Several firms race to invent; the first success wins the patent. Then watch the profit flow decay while renewal fees rise, until the owner lets the patent lapse. The strip below shows expected payoff for different R&D budgets.
The science
Race. Each of n firms discovers the invention at a random time with constant hazard h = a·√x per year, where x is its R&D spending (diminishing returns). The first discovery time is the minimum of n exponentials, so τ ~ Exp(n h), mean 1/(n h), and each firm wins with probability 1/n. With continuous discount rate r, E[e−rτ] = n h/(n h + r).
Patent life. After the grant the profit flow decays, π(t) = π0 e−g t, for at most 20 years. A renewal fee fj = f0(1 + 0.35 j) is paid at the start of year j ≥ 1; the owner renews only while that year's profit flow exceeds the fee, so the patent lapses at the first year where it does not. Present value at grant: V = Σj [π0(e−(g+r)j − e−(g+r)(j+1))/(g+r) − fj e−rj].
Payoff per firm (winner-take-all, firms pay their R&D spend until the race ends): Π = V·E[e−rτ]/n − x·(1 − E[e−rτ])/r. The chart plots Π against x for the current n; where it is negative, entering the race loses money on average.
Independent checks: (A) 40,000 simulated races compare mean race length, win share and discount factor with the closed forms; (B) the closed-form yearly sum for V is compared with a separate fine-step (0.001 y) numerical integration of the same stream.
Simplifications: identical firms, constant hazard (no learning, no spillovers, no partial progress), no litigation, licensing or invalidity risk, a made-up fee schedule and product values in million dollars. Real patent fees, terms and renewal rules differ by country and patent type, so spot-check them for your jurisdiction. Educational only, not legal advice.