Google Cloud unveiled a Multi-Party Compute (MPC) cryptographic custody service on October 28, 2026, at its inaugural digital assets symposium in New York. The product, currently in private preview with commercial launch planned for early 2027, distributes private key shards across Google Cloud i...
"Digital asset adoption and tokenization are at an inflection point. As regulated financial institutions begin adopting blockchains and tokenization, they need scalable, proven infrastructure to secure their clients' assets." — Rich Widmann, Director of Digital Assets Strategy, Google Cloud
Google Cloud unveiled a Multi-Party Compute (MPC) cryptographic custody service on October 28, 2026, at its inaugural digital assets symposium in New York. The product, currently in private preview with commercial launch planned for early 2027, distributes private key shards across Google Cloud infrastructure, institutional hardware, and third-party cloud providers. Representatives from Morgan Stanley, Citi, Wells Fargo, and OKX were among the symposium speakers. Alphabet disclosed that it already uses the system to secure its own digital assets on Ethereum and Solana.
The timing is deliberate. Private-key compromise has become the dominant attack vector in decentralized finance, accounting for approximately 72% of DeFi losses in 2026, with at least $1.3 billion stolen in the first eight months of the year. The crypto custody provider market, valued at $3.69 billion in 2026, is projected to reach $27.2 billion by 2030, representing a 66.7% compound annual growth rate. Google is entering as an infrastructure provider — not a custodian — selling picks and shovels to the banks that want to hold digital assets but lack the cryptographic tooling to do so securely.
Neither AWS (31% cloud market share) nor Microsoft Azure (24%) currently offers a dedicated crypto custody MPC product. Google's move establishes it as the first hyperscale cloud provider to enter the space directly, placing it in competition with specialized firms including Fireblocks, BitGo, and Anchorage Digital. Over 70% of new institutional crypto products are already launching with MPC-first architecture.
The service distributes cryptographic key shards across three categories of infrastructure: Google Cloud, the client institution's own hardware, and third-party cloud providers. It supports configurable threshold schemes — 2-of-3 and 5-of-7 were cited — meaning a defined subset of shard holders must cooperate to authorize a transaction. No single party, including Google, holds enough material to sign independently.
A transaction policy engine sits between the signing request and execution. The engine verifies each transaction against configurable rules: transfer amount limits, approved destination addresses, multi-department approval requirements. On October 1, 2026, Google expanded the Bitcoin-specific rule set to evaluate transaction amounts, identify destination address types, and detect change outputs.
Alphabet has deployed the system internally to secure its own digital asset holdings on Ethereum and Solana. The company did not disclose the value of those holdings.
The product's distribution model follows what the industry calls "infrastructure-as-a-service" rather than custodial delegation. Banks retain control over their key shards and policy configurations. Google provides the cryptographic compute layer and cloud orchestration. This distinction is material for regulatory purposes: the bank, not Google, would be the qualified custodian under existing frameworks.
Anchorage Digital, the only federally chartered crypto bank in the United States, has been collaborating on Google Cloud's MPC architecture since 2025. Separately, Jack Henry, a banking technology provider serving community and mid-size banks, integrated with Google Cloud Universal Ledger (GCUL) as of October 16, 2026. Gemini Custody announced on October 8, 2026, that it would begin transitioning all institutional accounts to MPC architecture.
The data explains Google's market thesis. Compromised private keys accounted for 43.8% of all crypto theft in 2024, according to Chainalysis. That figure has worsened. In 2026, key and credential theft accounts for approximately 72% of DeFi losses, with at least $1.3 billion stolen in the first eight months of the year.
Two incidents illustrate the velocity: Drift Protocol lost $285 million on April 1, and KelpDAO lost $290 million on April 18 — $575 million combined in 18 days, both attributed to key compromises.
Cumulatively, DeFi protocols have lost $16.69 billion to hacks, with approximately 40% attributed to private-key compromises, according to data from DeFiLlama. Approximately 66% of 2026 losses — roughly $643 million — have been attributed to North Korean state-linked actors.
The threat surface is expanding in other dimensions. AI-linked crypto scams averaged $3.2 million per operation in 2025, compared to $719,000 for non-AI scams, a 4.5x multiplier. Looking further ahead, an estimated 6.9 million BTC are considered potentially vulnerable to future quantum computing attacks on current elliptic curve cryptography.
Single points of key storage — whether hot wallets, poorly configured multi-signature schemes, or hardware security modules with insufficient access controls — remain the primary failure mode. MPC's core value proposition is the elimination of any single point of compromise: no complete key exists in any one location at any time.
