Nvidia reported $96 billion in Q2 revenue, up 106% year over year, while warning customers of 15%+ AI server price hikes in 2027. Here is what service businesses need to know.
Ido Cohen · Published 2026-08-28 · AI News
Nvidia just reported $96.2 billion in quarterly revenue — up 106% year over year — and simultaneously warned its biggest customers that AI server prices are going up more than 15% starting in early 2027. For a plumber or a dentist, that sounds like Wall Street noise. It is not. These two data points, read together, tell you everything about where AI tool costs are heading and what to do about it before your SaaS bill catches the wave.
The numbers are staggering. Nvidia reported Q2 fiscal year 2027 revenue of $96.2 billion, beating analyst expectations of $92.17 billion and coming in 5.7% above the company's own guidance. Data Center revenue — the division that sells the chips powering every AI tool you use — hit $89.0 billion for the quarter, up 117% from the same period a year earlier. Gross margin held steady at 75.0%.
CEO Jensen Huang's quote from the earnings statement says the quiet part loud: "AI has reached its inflection point. Its tokens are productive and profitable. Now, compute is revenue."
That last phrase — "compute is revenue" — is the key signal. When Nvidia frames GPU processing power as revenue-generating infrastructure rather than R&D expense, it is telling you that the entire AI industry has locked in its bet. The buildout is not slowing down. In fact, Nvidia guided Q3 revenue to $108 billion — another $12 billion sequential jump — and issued a rare full-year outlook forecasting 70% revenue growth for fiscal 2028, well above the 44% analysts expected. Shares jumped more than 5% in after-hours trading after the earnings call.
The record earnings are the headline. The price hike is the story that actually matters to you.
Bloomberg first reported, and multiple outlets confirmed, that Nvidia has notified its biggest customers — the server builders that supply Microsoft, Google, Oracle, and AWS — that AI server system prices will rise more than 15% on shipments starting in early 2027. The increases apply specifically to systems containing Nvidia's flagship Grace Blackwell and Vera Rubin chips. The exact percentage depends on chip generation and memory configuration.
The root cause is not Nvidia flexing pricing power. According to Nvidia CFO Colette Kress on the earnings call, "Memory scarcity today is being driven in large part by the AI buildout itself." The three major high-bandwidth memory (HBM) manufacturers — SK Hynix, Samsung, and Micron — have reportedly committed most of their 2026 HBM production capacity already. When memory gets scarce, everyone up and down the supply chain passes the cost along.
Here is what that chain looks like:
Analyst Dan Ives told Bloomberg that demand for advanced chips may be running at up to 15 times supply. He does not expect supply and demand to reach equilibrium until mid-to-late 2028.
You do not buy Nvidia servers. You buy subscriptions to ChatGPT, Claude, Jasper, or whatever AI writing and lead-gen tool your marketing agency uses. So why does this matter?
Because every AI tool you use runs on rented Nvidia hardware, and the rent is going up.
The path from chip factory to your monthly software bill typically takes 12 to 18 months to work through the stack. The 15% hardware price increase hitting in early 2027 will pressure AI SaaS companies' infrastructure costs through 2027 and into 2028. Companies with thin margins — and most AI startups are burning cash to subsidize cheap pricing — will eventually have to raise prices or get acquired by players with better hardware deals.
There is a counterforce worth naming: per-token inference costs have been falling, even as hardware gets more expensive. That paradox exists because newer chips are dramatically more efficient. A Vera Rubin chip does far more inference per watt than a previous-generation GPU, partially offsetting the higher unit cost. But efficiency gains do not help a company whose memory budget just went up 15% overnight.
The honest answer is that AI tool pricing for service businesses will be volatile through 2028. Some platforms will absorb the cost. Others will not.
Here is the strategic read that most marketing blogs will miss: this is the best possible time to lock in annual AI contracts, not monthly ones.
When infrastructure cost pressure is heading upstream (from Nvidia to the cloud to SaaS), software companies have an incentive to collect annual cash upfront before their own costs rise. That means annual plans are likely to stay cheaper longer than month-to-month plans — the same way airlines discount advance bookings and charge full price at the gate.
Practically, this means:
1. Audit what you are paying monthly vs. annually. If you are on monthly billing for ChatGPT Plus, Claude Pro, Google Workspace AI features, or any AI-assisted CRM, calculate the annual equivalent. Most platforms are offering 15-20% discounts for annual commitments right now.
2. Lock in your primary AI platform for 12 months before Q1 2027. The hardware price hikes hit shipments in early 2027. SaaS repricing tends to follow within two to three quarters.
