Broadcom (AVGO) deep dive, April 12, 2026: 18.1% weekly surge on Google TPU deal and Anthropic compute pact

NEW YORK, April 12, 2026. Broadcom (Nasdaq: AVGO) delivered one of the largest single-week gains in its history as a public company, rising 18.1% from $314.43 to close at $371.55 across the April 7–11 trading week — a gain of approximately $57 per share adding roughly $270 billion in market capitalisation in five sessions. The catalyst arrived after Monday’s market close: an SEC 8-K filing disclosing two formally structured, multi-year agreements that the market immediately understood for what they are — not just revenue announcements, but a structural transformation of how Broadcom’s AI semiconductor business is priced. The first agreement locks Broadcom as Google’s primary custom silicon partner for TPU design, supply, and networking through 2031. The second expands the collaboration between Broadcom, Google, and Anthropic to give the Claude developer access to approximately 3.5 gigawatts of next-generation TPU-based compute capacity starting in 2027. Together, the two deals convert Broadcom’s most important revenue stream — its AI ASIC business — from a business that must re-compete annually for customer wins into one with contractually anchored demand visibility across multiple product generations. The market spent the week repricing that transformation, and Seaport Global’s contrarian downgrade to Neutral (“gains are fully factored in”) was the only dissenting institutional voice in an overwhelming bull consensus.

The weekly price action — day by day

AVGO entered the week at $314.43 on Monday’s close — down approximately 24% from its December 10, 2025 all-time closing high of $411.32 and sitting near multi-month lows, having absorbed the broader technology sector’s Iran war-driven de-risking alongside its own investor concern about customer concentration and hyperscaler capital expenditure sustainability. The 8-K filing after Monday’s close was the trigger. Tuesday’s open at approximately $327 (+4% from Monday’s close) was driven by early institutional recognition of the deal’s significance; the stock continued grinding higher through the session, closing the Tuesday session with approximately 6% gains as the analyst community digested the filing’s implications. By Thursday, the ceasefire-driven broader market relief rally added a macro tailwind to the company-specific momentum: AVGO closed Thursday at $354.91 (+1.22% on the session, up approximately 12.8% from Monday’s close). Friday’s final leg — driven by the “bifurcated” CPI reading that provided relief to rate-sensitive growth stocks and a broader tech sector rally — lifted AVGO to $371.55, a +4.69% Friday session on volume of 29.96 million shares (above the 24.16 million daily average). The intraday range on Friday: $357.01 to $376.54. Market capitalisation at Friday’s close: approximately $1.76 trillion. The 52-week range now reads $161.61 (April 21, 2025 low) to $414.61 (December 10, 2025 high), with the current price sitting approximately 10% below the all-time high after an 18.1% weekly gain.

The 8-K deal mechanics — what Broadcom actually filed

The April 7 8-K disclosed two distinct but structurally connected agreements. The first is a long-term agreement with Google LLC to design and supply custom Tensor Processing Units for future generations of Google’s TPU programme, along with a supply assurance agreement covering networking components and other silicon for Google’s next-generation AI racks — both running through up to 2031. The partnership between Broadcom and Google on custom silicon predates the 8-K by a decade: the two companies have co-designed TPUs since 2016 across seven generations, with the current seventh-generation TPU, internally codenamed “Ironwood,” now in production. The agreement codifies what had previously been an ongoing partnership into a formal contractual framework — removing the single most consequential bear case against AVGO’s AI revenue thesis, which was the possibility that Google might gradually internalise its chip design capabilities or shift to a competing ASIC vendor. The 2031 horizon converts what was a relationship-dependent revenue stream into a contracted backlog with approximately five years of forward visibility across multiple TPU generations.

The second agreement expands the tripartite collaboration between Broadcom, Google, and Anthropic. Under the expanded terms, Anthropic will access approximately 3.5 gigawatts of next-generation TPU-based AI compute capacity through Broadcom beginning in 2027, as part of Anthropic’s multi-gigawatt total compute commitment. CEO Hock Tan had previously disclosed on the Q1 FY2026 earnings call that Anthropic’s demand for TPU compute is expected to scale from approximately 1 GW in 2026 to over 3 GW in 2027. The 3.5 GW 8-K figure formalises and exceeds that guidance with a structural commitment. The 8-K specifies that Anthropic’s actual utilisation of that capacity “depends on its continued commercial success” — a contingency clause that is standard in infrastructure capacity agreements of this type and that carries minimal risk given Anthropic’s current trajectory: the company’s annualised revenue run rate has surpassed $30 billion (up from approximately $9 billion at year-end 2025), it now counts over 1,000 business clients spending more than $1 million annually, and Anthropic CFO Krishna Rao described the deal as “our most significant compute commitment to date to keep pace with our unprecedented growth.” Anthropic had previously placed a $10 billion custom chip order with Broadcom in late 2025, and the 3.5 GW agreement represents the next generation of that capital commitment.

