Barron’s put four names on its buy list heading into the heart of tech earnings season on October 19, 2026: NVIDIA Corporation, Intel Corporation, Micron Technology, Inc., and Monolithic Power Systems, Inc. The call, built around a KeyBanc Capital Markets analysis from semiconductor analyst John Vinh, leans on a simple data point that traders have been watching all month: chip shipments are running 14.9% above trend on a three-month moving-average basis, measured through August. For investors, that is a tidy stock-picking story. For software developers, it is something closer to an early warning.
The same memory and accelerator tightness that is lifting semiconductor stocks is the raw material behind every AI coding assistant on the market today, from GitHub Copilot to Cursor to Anthropic’s Claude Code to OpenAI’s Codex CLI. When DRAM, high-bandwidth memory (HBM) and NAND flash get scarce, cloud GPU capacity gets more expensive and harder to reserve, and that cost eventually lands somewhere, whether in usage caps, pricing tiers, or slower queues for autonomous coding agents. This is a story about four stock tickers, but it is also a story about what engineering teams will be paying for AI-assisted development tools in the months ahead.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Barron’s Names Four Chip Stocks as Earnings Season Opens
The Barron’s headline, “Nvidia, Intel, Micron, Monolithic Power Are Buys Ahead of Tech Earnings,” landed as the sector’s most closely watched earnings window of the year got underway. According to Business Insider’s coverage of the same KeyBanc note, analyst John Vinh identified the four companies as his top semiconductor picks heading into third-quarter results, citing broad-based strength in chip demand rather than a single product cycle.
What makes the call notable is the breadth of the companies involved. Nvidia represents the AI accelerator market. Intel represents the traditional CPU and foundry business attempting a turnaround. Micron represents the memory supply chain that feeds every AI data center. And Monolithic Power Systems represents the power-delivery components that keep high-density GPU racks running without melting down. Grouping all four into one “buy” call signals that KeyBanc views the current cycle as a supply-chain-wide phenomenon, not an isolated Nvidia story.
That distinction matters for anyone trying to understand why AI infrastructure costs keep climbing even as individual chip vendors post strong results. A shortage anywhere in the stack, whether it is HBM capacity, DRAM contract pricing, or power-management silicon, can bottleneck how many GPUs a cloud provider can actually deploy, regardless of how many accelerator dies Nvidia ships.
Who’s Who: Nvidia, Intel, Micron and Monolithic Power
Each of the four companies plays a distinct role in the pipeline that ultimately determines how much it costs to run an AI coding agent in the cloud. NVIDIA Corporation supplies the GPUs (H100, B200 and successor architectures) that run both the large language models behind tools like Claude Code and Codex CLI, and the inference workloads that power code completion in editors like Cursor. Intel Corporation supplies CPUs for the host servers that orchestrate those GPU workloads and is attempting to rebuild its foundry business as a second source for advanced chips. Micron Technology, Inc. supplies the DRAM, HBM and NAND flash that every one of those systems depends on for memory bandwidth and storage. Monolithic Power Systems, Inc. supplies power-management chips that regulate voltage for high-density server racks, a less visible but increasingly critical link as GPU power draw keeps climbing generation over generation.
Micron’s role is arguably the most consequential for the near-term AI coding tool story. Tech Insider covered the company’s accelerating HBM4 sales growth ahead of its October 21 investor webinar, a separate but related data point showing just how tight next-generation memory supply has become heading into the holiday quarter.
Inside KeyBanc’s Call: John Vinh and the 14.9% Shipment Signal
The headline figure behind the KeyBanc note is the 14.9% above-trend semiconductor shipment reading for August, calculated on a three-month moving-average basis. In plain terms, chip shipments have been running meaningfully ahead of their historical trend line for months, a pattern that typically signals either unusually strong end demand, a restocking cycle among buyers worried about future scarcity, or both at once.
John Vinh’s framing, as reported, treats that shipment strength as a tailwind for the entire group rather than a red flag of overbuilding. That is consistent with how the AI infrastructure buildout has been described all year: hyperscalers and cloud GPU resellers are not slowing down capital spending, they are racing to secure capacity before memory and accelerator supply gets even tighter. The retrieved reporting does not include a verbatim quote from Vinh describing his reasoning in his own words, so any specific rationale beyond the shipment data and the four-stock pick itself should be treated as the analyst’s documented recommendation rather than a direct statement.
