Monolithic Power Systems

Q2 2026 is the quarter that broke the bear case. Revenue grew 47.6% y/y to a record $980.6M with Enterprise Data up 164.3% y/y, GAAP operating margin expanded 6.2 points to 31.0% on flat gross margin (pure operating leverage), the Q3 guide implies ~56% y/y growth, and all of it is funded from a zero-debt balance sheet with $1.41B of net cash — the AI power-content thesis is compounding regardless of any single platform socket.

• Enterprise Data revenue of $380.6M was up 164.3% y/y and 44.8% q/q, and is now 38.8% of the company. Whatever share MPWR lost on one AI platform generation, the segment has more than tripled — the 'AI ceiling' framing is contradicted by the filed segment table
• Growth is broadening, not narrowing: ALL six end markets grew sequentially, and Communications was up 78.3% y/y on optical-module and switch power. AI datacenter interconnect is a second, independent AI-levered growth vector alongside server power
• Operating leverage is real and clean — GAAP operating margin went from 24.8% to 31.0% y/y while gross margin moved just +0.1pt. Opex grew 17.8% y/y against 47.6% revenue growth, so the incremental margin is structural, not a one-off
• Gross margin held at 55.2% GAAP / 55.6% non-GAAP through a violent mix shift toward AI, and Q3 is guided 55.4-56.0% non-GAAP. That is evidence of pricing power in a segment where a share-loser would be discounting
• The Q3 guide of $1,140-1,160M implies ~+17% sequential and ~+56% y/y (derived) — acceleration off a record quarter, not a peak-quarter fade
• Capacity commitment was raised again, now 'significantly beyond $6B' (up from the ~$6B target that itself replaced ~$4B), plus front-end foundry AND complex back-end 3D-power-module assembly — management is spending against multi-year visibility
• The product roadmap is widening the addressable market beyond core server power: initial orders for high-speed DDR5 memory interface components (SAM expansion into 2027) and first samples of high-voltage AC-to-DC parts for 800V DC datacenter architectures
• Fortress balance sheet with optionality: $1,413.8M cash and short-term investments, zero debt, and the buyback authorization doubled to $1B — the company can fund capacity, R&D and repurchases through any AI air-pocket without financing
• Automotive is quietly compounding underneath the AI story — over 1,500 new sockets shipped year-to-date in ADAS and other in-vehicle applications, a design-win pipeline that converts to revenue over multiple years

The operating result is excellent and the price already assumes it: ~20.2x TTM sales and ~82x TTM GAAP earnings (both derived) for a company whose cash generation went BACKWARDS in its record quarter, whose inventory is at $675.8M, and which is still carrying an unremediated material weakness after restating FY2024 and every 2025 quarter.

• Cash conversion broke down in the best quarter the company has ever printed: operating cash flow of $227.9M was LOWER than Q1 2026's $250.3M and Q2 2025's $237.6M, even though GAAP net income rose 90.6% y/y. OCF was just 0.89x net income versus 1.76x in Q2 2025 (derived) — the earnings are real but they are sitting in working capital, not the bank
• Inventory rose to $675.8M from $619.2M q/q and $490.6M a year ago, and receivables rose to $343.6M from $194.8M. Days of inventory fell only because revenue ran hot — if the Q3 guide is the peak, 140 days of inventory becomes the problem
• Valuation leaves no room: ~$66.1B market cap on ~$3,273M TTM revenue is ~20.2x sales, and ~82x TTM GAAP net income of ~$802M (both derived). The net cash pile of $1.41B is only ~2.1% of the price, so it offers essentially no valuation cushion
• Accounting credibility is impaired and not yet repaired. On 2026-02-26 the audit committee determined that the FY2024 10-K and EVERY 2025 10-Q should no longer be relied upon; the restatement cut FY2024 income tax benefit and deferred income taxes by $194.6M. The company still lists 'ability to timely and adequately remediate its material weakness' as a live risk factor in the Q2 2026 release's safe harbour — and management made no comment on remediation on the Q2 call
• GAAP-to-non-GAAP gap is wide and structural: $6.50 non-GAAP vs $5.22 GAAP diluted EPS, a ~$62.8M quarterly difference driven mainly by $53.5M of stock-based compensation. On the guided Q3 numbers SBC stays at $53.2-55.2M, so ~20% of non-GAAP earnings is a real, dilutive cost
• Concentration risk is asserted away rather than disclosed away. Management stated on the call that Enterprise Data 'does not have customer concentration', but no customer-percentage disclosure accompanies the Q2 release, and 38.8% of revenue now sits in the single most cyclical, most platform-cadence-dependent end market
• The two end markets that would cushion an AI digestion are not growing fast enough to do so: Storage & Computing +2.3% y/y and Consumer -4.8% y/y. Automotive at +8.2% y/y is solid but is only 16.0% of revenue
• Capex intensity is rising off a low base (6.2% of revenue in FY2025 to 8.8% in Q1 2026, derived) as MPWR builds back-end 3D-module assembly. The capital-light fabless argument weakens as it becomes a module and 'full-solution' provider — and Q2 capex is not yet disclosed, so the current run-rate is unknown
• MPWR guides only one quarter forward and gives no annual outlook, so there is no company-provided visibility beyond 2026-09-30 to underwrite a ~20x-sales multiple