Why the 2031 horizon changes Broadcom’s valuation framework

The most analytically important dimension of the 8-K is not any individual revenue number — it is the time horizon. Semiconductor supply agreements of five-year duration are structurally unusual and historically command premium valuation multiples precisely because they reduce the earnings volatility that typically forces semiconductor companies to trade at discounts to software peers. Standard chip supply contracts operate on 12–24 month cycles that align with product generations and customer capital expenditure planning windows. When Broadcom discloses a 2031 TPU agreement with Google, it is not announcing a one-time order — it is announcing that its largest single revenue relationship has been converted from cycle-by-cycle re-competition into a long-term contracted framework spanning multiple processor generations, two or more capital expenditure cycles, and an unknown number of AI model training generations. HSBC estimates that TPU-related shipments account for approximately 78% of Broadcom’s total ASIC revenue. Securing that relationship through 2031 means securing approximately 78% of the most rapidly growing component of the company’s revenue against the single most frequently cited institutional bear case.

The networking component of the Google agreement adds a second contracted revenue stream that compounds alongside the silicon business. The 8-K explicitly includes “networking and other components for Google’s next-generation AI racks” — Broadcom’s switching and interconnect silicon that connects TPU clusters into the large-scale training and inference infrastructure Google runs for Gemini and other AI products. Broadcom recently shipped the world’s first 102.4 Tbps switch in production volume — a networking product milestone that positions it as the dominant supplier of the bandwidth infrastructure required to connect next-generation AI accelerator clusters. As Google builds out TPU “Ironwood” clusters at scale through 2031, the networking layer scales proportionally, creating a revenue stream that is multiplicative rather than additive relative to the silicon business alone. CEO Hock Tan’s stated target — “line of sight to achieve AI revenue from chips, just chips, in excess of $100 billion in 2027” — was already a headline-generating commitment at the Q1 earnings call. The 8-K provides the most concrete single contractual foundation that target has received. With a $73 billion AI backlog in 18-month visibility already disclosed, the 2031 Google deal adds structural durability to that backlog beyond any previous 18-month window Broadcom has guided investors to expect. For technology sector investors modelling Broadcom’s revenue trajectory, the 2031 horizon is the most important single element of the 8-K — not the gigawatt figures, not the networking scope, but the time contract itself.

Q1 FY2026 fundamentals — the earnings baseline the deal builds on

The 8-K arrived on the back of Broadcom’s strongest quarterly earnings report in its history. Q1 FY2026 (the quarter ended February 1, 2026, reported March 4) produced revenue of $19.31 billion against the $19.18 billion estimate — a 29% year-over-year increase. Adjusted EPS of $2.05 beat the $2.03 estimate. The AI-specific revenue figure — $8.4 billion, up 106% year-over-year from $4.1 billion in Q1 FY2025 — was the single most consequential data point in the print, demonstrating that the hyperscaler capex cycle that began accelerating in 2024 is not decelerating in the face of macro uncertainty. The semiconductor solutions segment delivered $12.52 billion (above the $12.25 billion estimate). Infrastructure software delivered $6.80 billion (below the $7.02 billion estimate) — the one miss in an otherwise strong quarter, reflecting the transition timing of VMware subscription migrations rather than any demand deterioration. Net income was $7.35 billion, or $1.50 per share GAAP, versus $5.50 billion a year earlier. The board authorised up to $10 billion in new share buybacks through 2026.

Q2 FY2026 guidance from the March 4 earnings call: revenue of approximately $22 billion (representing approximately 47% year-over-year growth), AI chip revenue approaching $10.7 billion — a near-30% sequential increase from Q1’s $8.4 billion — and adjusted operating margin of approximately 68%, above the 66% StreetAccount consensus. The $22 billion Q2 revenue guide was itself a landmark: it implies annualised revenue approaching $88 billion for the fiscal year, a scale that no semiconductor company has achieved in its core chip business outside of Nvidia. Mizuho projects full-year FY2026 AI revenue of $40.4 billion — approximately 92% growth from FY2025 — treating the Q1 and Q2 guidance as a durable run rate. At $40 billion in AI revenue alone, Broadcom’s AI segment would represent the second-largest AI chip franchise in the world by revenue, approximately one-third of Nvidia’s scale but growing at comparable percentage rates from a larger base. The $73 billion committed AI backlog over 18 months — disclosed on the March earnings call — is the financial expression of the same structural demand that the 8-K’s 2031 Google agreement institutionalises at the contract level.