The Earnings Calendar Investors Are Watching
All four names report within roughly a month of each other, giving the market an unusually concentrated window to test whether the bullish thesis holds. Monolithic Power Systems and Intel are both scheduled to report on October 29, 2026, while Nvidia’s fiscal 2027 third-quarter results land nearly three weeks later, on November 17, 2026. Micron’s exact next report date was not specified in the retrieved reporting, though its HBM4 momentum is already drawing investor attention ahead of its own mid-cycle update.
| Company | Report Date | EPS Estimate | Revenue Estimate |
|---|---|---|---|
| Monolithic Power Systems | October 29, 2026 | $7.71 | $1.15 billion |
| Intel Corporation | October 29, 2026 | ~$0.40 | ~$16.48 billion |
| NVIDIA Corporation (FY27 Q3) | November 17, 2026 | ~$2.47 | ~$108.80 billion |
| Micron Technology | Not specified in reporting | Not specified | Not specified |
Wall Street’s expectations for Nvidia alone, at roughly $108.80 billion in quarterly revenue, illustrate the scale of the AI buildout that is driving the memory and accelerator shortage in the first place. That figure, if it holds, would represent one of the largest single-quarter revenue totals ever reported by a semiconductor company, and it is the backdrop against which every cloud GPU price and every AI coding tool subscription tier is currently being set.
Why This Story Matters Beyond Wall Street: The AI Coding Tool Connection
Here is the part of the story that rarely makes it into earnings coverage: the same chip shortage lifting these four stocks is the direct input cost for every AI-assisted coding product on the market. GitHub Copilot, Cursor, Claude Code and OpenAI’s Codex CLI all run on rented or owned GPU capacity. Autocomplete suggestions are relatively cheap to serve, but the autonomous “agent” features that have become the industry’s biggest selling point, repository-wide code search, multi-file refactors, automated test execution, and background pull-request generation, are dramatically more GPU-intensive than a simple inline suggestion.
That distinction is why the semiconductor earnings story is a developer-tools story in disguise. A coding assistant vendor offering “unlimited” agent usage on a flat monthly fee is making a bet on its own GPU cost structure. If that cost structure gets squeezed by memory and accelerator scarcity, something has to give: usage caps tighten, premium tiers get more expensive, or queue times during peak hours get longer. None of that requires a dramatic announcement. It shows up quietly, in rate limits and fine print.
Memory Is the Bottleneck: DRAM, HBM and NAND Tightness
Modern AI accelerators are only as fast as the memory feeding them. High-bandwidth memory stacks sit directly on the same package as the GPU die, and HBM capacity, not raw compute, is frequently the limiting factor in how many usable AI servers a cloud provider can actually assemble. That is why Micron’s HBM4 sales trajectory matters as much to developers renting GPU time as it does to Micron shareholders.
Conventional DRAM and NAND flash contract prices have also moved sharply this year, and reporting on memory markets has generally not expected the shortage to ease before 2027 at the earliest. For teams running their own inference infrastructure rather than renting from a hyperscaler, that pressure also hits closer to home: Tech Insider has reported separately on DDR5 RAM prices climbing toward $450 for 32GB kits as the broader AI-driven memory shortage works its way down to consumer and workstation components, including the machines developers use to run local models and build AI-assisted tooling.
The GPU Cloud Pricing Gap Developers Are Feeling
Cloud GPU pricing in October 2026 shows just how uneven the market has become. On-demand H100 instances range from roughly $2.89 per GPU-hour on budget providers like RunPod up to $12.29 per GPU-hour on Azure’s ND H100 v5 instances, a spread of more than four times for broadly comparable hardware. AWS sits closer to the middle at roughly $6.88 per GPU-hour on its eight-GPU p5.48xlarge instance, while Google Cloud’s a3-highgpu-8g configuration lists around $11.06 per GPU-hour.