The institutional custody market segments along cryptographic architecture, regulatory status, and pricing. The following table summarizes the four primary competitors Google now faces.
| | Google Cloud MPC | Fireblocks | BitGo | Anchorage Digital | |---|---|---|---|---| | Architecture | MPC, distributed shards (2-of-3, 5-of-7) | MPC-CMP | HSM + multi-sig | HSM-based custody | | Status | Private preview; commercial launch early 2027 | Operational; 2,400+ institutional clients | Operational; conditional OCC national bank charter | Operational; full OCC federal bank charter (Jan 2021) | | Regulatory Posture | Infrastructure provider (not custodian) | Technology provider | NY trust charter, SD trust charter, HK TCSP license, conditional OCC charter | Only federally chartered crypto bank in U.S. | | Scale Metrics | Alphabet internal deployment (ETH, SOL) | $4T+ annual transfer volume | Not disclosed | $310B+ custodied (with Coinbase Custody combined) | | Pricing (Mid-Tier) | Not disclosed | $10K–$50K/month | Not disclosed | Not disclosed | | Key Differentiator | Hyperscale cloud integration; bank retains custody | Largest installed base; broadest protocol coverage | Regulatory charter breadth | Federal bank charter; regulatory-first model |
Fireblocks dominates by client count and transfer volume. However, the platform faces security scrutiny. In July 2026, researchers identified 18 vulnerabilities in the Fireblocks MPC library, including 2 rated critical and 4 high-severity. On October 26, 2026, CERT/CC disclosed CVE-2026-90975 in Fireblocks' mpc-lib: a hardened-derivation flaw in which a compromised descendant signer share combined with its chain code can expose ancestor shares. The vulnerability is architecturally significant — it means a breach at one level of the key hierarchy could propagate upward.
BitGo has pursued regulatory licensing aggressively, holding charters and licenses across multiple jurisdictions. Its reliance on HSM and multi-signature architecture, rather than MPC, reflects an older cryptographic model that some institutions consider more battle-tested but less operationally flexible.
Anchorage Digital occupies a unique regulatory position as the only federally chartered crypto bank, with its OCC charter granted in January 2021. Its collaboration with Google Cloud on MPC architecture since 2025 suggests a potential hybrid model: Anchorage's regulatory charter combined with Google's infrastructure.
Google's entry reframes the competitive dynamic. The company is not competing for custody mandates — it is competing for the infrastructure layer beneath them. The distinction matters.
AWS leads the cloud market at 31% share and hosts approximately 25% of Ethereum nodes. Azure holds 24%. Neither offers a dedicated crypto custody MPC product. Google is not the largest cloud provider, but it is the first to package cryptographic key management, threshold signing, and policy enforcement into a single institutional-grade product.
The banking technology integration signals are concrete. Jack Henry's October 16 integration with Google Cloud Universal Ledger creates a distribution channel into community and mid-size banks — institutions that lack the internal engineering capacity to build custody infrastructure but face client demand for digital asset services.
The symposium speaker roster — Morgan Stanley, Citi, Wells Fargo, OKX — indicates that Google is targeting both traditional financial institutions and crypto-native exchanges. The infrastructure-provider positioning allows Google to serve both segments without the regulatory burden of being a custodian.
The over-70% adoption rate of MPC-first architecture among new institutional crypto products indicates that the market has already converged on MPC as the default approach. Google is not evangelizing the technology; it is arriving after the architectural consensus has formed, offering to provide it at cloud scale.
MPC eliminates the single-key-compromise vector. It does not address the full spectrum of institutional custody risk.
Poorly designed approval policies remain a vulnerability regardless of the underlying cryptographic scheme. A 2-of-3 threshold is only as secure as the governance framework determining who holds the three shards and under what conditions they sign. MPC does not prevent insider collusion — if two of three shard holders cooperate maliciously, the threshold is met and the transaction executes.
Software coordination flaws represent another category of risk that MPC does not inherently resolve. The Fireblocks CVE-2026-90975 disclosure demonstrates that implementation vulnerabilities in MPC libraries can undermine the theoretical security guarantees of the threshold scheme itself. Google's system is not immune to analogous implementation risks.
Google's dual role as infrastructure provider and operator of one of the three shard environments introduces concentration questions. If a bank distributes shards across Google Cloud, its own hardware, and a third-party cloud provider, and the third-party cloud provider is also running on Google Cloud infrastructure, the effective distribution is narrower than the 2-of-3 scheme suggests.
The product is in private preview. Pricing, SLA terms, liability frameworks, and insurance coverage have not been disclosed. Regulated institutions will require clarity on all four before committing production workloads.
Quantum risk remains unaddressed by current MPC implementations. The 6.9 million BTC estimated as potentially vulnerable to future quantum attacks represent a systemic risk that threshold signing schemes, in their current form, do not mitigate.
Google's entry into crypto custody infrastructure marks a structural shift. The company is not competing with Fireblocks or Anchorage for custody clients — it is offering to become the compute and cryptographic layer those firms and their bank clients run on. The $3.69 billion custody market is projected to grow sevenfold by 2030. Google is positioning to capture a share of that growth without taking on custodial liability or regulatory obligations.
The timing aligns with a measurable crisis: $1.3 billion in key-compromise losses in eight months of 2026, up from 43.8% of theft volume in 2024 to approximately 72% today. Institutions have a quantifiable reason to move away from single-key architectures.
Whether Google's product proves robust in production remains to be determined. The private preview period, the undisclosed pricing, and the inherent limitations of MPC — which secures key material but not governance, policy design, or implementation quality — are open variables. The Fireblocks CVE disclosure is a reminder that MPC implementations carry their own risk surface. Google will need to demonstrate that its system is not merely architecturally sound but operationally resilient at institutional scale.