The custom silicon competitive landscape — AVGO vs NVDA vs MRVL

Broadcom’s competitive positioning in the AI custom silicon market is structurally different from Nvidia’s, and that difference is why the 2031 Google agreement matters for sector investors beyond the AVGO-specific thesis. Nvidia dominates the general-purpose AI GPU market — its H100 and B200 series are the default training accelerators for virtually every AI lab that does not have the scale to justify building its own custom chip. The Nvidia GPU’s strength is its software ecosystem (CUDA), its broad applicability across model architectures, and its production volume. Its weakness is that it is a general-purpose tool: hyperscalers running billions of inference requests per day on known, stable model architectures can design custom ASICs — application-specific integrated circuits — that outperform Nvidia GPUs on the specific workloads they run at lower power consumption and cost per inference. Google’s TPUs, Meta’s MTIA chips, and Amazon’s Trainium/Inferentia are all expressions of this same strategy: deploy Nvidia GPUs for training and rapid iteration at the frontier, deploy custom ASICs for the high-volume, stable inference workloads at scale.

Broadcom is the primary designer and supplier of those custom ASICs for Google and Meta, and has recently added OpenAI and Anthropic to its customer roster. Marvell Technology (Nasdaq: MRVL) is Broadcom’s closest ASIC competitor, supplying custom silicon to Amazon (for Trainium and Inferentia) and Microsoft (for its internal AI infrastructure). The custom ASIC market is not winner-takes-all — each hyperscaler requires a design partner that can invest in understanding its specific model architectures, data centre infrastructure, and supply chain requirements. Broadcom’s decade-long Google relationship, its deep integration with Google’s infrastructure planning, and its Ironwood TPU track record give it advantages in the Google-and-Anthropic ecosystem that Marvell cannot replicate quickly. The 2031 Google agreement is the formal expression of those advantages as a contractual moat. Hock Tan’s confirmation on the Q1 call that Meta’s MTIA roadmap is “alive and well” and that Meta is targeting multiple gigawatts of custom accelerator capacity in 2027 adds a second hyperscaler anchor to what is increasingly a multi-customer, multi-gigawatt AI silicon franchise. For investors comparing AVGO and NVDA as AI infrastructure exposures, the 8-K’s significance is that it converts AVGO’s customer concentration from a risk factor (what if Google internalises chip design?) into a competitive moat (Google has contractually locked in Broadcom through 2031). Understanding how analyst ratings reflect competitive positioning assessments is relevant context for reading the 47-of-49 buy consensus in this light.

VMware — the underappreciated second act

The AI narrative has so dominated Broadcom’s investor story that the VMware integration has received disproportionately little coverage relative to its contribution to the investment thesis. The $69 billion VMware acquisition, closed in November 2023, is 18 months post-close — and by every measure available, it is executing above the targets Broadcom set during the deal process. More than 300,000 enterprise customers have migrated from perpetual licences to recurring subscription contracts. Operating margins from the Infrastructure Software segment have expanded materially as Broadcom applied its standard playbook: eliminate redundant product lines, consolidate sales teams, and concentrate R&D on the highest-margin SKUs. CEO Tan’s explicit pushback against analyst concerns about AI disruption to VMware on the Q1 call — “Our infrastructure software is not disrupted by AI” — addressed the most frequently raised bear case against the software segment. The infrastructure virtualisation layer that VMware provides is not being replaced by AI tools; it is increasingly required to orchestrate the complex multi-cloud AI infrastructure that enterprises are building. As Broadcom noted, the $10 billion board-authorised buyback through 2026 reflects confidence in the combined entity’s cash generation — both the AI chip revenues and the VMware subscription stream. At $6.80 billion in Q1 Infrastructure Software revenue (annualised pace above $27 billion), VMware provides a stable, high-margin recurring revenue base that would be a standalone public company of significant scale. Combined with the AI chip business, it creates a dual-engine growth story that is genuinely without peer in the semiconductor sector.