Newer Blackwell-generation B200 capacity shows a similar pattern. AWS prices an eight-GPU B200 configuration at roughly $14.24 per GPU-hour, while Google Cloud’s flex-start B200 offering comes in lower, around $8.06 per GPU-hour, and Lambda has listed B200 capacity in the $6.69 to $6.99 per GPU-hour range depending on configuration size. Tech Insider has tracked this same three-way pricing gap in depth in its AWS vs. Azure vs. GCP GPU cloud pricing comparison, and the spread has direct consequences for any AI coding tool vendor deciding where to host its inference workloads.
| Provider | H100 (8-GPU, $/hr per GPU) | B200 (8-GPU, $/hr per GPU) |
|---|---|---|
| AWS | $6.88 | $14.24 |
| Microsoft Azure | $12.29 | Not widely listed |
| Google Cloud | $11.06 | $8.06 (flex-start) |
| Lambda | Not widely listed | $6.69–$6.99 |
| RunPod | $2.89 | Not widely listed |
For a coding-agent provider weighing where to run its background tasks, repository indexing and test-execution loops, that pricing spread is the difference between a sustainable margin and a money-losing “unlimited” plan. It also explains why so many vendors have started segmenting their products into tiers that charge more for heavier agent usage rather than offering one flat price for everything.
How Chip Costs Flow Into Your AI Coding Assistant Bill
Pricing across the major AI coding assistants already reflects a tiered structure built around how much GPU time a given workflow consumes. Cursor’s individual plans run from a free Hobby tier up to $20 per month for Pro, $60 for Pro Plus, and $200 for Ultra, with each step up unlocking more agent usage, cloud-agent capacity, and extended context. GitHub Copilot offers a free tier capped at 2,000 completions per month, a $10 Pro plan, a $39 Pro+ tier, and a $100 Max tier for teams that need larger premium-model credit pools. Tech Insider recently covered how GitHub Copilot retired six older models in favor of GPT-5.6, a move that also reshuffled which models are available at which tier.
Claude Code is bundled into Claude Pro at $20 per month billed monthly, with higher-volume Max tiers running roughly £90 per month for Max 5x and £180 per month for Max 20x in markets where that pricing has been reported. OpenAI’s Codex CLI uses a more usage-linked model, reported in the $8 to $20 per month range plus consumption-based credits, rather than a flat unlimited subscription. That credit-based structure is itself a signal: usage-based pricing is a hedge against exactly the kind of GPU cost volatility that a chip shortage produces.
| Tool | Free Tier | Mid Tier | Top Tier |
|---|---|---|---|
| GitHub Copilot | 2,000 completions/mo | $10/mo (Pro), $39/mo (Pro+) | $100/mo (Max) |
| Cursor | Hobby (limited) | $20/mo (Pro), $60/mo (Pro Plus) | $200/mo (Ultra) |
| Claude Code | Not offered standalone | $20/mo (bundled with Claude Pro) | ~£90–£180/mo (Max tiers) |
| OpenAI Codex CLI | Limited credits | $8–$20/mo + usage credits | Scales with credit consumption |
Engineering teams trying to forecast their own AI tooling spend increasingly treat these subscriptions the way they treat cloud compute bills, something to monitor and cap rather than a fixed line item. Tech Insider’s guide to building a Cursor Composer cost guardrail walks through one way teams are already doing that proactively, in anticipation of exactly the kind of price and usage-cap volatility this earnings cycle could accelerate.
Historical Context: From Chip Glut to Chip Squeeze
It is worth remembering how recently the semiconductor industry was dealing with the opposite problem. Memory makers including Micron spent much of 2022 and 2023 working through a severe DRAM and NAND glut, cutting production and writing down inventory as PC and smartphone demand slumped. The pivot to today’s tightness has been driven almost entirely by one force: the buildout of AI data centers, which consume HBM and DRAM at a rate that traditional computing demand never approached.
That pivot has happened faster than most memory executives projected even a year ago. Capacity that was being idled for lack of demand in 2023 is now the scarcest resource in the entire technology supply chain, and pricing power has shifted decisively back to memory makers for the first time in several product cycles. The AI coding tool market has effectively been built, grown, and priced entirely within this new scarcity regime, meaning most developers have never experienced what these products would cost in a looser memory market.