3. Diversify away from single-provider risk. The A2A protocol joining the Agentic AI Foundation alongside Anthropic's MCP signals that the industry is building interoperability standards so you can switch AI vendors without rebuilding your stack. That's good for you. Do not build workflows so deep into one vendor that you cannot exit.
4. Treat per-token costs as a budget line. If you are using AI to generate content, summarize reviews, answer lead inquiries, or draft proposals, you are spending tokens. Know your monthly token spend. When pricing shifts, you will know whether to cut usage or find a cheaper model.
5. Consider cheaper open-source models for low-stakes tasks. The open-source ecosystem — Llama 4, Qwen3, and similar models — is increasingly capable. Running inference through cheaper providers (Groq, Together AI, Ollama for local use) for tasks like FAQ generation or email drafts can reduce exposure to Nvidia's price cycle entirely.
Nvidia guided fiscal 2028 revenue growth at 70% — more than 25 percentage points above what analysts expected. That is not a company hedging. That is a company that can see its own order book and is telling the market: the AI buildout is not plateauing.
For service businesses, that guidance tells you something important: AI infrastructure spending is not a bubble that pops in 2026 or 2027. The money going into AI hardware is real, sustained, and accelerating. That means the AI tools you are adopting today will get meaningfully more powerful over the next 24 months, not less.
The question is not whether to use AI in your business. That question is settled. The question is whether you are buying it smart — locking in pricing now, understanding which costs are fixed versus variable, and not over-indexing on any single platform at a moment when the underlying cost structure is shifting.
There is also a subtler point here. When Jensen Huang says "compute is revenue," he is describing a world where the companies that control AI infrastructure control the economics of AI services. For a small HVAC company or a real estate agency, that means your AI vendor relationships are now a real procurement decision, not a software purchase. Treat them accordingly.
The Nvidia earnings and server price hike news landed Wednesday. You have a narrow window before the repricing signal moves up the stack. Here is a concrete action list:
The businesses that get blindsided by AI price increases in 2027 will be the ones treating their AI subscriptions as trivial software costs today. The ones that win will be the ones who understood that "compute is revenue" applies to their vendor's business model — and bought accordingly.
---
Will AI tools like ChatGPT or Claude actually get more expensive in 2027?
There is no guarantee, but the cost pressure is real and moving up the supply chain. Nvidia's 15%+ server price hikes will reach cloud providers in 2027, which in turn pressures AI SaaS companies. Some platforms with strong margins or efficiency gains may absorb the cost; others will pass it along through price increases or reduced free-tier access. Locking in annual contracts now is a reasonable hedge.
What is HBM and why does it affect my AI tools?
HBM stands for high-bandwidth memory — the specialized, extremely fast memory chips that sit alongside Nvidia's GPU accelerators inside AI servers. Large language models (the AI behind tools like ChatGPT and Claude) require enormous amounts of this memory to run inference quickly. When HBM supply tightens, the cost of running any AI server — and therefore any AI tool — goes up. The three dominant HBM makers (SK Hynix, Samsung, Micron) have reportedly committed most of their 2026 capacity already.
Does Nvidia's record revenue mean AI is working for regular businesses?
It means AI infrastructure spending is real and sustained — enterprises and cloud providers are committing enormous capital budgets to it. Nvidia's revenue is a leading indicator for how much AI capacity is being built, which translates into what AI tools will be available to small businesses 12 to 24 months from now. The fact that Nvidia guided 70% revenue growth for fiscal 2028 suggests the buildout continues well into next year, meaning AI tools should get more capable, not plateau.
Should I buy AI hardware myself instead of relying on cloud-based AI tools?
For the vast majority of service businesses — plumbers, dentists, lawyers, med spas, real estate agents — the answer is no. Building and maintaining your own AI inference hardware is a full-time engineering job. The economics of renting AI through API or SaaS subscriptions still beat ownership for any business processing under millions of documents per month. The right response to rising infrastructure costs is smarter purchasing of cloud-based AI, not capital investment in servers.
What is the A2A protocol and should I care about it?
A2A (Agent2Agent) is an open standard that lets AI agents built by different companies talk to each other and hand off tasks without custom-built connections. Google transferred it to the Agentic AI Foundation in August 2026, placing it alongside Anthropic's MCP (Model Context Protocol) under neutral governance. For service businesses, this matters in 18 to 24 months: as AI agents become central to marketing, scheduling, and lead qualification workflows, A2A means you will be able to mix tools from different vendors without rebuilding your stack when you switch providers.
---
Sources: