diff --git a/src/deploy/functions/prepare.ts b/src/deploy/functions/prepare.ts
index 7572b1f8325..cb66dc06685 100644
--- a/src/deploy/functions/prepare.ts
+++ b/src/deploy/functions/prepare.ts
@@ -213,7 +213,7 @@ export async function prepare(
   for (const [codebase, { wantBackend, haveBackend }] of Object.entries(payload.functions)) {
     inferDetailsFromExisting(wantBackend, haveBackend, codebaseUsesEnvs.includes(codebase));
     await ensureTriggerRegions(wantBackend);
-    resolveCpu(wantBackend);
+    resolveCpuAndConcurrency(wantBackend);
     validate.endpointsAreValid(wantBackend);
     inferBlockingDetails(wantBackend);
   }
@@ -331,17 +331,7 @@ export function inferDetailsFromExisting(
     // if there are >1 CPU because this could expose race conditions. They
     // either need to start on concurrency where they've always needed to
     // handle race conditions, or they should explicitly enable.
-    if (typeof wantE.concurrency === "undefined") {
-      const explicitlySmallCPU = typeof wantE.cpu === "number" && wantE.cpu < 1;
-      // !availableMemoryMB means we'll default to 256
-      const implicitlySmallCPU =
-        wantE.cpu === "gcf_gen1" && (wantE.availableMemoryMb || 256) < 2048;
-      if (explicitlySmallCPU || implicitlySmallCPU) {
-        wantE.concurrency = 1;
-      } else if (haveE.concurrency) {
-        wantE.concurrency = haveE.concurrency;
-      }
-    }
+    // We'll hanndle this in setCpuAndConcurrency
 
     wantE.securityLevel = haveE.securityLevel ? haveE.securityLevel : "SECURE_ALWAYS";
 
@@ -419,7 +409,7 @@ export function inferBlockingDetails(want: backend.Backend): void {
  * provided and sets concurrency based on the CPU level if not provided.
  * After this function, CPU will be a real number and not "gcf_gen1".
  */
-export function resolveCpu(want: backend.Backend): void {
+export function resolveCpuAndConcurrency(want: backend.Backend): void {
   for (const e of backend.allEndpoints(want)) {
     if (e.platform === "gcfv1") {
       continue;
@@ -429,5 +419,9 @@ export function resolveCpu(want: backend.Backend): void {
     } else if (!e.cpu) {
       e.cpu = backend.memoryToGen2Cpu(e.availableMemoryMb || backend.DEFAULT_MEMORY);
     }
+
+    if (!e.concurrency) {
+      e.concurrency = e.cpu >= 1 ? backend.DEFAULT_CONCURRENCY : 1;
+    }
   }
 }
diff --git a/src/test/deploy/functions/prepare.spec.ts b/src/test/deploy/functions/prepare.spec.ts
index 280a96e9820..ca96c93beef 100644
--- a/src/test/deploy/functions/prepare.spec.ts
+++ b/src/test/deploy/functions/prepare.spec.ts
@@ -123,7 +123,7 @@ describe("prepare", () => {
       expect(want.availableMemoryMb).to.equal(512);
     });
 
-    it("downgrades concurrency if necessary", () => {
+    it("downgrades concurrency if necessary (explicit)", () => {
       const have: backend.Endpoint = {
         ...ENDPOINT_BASE,
         httpsTrigger: {},
@@ -137,6 +137,42 @@ describe("prepare", () => {
       };
 
       prepare.inferDetailsFromExisting(backend.of(want), backend.of(have), /* useDotEnv= */ false);
+      prepare.resolveCpuAndConcurrency(backend.of(want));
+      expect(want.concurrency).to.equal(1);
+    });
+
+    it("downgrades concurrency if necessary (implicit)", () => {
+      const have: backend.Endpoint = {
+        ...ENDPOINT_BASE,
+        httpsTrigger: {},
+        concurrency: 80,
+        cpu: 1,
+      };
+      const want: backend.Endpoint = {
+        ...ENDPOINT_BASE,
+        httpsTrigger: {},
+        cpu: "gcf_gen1",
+      };
+
+      prepare.inferDetailsFromExisting(backend.of(want), backend.of(have), /* useDotEnv= */ false);
+      prepare.resolveCpuAndConcurrency(backend.of(want));
+      expect(want.concurrency).to.equal(1);
+    });
+
+    it("upgrades default concurrency with CPU upgrades", () => {
+      const have: backend.Endpoint = {
+        ...ENDPOINT_BASE,
+        httpsTrigger: {},
+        availableMemoryMb: 256,
+        cpu: "gcf_gen1",
+      };
+      const want: backend.Endpoint = {
+        ...ENDPOINT_BASE,
+        httpsTrigger: {},
+      };
+
+      prepare.inferDetailsFromExisting(backend.of(want), backend.of(have), /* useDotEnv= */ false);
+      prepare.resolveCpuAndConcurrency(backend.of(want));
       expect(want.concurrency).to.equal(1);
     });
   });