Market Impact: What a Beat or Miss Means for Developer Tools
The market impact of this earnings cycle extends well past share prices. If Monolithic Power Systems, Intel, Micron and Nvidia all post results consistent with the bullish KeyBanc thesis, it effectively confirms that AI infrastructure demand and memory scarcity are not easing in the near term, which gives coding-tool vendors less room to delay price adjustments or loosen usage caps. If one or more of the four misses expectations or signals softening demand, that would be the first real evidence this year that the AI infrastructure cycle is decelerating, which could ease some of the margin pressure on coding-agent providers.
Either outcome is being watched closely by developer-tool vendors themselves, several of which have reportedly deepened partnerships with specific cloud providers to lock in GPU capacity ahead of further price moves. Nvidia’s own next earnings report, expected November 17, 2026, will be the single most-watched data point in that chain, given how directly its guidance shapes expectations for GPU availability into 2027.
Competitive Comparison: Coding Assistants Under Cost Pressure
Not every AI coding assistant is equally exposed to GPU cost swings. Tools built around smaller, cheaper models for routine autocomplete, reserving larger frontier models only for complex agentic tasks, are better insulated than tools that route every request through the most expensive available model. GitHub Copilot’s free tier, for instance, uses lighter-weight models for its completion feature, reserving premium models for paid tiers. Cursor and Claude Code both lean on model routing to varying degrees, sending simple edits to smaller models while escalating complex, multi-file tasks to flagship models that consume far more GPU time per request.
This is also where competition between coding assistants and the underlying large language models themselves becomes relevant. Tech Insider’s comparison of Grok 4.7, Claude Sonnet 5.5 and DeepSeek V4.1 Flash found price gaps of more than 13x between models with broadly comparable coding capability, underscoring how much room coding-tool vendors have to manage costs simply by choosing which model answers a given request. As GPU capacity gets scarcer and more expensive across the board, that model-routing flexibility becomes a competitive advantage, not just a technical detail.
Developers evaluating which coding assistant to adopt can get a rough sense of a vendor’s exposure to this dynamic by checking which cloud provider and GPU generation they rely on. A simple way to track current on-demand GPU rates, which several coding-tool teams reportedly monitor internally when deciding where to route expensive agent workloads, looks like this:
# Example: checking current on-demand GPU instance pricing via AWS CLI
aws pricing get-products \
--service-code AmazonEC2 \
--filters "Type=TERM_MATCH,Field=instanceType,Value=p5.48xlarge" \
--region us-east-1That kind of pricing check, run against whichever region and instance type a vendor actually uses, is a practical way for engineering leads to estimate how exposed their AI coding tool stack is to the chip shortage playing out in this week’s earnings calendar.
Expert Voices on the Economics of AI Compute
Nvidia CEO Jensen Huang has repeatedly framed the current AI buildout, and the compute economics underneath it, in terms that echo exactly the dynamic now showing up in coding-tool pricing. Huang has said that “generative AI is the largest TAM expansion of software and hardware that we’ve seen in several decades,” a framing that explains why memory and accelerator makers are racing to add capacity even as shortages persist.
Huang has also pushed back on the idea that a GPU is a simple commodity part, noting that “people think that the GPU is a chip. But the HGX H100, the Hopper HGX has 35,000 parts, it weighs 70 pounds,” underscoring why power-delivery and memory components, not just the GPU die itself, end up gating how much accelerator capacity the industry can actually build.
On cost trends specifically, Huang has pointed to dramatic inference efficiency gains even amid rising hardware prices, saying that “our inference performance, another way of saying inference cost, just reduced by a factor of four within about a year’s time.” More recently, in materials tied to Nvidia’s fiscal 2027 second-quarter results, Huang distilled the company’s broader thesis into a single line: “compute is revenue.” He has also described the long-term goal for the industry as wanting “intelligence costs to decrease by a factor of two or 20 every single year because the consumption of it is going to more than make up for it,” a statement that captures the tension at the center of this week’s earnings story: hardware scarcity can raise near-term costs even while the long-run trend for AI compute, including the AI compute behind coding assistants, continues to point toward cheaper intelligence per dollar.
