Devin Parekh has a deal he didn't win, and he's not losing sleep over it. The Insight Partners managing director recently watched Legora — a legal AI startup his firm had pursued — land with General Catalyst instead. In an industry where a single missed allocation to a hot AI company can define a partner's year, Parekh's public equanimity is unusual. So is the strategy behind it.
Insight Partners, a $90 billion firm, is deliberately refusing to concentrate its AI bets on the two labs absorbing the overwhelming majority of venture capital: OpenAI and Anthropic. While much of Sand Hill Road races to write ever-larger checks into those two names, Insight is spreading its exposure across competing frontier labs and application-layer companies. The position is contrarian by construction, and Parekh is comfortable defending it.
Devin Parekh's Case for Staying Diversified
The Legora loss is the telling detail. Rather than treat it as a failure to be corrected with more aggressive concentration, Parekh frames competitive deal dynamics as a feature of a market where no single outcome can be reliably predicted. Losing a contested asset to General Catalyst is, in this reading, the cost of not overpaying for any one position — and a reminder that conviction in a single company is not the same as conviction in a category.
That distinction matters. Insight's stance is not a rejection of AI. It is a rejection of the assumption that the AI trade must be expressed through two balance sheets. Parekh has said he is fine holding stakes in rival AI labs — a portfolio posture that accepts internal competition among his own positions rather than treating it as a conflict to be resolved. If one lab's gains come at another's expense, the firm still participates in the category's expansion.
Portfolio theory gives this instinct a formal name. Concentration amplifies both upside and downside; a fund that puts a disproportionate share of capital into a single company is effectively making a directional bet on that company's execution, its governance, its financing runway, and the durability of its competitive moat. Venture returns are famously power-law distributed, which is often cited as justification for doubling down on winners. But the power law describes outcomes across a portfolio, not the wisdom of collapsing that portfolio into one or two names. Diversification within a high-variance category is how investors capture the power law without being destroyed by its left tail.
The Risk of Betting the Farm on a Single AI Frontier Lab
The scale of the pile-on is the context Parekh is pushing back against. Capital flowing into OpenAI and Anthropic rounds during 2025 and 2026 reached levels that dwarf anything in venture history. Data from CB Insights and PitchBook has tracked a market in which a handful of frontier labs absorb a disproportionate share of all venture dollars, with mega-rounds routinely measured in the tens of billions. When a single funding event can exceed the total annual deployment of most established firms, the concentration is structural, not incidental.
Read next Top Technology Trends in 2026 You Need to KnowThat concentration carries specific, identifiable risks. Valuation marks in private AI rounds are set by negotiation, not by public markets, which means they can lag reality in both directions. Governance at frontier labs remains unsettled — corporate structures at several of these companies have already been reorganized, and their long-term control arrangements are still being tested. Compute costs impose a capital intensity that traditional software never demanded. And competitive positions can shift quickly: a model capability lead that looks decisive in one quarter can narrow the next.
For a $90 billion firm, the arithmetic of a concentrated bet is unforgiving. A fund that ties a meaningful share of its returns to one lab's continued dominance is exposed to regulatory action, talent departures, strategic pivots, and financing conditions that no LP can underwrite with confidence. Parekh's approach treats those as correlated risks rather than independent ones — the same shock can hit every position if the entire thesis depends on one company's trajectory.
How a $90 Billion Firm Navigates the AI Investment Landscape
Insight's scale changes the calculus in ways smaller funds cannot replicate. A firm managing $90 billion has the balance sheet to hold positions across multiple labs, application companies, and infrastructure plays simultaneously. It does not need any single deal to return the fund. That freedom is precisely what makes diversification affordable — and what makes concentration, for Insight, a choice rather than a necessity.
The Legora episode illustrates how this plays out in practice. Legal AI is a category where model providers and application companies compete directly, and where a lab's capability release can reshape an application company's economics overnight. Insight's willingness to hold rival positions, rather than commit fully to one horse, reflects an operating view: in a market where the technology stack is still consolidating, the identity of the eventual winner in any given vertical is genuinely uncertain. That uncertainty is not a problem to be solved with more conviction. It is a condition to be managed with more positions.
There is a second-order benefit. A firm holding stakes across competing labs sees deal flow, pricing dynamics, and technical developments from multiple vantage points. That informational advantage compounds. It also keeps Insight in conversations it might otherwise lose — including ones it lost on Legora, where the relationship with founders and the category knowledge both survive the missed allocation.
Implications for the Broader Venture Capital Industry
The venture industry's current posture is a bet on a small number of names. Fund letters and public commentary increasingly tie LP returns to whether one or two AI labs deliver on valuations set in private markets. That is a historically unusual degree of convergence for an asset class whose returns have always depended on dispersion.
Parekh's position is a reminder that the opposite strategy has a coherent logic. If the market has over-concentrated, the firms holding diversified AI exposure may end up better positioned on a risk-adjusted basis — not because they will necessarily produce the highest gross return, but because their outcomes are less dependent on a single company's execution. In venture, where the median fund struggles and the top decile captures most of the value, avoiding catastrophic correlation is not a modest goal. It is the core of the job.
Whether Insight's approach outperforms the concentrated bet will not be known for years. Fund performance in AI will be settled by exits — IPOs, acquisitions, and secondary sales — and the timing of those events remains uncertain. What is already observable is the divergence in strategy. A $90 billion firm publicly declining to join the pile-on is a data point about how at least one sophisticated LP base is being advised on risk.
What This Signals for AI's Next Investment Cycle
The next phase of AI investing will be defined less by which lab raises the largest round and more by which investors can hold a coherent position through a volatile period. Model capability gaps are narrowing; capital requirements are rising; and the path from private valuation to public-market validation is still unproven for the sector's largest players. In that environment, the case for spreading exposure across multiple labs and application companies strengthens rather than weakens.
Insight's strategy is not a prediction that OpenAI or Anthropic will fail. It is a refusal to bet the firm on whether they will succeed. For an industry that has spent two years treating concentration as conviction, that restraint may prove to be the more disciplined position — and the Legora loss, rather than a cautionary tale, may simply be the price of a portfolio built to survive outcomes no one can yet forecast.
Source: TechCrunch
