate a backlash against AI even if another factor caused it, particularly when industry leaders themselves discuss possible job displacement.

Lemkin says clear evidence of job losses would intensify public concern, while O’Driscoll thinks concern could grow before causal evidence is available. Both describe possibilities and political reactions, not an established labor-market result.

O’Driscoll closes by warning that AI could attract blame during a broader economic downturn even if its short-term effect were smaller than expected. The guests’ jokes about investor gains underline their perspective as investors rather than resolve the distributional question.