What Analysts and Reports Are Not Saying
It is worth being precise about what this week’s reporting does and does not confirm. The retrieved Barron’s and KeyBanc coverage names the four stock picks, the shipment statistic, and the earnings dates and estimates listed above. It does not include a confirmed price increase announcement from any AI coding tool vendor tied directly to this earnings cycle, and it does not include specific price-target figures or exact recommendation wording (strong buy versus buy, for example) for each of the four stocks beyond the “buy” framing in the headline itself.
Framed honestly, the connection between this earnings story and AI coding tool pricing is a structural one, not a confirmed cause-and-effect announcement. Memory and GPU scarcity create the cost pressure; how individual vendors respond, through tighter usage caps, new pricing tiers, or simply absorbing the cost, remains a company-by-company decision that has not yet played out in full.
Predictions: Where AI Coding Tool Pricing Heads Next
- Expect more aggressive usage caps on “unlimited” agent plans as vendors absorb higher GPU-hour costs rather than raising headline subscription prices outright.
- Expect wider price differentiation between simple autocomplete requests and autonomous, multi-step agent tasks, since the latter consume dramatically more compute per request.
- Expect coding-tool vendors to lean harder into model routing, sending routine edits to smaller, cheaper models while reserving frontier models for complex tasks, as a direct hedge against GPU scarcity.
- Expect Micron’s and Nvidia’s upcoming earnings calls to include direct commentary on HBM and GPU allocation for 2027, which will shape how much room cloud providers, and by extension coding-tool vendors, have to avoid further cost increases.
- Expect continued regional and provider-level pricing divergence, meaning two AI coding assistants built on similar underlying models could see meaningfully different cost structures depending purely on which cloud and GPU generation they standardized on.
None of these are guaranteed outcomes. They are directional expectations based on the documented shipment data, the confirmed earnings calendar, and the pricing structures already in place across the major coding assistants as of October 2026.
Frequently Asked Questions
Why did Barron’s name Nvidia, Intel, Micron and Monolithic Power as buys ahead of earnings?
The call, attributed to KeyBanc analyst John Vinh, cited semiconductor shipments running 14.9% above trend in August on a three-month moving-average basis, a signal of broad-based strength across the chip supply chain rather than one company’s product cycle.
When do these four companies report earnings?
Monolithic Power Systems and Intel are both scheduled to report on October 29, 2026. Nvidia’s fiscal 2027 third-quarter results are expected November 17, 2026. Micron’s next report date was not specified in the available reporting.
How does a chip shortage affect AI coding assistants like Copilot or Cursor?
AI coding assistants run on rented or owned GPU capacity. When DRAM, HBM or NAND supply tightens, cloud GPU capacity gets scarcer and more expensive, which can pressure vendors to tighten usage caps, raise prices on premium tiers, or slow response times during peak demand.
Has any AI coding tool vendor confirmed a price increase tied to this chip shortage?
No. The available reporting does not include a confirmed price increase from GitHub Copilot, Cursor, Claude Code or OpenAI’s Codex CLI tied directly to this earnings cycle. The connection described in this story is structural, based on shared reliance on the same constrained GPU and memory supply chain.
What is Monolithic Power Systems and why is it included with Nvidia and Intel?
Monolithic Power Systems makes power-management chips that regulate voltage for high-density server racks, including the ones running Nvidia GPUs. Its inclusion in the same buy call reflects how power delivery has become as critical a bottleneck as the GPUs or memory themselves.
Which cloud provider has the cheapest GPU pricing right now?
Pricing varies widely by provider and GPU generation. As of October 2026, on-demand H100 pricing ranged from roughly $2.89 per GPU-hour on RunPod to $12.29 per GPU-hour on Azure, while B200 pricing ranged from roughly $6.69 per GPU-hour on Lambda to $14.24 per GPU-hour on AWS.
Is the memory shortage expected to ease soon?
Reporting on memory markets has generally not expected DRAM and NAND shortages to ease before 2027 at the earliest, suggesting cost pressure on AI infrastructure, and the tools built on top of it, is likely to persist through at least the next several quarters.
Should developers expect their AI coding tool subscriptions to get more expensive?
There is no confirmed price increase yet, but the underlying cost pressure, tight GPU and memory supply amid surging AI infrastructure demand, is real and documented. Usage caps and tier restructuring are more likely near-term responses than outright price hikes.