The CFO transition and governance signal

One detail from the April 8 announcement period that received less attention than the deal mechanics but carries meaningful governance signal: Broadcom recently named Amie Thuener — Google’s former Chief Accounting Officer — as its incoming CFO, effective June 12, 2026. The hire carries dual significance. First, it signals that Broadcom’s management team is confident enough in the company’s sustained AI-driven growth to invest in finance leadership with direct, senior experience in large-scale AI infrastructure transactions — exactly the domain the Google and Anthropic agreements require. Second, it deepens the operational relationship between Broadcom and Google at the executive level: a CFO who spent years inside Google’s financial infrastructure has direct institutional knowledge of how Google structures its capital expenditure commitments, how it manages multi-year technology agreements, and how it evaluates the financial performance of its hardware partnerships. That knowledge is strategically valuable as Broadcom navigates what is likely to be an increasingly complex set of multi-year agreements with hyperscalers. The hire is consistent with a management team that views the current AI infrastructure cycle as a long-duration structural opportunity, not a cyclical trade.

Risks and the bear case — what could interrupt the thesis

The 18.1% weekly surge and near-unanimous analyst buy consensus do not mean the investment is without risk. Seaport Global’s downgrade — the contrarian call of the week — surfaces the most credible near-term concern: at approximately 65x trailing P/E and $371 per share, AVGO is pricing execution on the $100 billion FY2027 AI revenue target. If that target is delayed by one quarter due to supply chain constraints (TSMC CoWoS packaging capacity remains constrained, and Nvidia has reserved a large share), customer capital expenditure moderation, or a sustained period of geopolitical uncertainty that forces hyperscalers to defer infrastructure commitments, the stock’s premium multiple would face downward pressure even without any fundamental deterioration. The customer concentration risk is structurally reduced by the 2031 agreement but not eliminated: Google accounts for approximately 78% of ASIC revenue and any deterioration in Google’s own AI capex programme — whether from regulatory headwinds, competitive pressure from OpenAI, or macro-driven capex reduction — would disproportionately impact Broadcom’s revenue even with contractual protections in place.

The EU regulatory overhang on the VMware acquisition remains unresolved, and while management has expressed confidence in eventual approval, any regulatory action that requires Broadcom to divest VMware product lines would impair the Infrastructure Software segment’s margin profile. The Iran war’s energy shock is a minor but real headwind for Broadcom’s hyperscaler customers: data centres are power-intensive, and elevated energy costs pressure the operating economics of the same AI infrastructure investments that drive Broadcom’s chip demand. Most hyperscalers have hedged energy exposure through long-term power purchase agreements, but sustained $100+ crude oil puts upward pressure on electricity cost structures over multi-year contract renewals. None of these risks invalidates the core thesis — a business with $73 billion in AI backlog, a $100 billion FY2027 revenue target, a 2031 Google contract, and 3.5 GW of Anthropic compute demand is structurally positioned among the strongest in global technology. They are reasons to size the position with appropriate precision rather than maximum conviction. For tracking AVGO’s ongoing price action, PreMarket Daily covers pre-market moves and analyst developments before each session.

What comes next — the Q2 FY2026 earnings inflection

Broadcom’s next major catalyst is Q2 FY2026 earnings, expected in early June 2026 (the quarter ending May 3, 2026). The consensus is now pricing the $22 billion revenue guide and approximately $10.7 billion in AI chip revenue as the floor rather than the target — the 8-K’s 2031 Google agreement and 3.5 GW Anthropic commitment provide structural confirmation that the demand runway supporting those numbers is durable. The Q2 print will be the first earnings report that investors can read with the 2031 context in mind, and the primary question will be whether AI revenue lands at or above the $10.7 billion Q2 guide — a sequential jump of approximately 27% from Q1’s $8.4 billion that would itself be the largest single-quarter AI revenue growth in the company’s history. Any upward revision to the FY2026 AI revenue trajectory or the $100 billion FY2027 AI chip target would be the single most powerful catalyst available to a stock that has already re-rated 18.1% in a single week. Any disappointment on the guidance — particularly if the Anthropic capacity utilisation contingency clause becomes the subject of investor concern — would test whether the week’s buyers were pricing in execution or aspiration. Between now and June, the weekly Islamabad ceasefire negotiations, the Federal Reserve’s April 28–29 FOMC decision, and the broader S&P 500’s trajectory through bank earnings season will set the macro context for how AVGO’s company-specific fundamental story lands with institutional investors.